Side Finds

Western Digital repurchased $1.92 billion of its own stock in nine months — yet the share count fell only from 347 to 345 million. The reason sits in the debt footnotes: the convertible note ($37.72 conversion price, $50.41 cap) and the forced conversion of the preferred stock keep feeding new shares into the system — the buybacks are fighting the company's own capital structure.

Western Digital still holds a block of shares in its former subsidiary Sandisk — and uses it as currency: per a current report (8-K) dated June 11, 2026, the company swapped roughly one million Sandisk shares directly for its own shares. A buyback that costs no cash, but cashes in the spin-off's inheritance instead.

In the patent case brought by MR Technologies, Western Digital booked a $384 million charge in fiscal year 2024. In April 2025 came the global settlement — for $130 million. Releasing the roughly $200 million of excess reserve flattered the quarterly result in which it was recorded.

In October 2024 a jury awarded the firm SPEX Technologies $316 million in damages against Western Digital — with interest, the claim grew to roughly $553 million. On June 16, 2025 the court threw out the verdict for lack of a sound damages theory and set nominal damages of one dollar instead. Both sides are appealing; Western Digital has not booked a reserve for the case.

From the exchange-hype era of 2019 through 2021, Virgin Galactic drags along a whole bundle of shareholder lawsuits — the class action and several derivative suits carry case names like "… v. Branson et al.". In July 2025 the securities class action was settled for $8.5 million, of which the company expects $6.25 million from its insurers; net, $2.25 million remained as an expense in 2025. The court preliminarily approved the settlement on March 11, 2026, and the payments are to flow by the second quarter of 2026. In April 2026 a settlement agreement followed for two derivative suits as well. Old promises can have expensive afterlives — here, at least, mostly at the insurers' expense.

Almost in passing, the quarterly report (10-Q) mentions a second business model: Virgin Galactic is evaluating using a derivative of its twin-fuselage carrier aircraft as a HALE aircraft ("High-Altitude, Long-Endurance") — an aircraft that flies very high and very long and, per the report, could be suitable "for several types of government and research applications". After the design work on the new spaceships was completed, engineers were already reassigned to the next carrier-aircraft generation. For now this is pure future music without revenue — but it is the only idea documented in the filings by which the expensive aircraft technology could earn money beyond space tourism.

The trademark license agreement (Amended TMLA) runs until October 2044 — but it contains a remarkable exit clause: Virgin Enterprises may terminate if the commercial launch does not happen by a set date, or if the company afterwards cannot conduct commercial flights with paying passengers for a defined period of time (pauses due to significant safety issues excepted). After a termination, 90 days would remain to destroy all materials carrying the Virgin logo and to change the company name. For a company that has not flown since June 2024, this is more than a footnote: the global brand, too, hangs on the restart date.

Virgin Galactic does not own the name "Virgin": the brand is licensed from Virgin Enterprises Limited, a company from the business empire of Sir Richard Branson. License fees flow in return — a low single-digit percentage of revenue, but at minimum a fixed base amount. In 2025, according to the annual report (10-K), the brand fees added up to $2.5 million — more than the complete annual revenue of $1.5 million. A company that pays more for its name than it takes in with its business: that should be rare even on the stock market.

In the middle of the cash squeeze, TuHURA still bought more: through the TuHURA-Kineta merger (agreement of May 5, 2025, cash-and-stock), the company acquired the private Kineta, Inc. and with it a second drug candidate — TBS-2025, a bifunctional, bispecific antibody-drug-conjugate ("ADC") approach meant to shut down myeloid-derived suppressor cells (MDSCs) in the tumor environment and prevent resistance to checkpoint inhibitors.

Two readings are honestly to be set side by side. The opportunity: a second leg to stand on, should IFx-2.0 stumble. The price: "acquisition-related costs" of $3.7 million in 2025 alone and further dilution — the number of shares outstanding rose within a year from 12.2 million (end of 2024) to 59.3 million (end of 2025) and 63.6 million (March 31, 2026). Growth paid for with fresh shares and bolted-on companies is rarely free.

In the risk section of the annual report sits an episode that shows how close it was at times: TuHURA's stock closed 30 consecutive trading days below one U.S. dollar, breaching the Nasdaq minimum-bid-price rule (Listing Rule 5550(a)(2)). The company received the usual 180-day grace period — until July 28, 2026 — to get back above the mark.

It succeeded, if narrowly: on February 26, 2026, Nasdaq notified TuHURA that it had regained compliance with the minimum-bid requirement. The report adds soberly, however, that there is no guarantee it stays that way. For a stock whose price hangs heavily on individual trial and regulatory news, the one-dollar threshold is therefore less a hurdle cleared than a recurring risk.

TuHURA Biosciences is not yet two years old — at least under this name. The listed vehicle behind it is the former Kintara Therapeutics, Inc., which traded under the ticker "KTRA" on Nasdaq and whose own cancer programs did not make it. On October 18, 2024, Kintara first carried out a 1-for-35 reverse stock split (35 old shares became one new one), then merged with the private legacy TuHURA and immediately renamed itself "TuHURA Biosciences, Inc."

For investors this is more than a footnote: a 1-for-35 reverse split is the classic maneuver to lift a crashed price optically out of penny-stock territory. The share history before October 2024 belongs to a different company with a different story — and the 2013 IPO year that appears in many databases is in truth the stock-market debut of the predecessor shell, not of TuHURA's business today.

When TuHURA ran out of money in April 2026, no outside investor stepped in — an insider did: the $50 million credit facility comes from Parkview Holdings One LLC, an affiliate of K&V Investment LLC — per the quarterly report (10-Q) "a holder of more than 5% of the Company's fully diluted capital stock and an entity owned by Vijay Patel." The loan is expensive and deeply anchored: 12 percent interest (plus 6 percent on default), secured by "substantially all assets" of the group, plus an annual commitment fee of 1.5 percent and the obligation to make repayments equal to 75 percent of net profits from drug sales.

The most remarkable part is in the fine print: Parkview additionally receives a royalty agreement — a low-to-mid single-digit annual license fee on the net revenue of future IFx-2.0 products, up to $450 million in revenue per year, continuing until the expiry of the last IFx-2.0 patent. Whoever rescues TuHURA thus secures not only a double-digit interest rate but a permanent stake in the greatest hope for the future — should the bet pay off.

Alongside BRIUMVI, TG Therapeutics is betting on a second hope: azer-cel, a cell therapy (CAR-T) the company licensed in from Precision BioSciences. What is remarkable is how closely the two firms are financially intertwined: TG Therapeutics holds not only the license but also shares of Precision BioSciences (about $1.3 million at fair value as of the balance-sheet date) — plus forward contracts to buy further Precision shares.

That is unusual: a company that licenses in technology from a partner is at the same time its shareholder and has contractually committed to buying more shares. Such cross-entanglements are not rare in the biotech world, but they bundle risks: if azer-cel development goes badly, it would hit TG Therapeutics twice — as licensee and as shareholder. A find that shows there are more connections behind the one-product image than a fleeting glance would suggest.

Between approval data and balance-sheet metrics, the annual report's "Business Highlights" chapter hides a remarkable marketing statement: TG Therapeutics launched the awareness platform "Next In MS" together with the actress Christina Applegate — accompanied by a commercial in Super Bowl LX, the most expensive advertising environment in the United States.

For a company that hangs on a single product and still carries roughly half a billion dollars in operating costs per year, a Super Bowl appearance is a confident — and expensive — bet on brand awareness. Applegate, who publicly lives with multiple sclerosis herself, lends the campaign credibility; at the same time the move shows how aggressively TG Therapeutics has to position BRIUMVI in the tight MS market against the brands of Roche and Novartis. Whoever sees the rising marketing costs in the income statement now knows where part of them flows.

A company that has just turned profitable and reports full coffers surely shouldn't need fresh debt? At TG Therapeutics it went differently. On March 18, 2026 the firm paid off its existing loan and closed a new $750 million loan with the financial investor Blue Owl Capital — three times as much as the previous $250 million, secured by "substantially all of the assets" of the company, bearing interest at a spread from 4.75 percentage points above the reference rate.

At the same time TG Therapeutics bought back its own shares: in March 2026 the board raised the running buyback program from $100 million to $300 million; $100 million had already been spent by quarter-end (average price $30.44), the treasury-stock balance stood at $200.2 million. Taking on debt and putting part of it into share buybacks while the valuation sits near a multi-year high — that is financial engineering that makes earnings per share look prettier but loads the balance sheet with interest cost and security interests. A detail that easily gets lost in the profitability euphoria.

The headline for fiscal year 2025 sounded like a dream: $447.2 million in net income — almost twenty times the prior year. But read one line higher in the income statement and you find the sober figure: pretax income was only $107.4 million. The difference of $339.8 million is not a sold drug but a tax bonus — and a non-cash one at that.

It arose because, after years of losses, TG Therapeutics released the so-called valuation allowance on its deferred tax assets: a company that writes red numbers for years may only recognize the resulting future tax benefits on the balance sheet once profits become probable. That is exactly what happened in 2025 — and the catch-up effect landed all at once as income in the profit. For investors it is a lesson in earnings quality: a value filter that bluntly looks at the price-to-earnings ratio therefore treats the stock as much "cheaper" than the operating business justifies. The bonus flows exactly once — next year the company pays normal taxes again.

Deep in the trial chapter stands the threshold that decides Taysha's future — and it is astonishingly concrete. The pivotal REVEAL trial enrolls 15 girls and young women aged 6 to under 22. Each patient is her own control; what is measured is how many regain at least one of 28 defined developmental milestones after treatment. The success threshold: a response rate of 33 percent — that is, 5 of 15 patients — suffices to statistically reject the null hypothesis.

The null hypothesis in turn holds that without treatment only about 1 of 15 patients (6.7 percent) would spontaneously reach such a milestone. In the early phase the response rate was 83 percent (5 of 6 patients on high dose). For investors that is the bet in its purest form: between jubilation and disappointment there may in the end lie two or three individual children whose progress blinded assessors rate on video.

It is the most instructive contradiction of this research: Taysha sits in our warning scanner "Going Concern (Distress-Proxy)", which hunts for the classic signs of a shaky balance sheet — an Altman Z-Score in the danger zone, no revenue, negative operating cash flow. All three apply. And yet as of March 31, 2026 the company sits on $276.6 million in cash, which per its own report lasts into 2028, and the auditor wrote no going-concern qualification into the opinion.

The reason lies in the mechanics of the metric: the Altman Z-Score punishes almost automatically firms without revenue and with accumulated losses — it was designed for classic industrial companies, not for pre-funded biotech labs that by definition burn money before they earn any. For investors that is the real lesson: a red warning signal is a reason to read on, not a verdict. At Taysha the original opinion contradicts the quantitative smoke detector — unlike, for instance, the otherwise similar Replimune, where auditor and scanner agreed.

In August 2025 Taysha drew $50 million from a new loan agreement with the specialty financier Trinity Capital (Tranche A of a framework of up to $100 million). For a company without revenue, debt is a double-edged sword: it does not dilute shareholders, but it hangs interest and repayment obligations on the company — and loan covenants that tighten its room to maneuver. One detail hardly any investor has on the radar: the loan is carried on the balance sheet at fair value ("fair value option", ASC 825) and was important enough to the auditor to be flagged as a "Critical Audit Matter".

What is interesting is the interplay with the timeline: the second loan tranche (a further $25 million) is available to the company only until March 31, 2028 — that is, exactly to the year in which, per the annual report, the cash also runs low. Anyone looking closely sees in the loan terms less a rescue anchor than a bridge that reaches precisely to the next capital need.

In the middle of the chapter on the warranty drama around shrinking cable insulation stands a half-sentence you have to read twice: "The Company does not maintain insurance for product warranty". The entire $73 million in remediation costs for the defective harnesses therefore ran, unchecked, through the company's own cash flow statement.

The hope of reimbursement rests solely on the lawsuit against the cable supplier Prysmian (filed in October 2023 in Nashville) — and under U.S. accounting rules (ASC 450) that may only appear in the books once success is all but certain. For investors this means: the money is demonstrably gone, the possible recovery remains a footnote until further notice.

While the whole world looks at cash and warranty costs, Shoals has been fighting a patent war since 2023 over its core invention, the "Big Lead Assembly" wiring system. In January 2025 the group filed new complaints against its competitor Voltage LLC with the U.S. trade authority ITC and before a federal court in North Carolina. On February 6, 2026 an ITC administrative judge ruled preliminarily that Voltage products infringe two Shoals patents; the final ITC decision is expected by June 2026, and before the district court Shoals additionally seeks damages.

The remarkable part sits in the accounting logic: Shoals treats the matter as a "gain contingency" — a possible gain that may only be booked once it is certain. In the numbers the Insolvency Radar sees, none of this trump card exists. The older ITC case from 2023, however, was initially lost and sits with the appeals court; the proceedings against co-defendant Hikam were ended by mutual agreement in February 2026.

Who leads the consolidated U.S. shareholder class action against the solar supplier from Tennessee? Not a Californian pension giant, but Erste Asset Management GmbH — the fund subsidiary of Austria's Erste Group. It was appointed lead plaintiff in "In re Shoals Technologies Group, Inc. Securities Litigation" by the U.S. federal court in Nashville; alongside it, plaintiffs include the pension plan of the municipal utility of Kissimmee, Florida.

The case shows, in passing, how international the plaintiffs' bench in U.S. securities litigation has become: a Viennese asset manager negotiated a $70 million settlement on behalf of all injured shareholders, which the court preliminarily approved on May 4, 2026 — $64.8 million of it is paid by Shoals' insurance.

In June 2024 Shoals felt strong: the board approved a share buyback program of up to $150 million (running through the end of 2025) and immediately executed an accelerated repurchase of $25 million — 3,908,387 shares at $6.40 each, handled through the investment bank Jefferies. For context: a few months earlier the group had fully repaid its term loan and upsized the credit line to $200 million.

The punchline was written by the balance sheet: the program never got beyond the $25 million opener, and as of March 31, 2026 there was $1.9 million of cash left in the books — against $181.8 million drawn on the credit line. The repurchased shares sit on the balance sheet today as treasury stock at $25.3 million. A lesson in how quickly "excess capital" turns into a liquidity buffer you would love to have back.

In the risk chapter of the annual report stands a paragraph that takeover speculators should know: the agreements with Kioxia on the joint Flash Ventures fabs contain clauses that, per Sandisk, could significantly impede shareholders' ability to benefit from future strategic transactions — including a takeover of Sandisk. A change of control can trigger rights of the Japanese partner; the report explicitly warns this could depress the share price and a possible takeover premium.

Translated: the joint venture that supplies Sandisk with practically its entire flash supply acts at the same time like a built-in poison pill — except it was not the board that adopted it but the supply contract. For the price fantasy "someday a big one buys them", that is a structural hurdle documented in the filing.

Shortly before the spin-off Sandisk parted with its assembly and test plant in Shanghai: 80 percent of SanDisk Semiconductor (Shanghai) Co. went to JCET, China's largest semiconductor packager; Sandisk kept 20 percent and has sourced the services as contract manufacturing since. The pre-tax gain on the sale: a modest $34 million — even though $382 million of goodwill allocated to the plant was handed over along with the net assets.

Strategically the deal is a quiet piece of de-risking: the company's own manufacturing presence in China shrinks to a minority stake while capacity remains secured through contracts. The 20 percent remainder stood on the books at $166 million as of April 3, 2026.

On March 25, 2026 Sandisk signed an agreement that does not fit the picture at first glance: the NAND specialist is buying about 139 million shares of the Taiwanese DRAM maker Nanya Technology for $972 million — about 3.9 percent of the company, via private placement. Per the quarterly report the purchase price sat 15 percent below the 30-day average price, in line with Taiwanese securities law.

What is remarkable is the direction: a flash maker that itself lives off the memory boom is putting almost a billion into the neighboring DRAM market — of all times in the most expensive phase of the industry's history, albeit at a discount. The supply chain of the AI boom thus keeps intertwining through cross-holdings; what Sandisk strategically intends with the stake, the report does not say.

Western Digital initially kept 19.9 percent of Sandisk (28,827,787 shares) at the February 2025 spin-off. What happened next is recorded soberly in the quarterly report: as early as June 9, 2025 — months before the price boom — WDC gave up 21,314,768 shares (14.6 percent), in exchange for its own debt held by WDC creditors; their banks immediately sold the shares on. In February 2026 another 5.8 million shares followed the same route. What remained were 1,691,884 shares, freely sellable since March 19, 2026.

Measured at the summer 2026 price level (about $2,335 per share, data as of July 8, 2026), the block given up in June 2025 alone would have been worth a double-digit billion amount years later. And one detail completes the footnote: "All expenses for these offerings were paid for by us" — per the report, all costs of these placements were borne by Sandisk itself, not by the selling former parent.

In the risk chapter of the annual report sits an incident hardly any investor has on the radar: in the third quarter of 2023, Sabre determined that an unauthorized actor had illegally extracted company data and published it on the dark web. The group called in forensic experts and U.S. federal law enforcement.

So far, per the report, the incident has had no material impact on the financial condition or results of operations — but there is no assurance that significant costs, lawsuits or regulatory proceedings will not yet follow. For a group that processes vast amounts of personal travel data every day, that remains an open item.

Since the first quarter of 2026, Sabre's revenue categories are no longer soberly called "Distribution" and "IT Solutions" but "Marketplace" and "Airline Technology" — per the quarterly report (10-Q) expressly to better reflect the company's evolving brand identity and market positioning. In the same breath, the report opens with the sentence that Sabre is an "AI-native technology leader".

The figures under the new labels are exactly the same as under the old ones: transaction fees per booking and SaaS fees per passenger boarded. Anyone comparing reports across several years should know about the renaming — and should not let fresh vocabulary promise a fresh balance sheet.

Deep in the legal-proceedings section of the annual report runs a case that has now reached its third decade: the Indian tax authority DIT asserts that Sabre's Singapore subsidiary SAPPL has a "permanent establishment" in India within the meaning of the Singapore–India double-taxation treaty — and issued tax assessments for the assessment years beginning March 2000. The Indian tax tribunal ITAT sided with Sabre (no taxable income), the authority moved on to the Bombay High Court.

Instead of ending, the dispute keeps growing: by now the assessment years through March 2016 as well as March 2018 through March 2021 carry similar assessments, and the appeals are pending. A proceeding older than the iPhone — one in which every new year adds a new assessment.

On Sunday, March 1, 2026, Sabre's board of directors adopted a "poison pill" (rights agreement) in a fast-track procedure: as soon as an investor crosses 15 percent of the shares, all other shareholders may buy in at a steep discount — a classic defensive weapon against unwanted takeovers. Only four days later, on March 5, came the peace accord: a "Strategic Governance Agreement" with Constellation Software, the Canadian serial acquirer of software companies, which had previously submitted a director nomination of its own.

The result: Damian McKay joins the board of directors as a new member (with a seat on the technology committee), Constellation commits to standing still (at most a 15 percent stake including economic exposure, voting in line with the board) — and the poison pill was buried again as of March 6. That the arguably most successful software acquirer in the world knocks on the door of a highly indebted travel-tech group is one of the most remarkable footnotes of this reporting year.

A small but treacherous peculiarity stands right at the start of the annual report: Replimune's fiscal year ends on March 31. "Fiscal year 2026" thus covers the period April 2025 to March 2026 — not the calendar year. In the same breath, however, the company clarifies that it reports its program and trial updates on a calendar-year basis.

Whoever reconciles figures and milestones from press releases with the reporting periods can easily be off by up to three quarters. For a company whose valuation hangs on individual dates, this double timekeeping is more than a footnote — it is an invitation to miscalculate.

Regulatory procedures are considered plannable — Replimune's approval saga is the opposite. After the first rejection notice (Complete Response Letter) in July 2025, the FDA had signaled at a meeting in September 2025 that a particular comparator arm (nivolumab plus relatlimab, trade name Opdualag) could be acceptable for the randomized confirmatory trial IGNYTE-3.

In the second rejection notice of April 10, 2026, however, the company writes itself, the FDA had backed away from this position again — and had moreover repeated points that, through the interim acceptance of the resubmission, actually counted as settled. For investors, that is the real lesson of this case: with a binary approval bet, not only the outcome is uncertain, but the rules of the game as well.

In Replimune's capital structure sits an instrument many retail investors overlook: "pre-funded warrants". Investors have already paid almost the full purchase price of the share; for the actual conversion into shares only a symbolic exercise price remains. As of the balance-sheet date, 14,058,153 shares were thus outstanding that can be converted into real shares at a tenth of a cent practically at any time.

The construct helps investors circumvent statutory ownership thresholds — but for the free float it means additional latent dilution. And because the annual share pool of the employee program automatically increases by 4 percent of the outstanding shares, these warrant shares even count toward it. The cake can therefore grow without fresh money coming in — at the expense of the slices already on the table.

Replimune's most important trial, the IGNYTE trial, tests its own drug RP1 in combination with nivolumab — an established immune checkpoint inhibitor of the pharma giant Bristol Myers Squibb (BMS). Whoever reads the fine print comes upon a remarkable arrangement: BMS has granted Replimune a non-exclusive, royalty-free license and supplies nivolumab at no cost for the trial.

This is opportunity and hidden dependency at once. Opportunity, because a tiny biotech could hardly shoulder such a combination trial without this free supply. Dependency, because the marketing application rests on exactly this combination — and in the second rejection notice the FDA rattled, of all things, at the question of which comparator arm is acceptable for the confirmatory trial. Getting a drug as a gift is rarely entirely free.

In the export-control chapter of the 2026 annual report there is a sentence that makes constitutional lawyers sit up: U.S. government officials expressed the expectation that the government would receive "15% or more of the revenue generated from licensed sales" of Nvidia's products — "but the USG did not publish a regulation codifying such requirement." A government revenue share as an informal expectation, nowhere bindingly regulated: that has rarely been seen in U.S. mandatory filings.

The rest of the chapter shows how small the business in question has become: under the H20 licenses granted in August 2025, Nvidia generated only about $60 million in revenue — after a $4.5 billion write-down on those very chips. And the H200 shipments allowed in February 2026 must be physically inspected in the United States before export, which triggers a 25 percent tariff on each chip at import; revenue under this through the filing of the quarterly report: zero. Anyone betting on a comeback of the China business is thus wagering not only against Beijing's counter-boycott — but also on the terms of a silent stakeholder in Washington.

In the derivatives note of the 2026 annual report a transaction appears that you would sooner expect at a bank: Nvidia guarantees the lease obligations of partners should they default — maximum gross exposure: $3.5 billion, terms of 5 to 7 years. The partners have posted $712 million as collateral. The consideration is the real twist: for the guarantees Nvidia receives warrants — that is, rights to acquire shares in the partners.

Economically this means: the chip supplier insures the leases of the very infrastructure partners who build data centers for its chips — and gets an equity interest in those partners in return. If the bet pays off, Nvidia earns twice; if a partner topples, Nvidia is on the hook precisely when its own business is weakening too. It is the same mechanism other AI companies use — Alphabet, according to its own quarterly report, guarantees a multiple of this for third-party data centers. The AI build-out is increasingly cross-insured, and the balance sheet traces of this entanglement grow faster than most investors can watch.

The chip war throws off strange blossoms: according to its annual report, Nvidia holds a — previously announced — stake in its former arch-rival Intel, and its price gains drove other income in fiscal year 2026: $11.1 billion, of which $8.9 billion were price gains on investments. In the first quarter of fiscal year 2027 this became a genuine profit driver: of $58.3 billion in net income, $15.9 billion came from valuation gains on securities — more than a quarter. The holdings of publicly traded stakes jumped from $12.9 to $30.2 billion within three months.

This repeats, at a chip giant, a pattern investors otherwise know from holding companies: a growing part of reported profit arises not from products sold but from the market valuation of stakes — and that swings both ways. The filing does the math itself: a hypothetical 10 percent decline in the publicly traded stakes would cost $3.9 billion in book value (as of April 26, 2026). Anyone comparing Nvidia's quarterly profits to the prior year should strip out this paper share.

The cash flow statement of the latest annual report contains a line Nvidia has never shown before: "Groq, Inc. — 13,000" — a $13 billion outflow in a single item. Behind it is not an acquisition but a non-exclusive license agreement signed in December 2025 for intellectual property of the chip startup Groq, which with its specialized inference processors was considered one of the most serious architectural challengers to Nvidia's GPUs. A further roughly $4 billion still sat on the balance sheet as an "accrued purchase obligation" as of April 26, 2026.

Notable is the candor of Nvidia's own risk chapter: the payments are described as "significant, nonrefundable," integrating the licensed technology into its own architectures requires "substantial engineering effort," may be delayed or never happen at all, and Nvidia may be "unable to recoup the associated costs or realize an adequate return." Translated: the market leader pays a double-digit billion sum with no right of return to bring a challenger's ideas in-house — and writes, itself, that success is open. For the question of how seriously Nvidia takes the competition from specialized inference chips, there is hardly a more expensive piece of evidence.

In the summer of 2026, "artificial intelligence" is the most-invoked phrase of the U.S. reporting season — hardly a company fails to hoist at least one AI platitude into its risk factors. We ran a full-text search across the six most recent SEC filings of Madison Square Garden Entertainment (two annual reports 10-K, four quarterly reports 10-Q): not a single hit. Neither "artificial intelligence" nor "machine learning", neither "generative" nor "AI" — nothing.

That is not carelessness but honest self-description: MSGE earns its money by people physically walking into a hall and experiencing live what cannot be streamed — Rockettes, Knicks, sold-out concerts. You can even read a quiet anti-AI thesis into it: the more interchangeable digital content becomes, the more valuable the uncopyable gets. On the stock market of 2026, where AI mentions are thrown around like confetti, a multi-billion-dollar company without a single AI word in its mandatory filings is a genuine rarity.

In the related-party section of the annual report sits a chapter you would not expect from an arena operator: aircraft arrangements with the owning family. MSG Entertainment provides "aircraft support services" for members of the Dolan family — among them executive chairman and CEO James L. Dolan himself — and in return, when needed, leases a Bombardier Challenger 350 from Brighid Air, a company of Patrick F. Dolan, the CEO's brother. The pilots come from the Dolan Family Office LLC, which is controlled by the estate of family patriarch Charles F. Dolan.

On top of that come time-sharing arrangements over aircraft criss-crossing with Sphere Entertainment, MSG Sports and AMC Networks — all companies under the same family's control. Every single arrangement is disclosed and documented in the customary way; taken together, however, they show how tightly the corporate till and the family ecosystem are interwoven. Whoever buys the stock should know: at MSGE, "corporate" and "family" is not a sharp line but a web.

Deep in the risk chapter of the annual report stands a number you have to read twice: the Madison Square Garden complex benefits from a property-tax exemption under a New York State law of 1982 — and in fiscal year 2025 that exemption was worth $43.0 million. For comparison: the group's net income in the same fiscal year was $37.4 million. The tax privilege is thus worth more than the entire annual profit.

And it wobbles: in January 2023, elected New York representatives demanded in an open letter that the exemption be reviewed; in July 2023 the city's Independent Budget Office followed up with a report pointing the same way. The punchline sits in the arena license agreements: the Knicks and Rangers teams would formally have to bear 100 percent of any property tax — but if the exemption falls, the annual license fee MSG Entertainment receives from the teams drops in return. One stroke of the pen by state lawmakers in Albany would therefore hit the landlord's revenues directly.

Micron's U.S. fab offensive is half a government project: up to $6.4 billion in direct grants from the CHIPS Act for new plants in Idaho, New York and Virginia, plus a 35 percent investment tax credit on qualified U.S. semiconductor investments. In total, $7.9 billion in committed government incentives from various governments (United States, India, Japan, Singapore) were still outstanding as of August 28, 2025; incentives already received have reduced the carrying value of property, plant and equipment by $5.04 billion.

The fine print has teeth: the grants are tied to milestones in construction, tool installation and wafer output — and on a miss they are "subject to reduction, termination, or clawback", in part including interest. The annual report explicitly names a "cyclical downturn" of the company's own business as a possible reason for missing them. Translated: in precisely the scenario in which Micron would need the money most, part of it could be demanded back.

Since May 2023, operators of "critical information infrastructure" in China have been barred from buying Micron products — so ruled China's cyberspace regulator, the CAC. Less well known is the second front: Yangtze Memory Technologies (YMTC), China's state-backed NAND maker, has been showering Micron with patent suits since November 2023 — before the federal district court in Northern California (eight U.S. patents, 3D NAND) and, since early 2024, also before the Beijing Intellectual Property Court (three Chinese patents, including a demand for sales bans in China).

The roles are remarkably distributed: first the Chinese regulator bans the U.S. maker from sensitive markets, then the Chinese state competitor sues it for injunctions — in the same country. Patent law as the continuation of the trade war by other means: for Micron, China is thus simultaneously sales market, production site, lost customer and courtroom opponent.

Hidden in the legal-proceedings section of the quarterly report is a juridical paradox: in May 2024 a Texas jury ordered Micron to pay $425 million for infringing a memory-module patent of the firm Netlist — plus $20 million for a second patent. The curious part: the U.S. Patent and Trademark Office (USPTO) had declared the sole asserted claim of the first patent unpatentable a month before the verdict; in July 2024 the same classification followed for all asserted claims of the second patent.

Now the appeals run crosswise: Micron is challenging the $445 million judgment, Netlist the unpatentability decisions. If the federal appeals court upholds the patent office's decisions, the judgment is, per the filing, unenforceable — the jury award would be waste paper. On the side, Netlist keeps suing: since May 2025 also against Micron's HBM products, the heart of the AI business. For investors, a lesson in how long and how contradictorily patent risks can live on in the footnotes.

In the investments note of the latest quarterly report stands a single, offhand sentence with astonishing content: after the May 28, 2026 balance-sheet date, Micron bought a stake in a "leading AI company" — non-marketable equity securities, meaning shares not listed on any exchange. Neither name nor purchase price nor purpose is disclosed.

What is remarkable is the direction of the money flow: the memory maker that lives off the AI boom is in turn buying into an AI company — while at the same time $22 billion in customer funds from take-or-pay contracts flow toward it. The AI boom's supply chain is thus increasingly intertwining through cross-holdings and prepayments — a pattern investors know from earlier investment cycles, and one that raises the question of how independent supply and demand in this boom still are of each other. What exactly Micron bought will probably only emerge in the next report.

KLA earns a third of its revenue in China — and at the same time feels the reverse thrust of the trade war: the annual report (10-K) names China's export controls on rare earths and related materials as a risk to its own supply chain. The equipment maker whose U.S. export licenses cap its China revenue depends, in the other direction, on Chinese raw-material approvals.

KLA manufactures significant parts of its systems in Migdal HaEmek and Yavne (Israel). The annual report (10-K) warns of rocket attacks and disrupted Red Sea shipping routes — and of employees being called up for reserve duty in the Israeli army: the consequences, it says, could be material. On top of that, an open assessment from the Israeli tax authority is pending (audit of 2019 through fiscal year 2022, objection filed).

While the core business celebrates records, KLA wrote off $230.4 million of goodwill on its former printed-circuit-board unit in fiscal year 2025 — after $263.1 million already in fiscal year 2024 (of which $192.6 million on PCB/display). The display business from the 2019 Orbotech acquisition was shut down entirely; the quarterly report (10-Q) still carries restructuring costs "as a result of exiting the Display business".

A curiosity on the margins of the reading: in Kirby's annual report for 2025, the Gulf of Mexico no longer appears under its traditional name. In its place stands, six times, "Gulf of America" — the renaming decreed by the U.S. government by executive order in early 2025 — for instance where offshore supply vessels and drilling rigs serviced by Kirby's service segment are concerned. The prior-year report, filed as early as February 2025, had already been converted completely.

For investors this is substantively irrelevant — and revealing precisely because of that: SEC filings are legally vetted documents in which every formulation is a considered choice. How quickly and consistently a Texan heritage company headquartered in Houston adopts the new official language rule, while international maps and news agencies keep writing "Gulf of Mexico", is a small time capsule of the 2025/2026 reporting season.

In the fourth quarter of 2024, Kirby had to write down $56.3 million pre-tax on inventories — above all on equipment for conventional diesel fracking. The reason, per the annual report: an industry-wide shift to electric fracking systems that made demand for the old diesel technology collapse. Parts of the inventory simply had "limited commercial opportunity" — hardly any market prospects left.

The punch line: Kirby stood on both sides of this upheaval. The company builds electric fracking systems itself and delivers them — the new orders thus devalue its own legacy business along the way. In 2025 the oil and gas share of the Distribution & Services segment shrank 32 percent to just 11 percent of segment revenue (2023: about a quarter). A textbook lesson in how technology shifts first become visible in the balance sheet exactly where nobody looks: in inventories.

An inland barge operator whose vessels may, by law, ply only American waters is the most down-to-earth business model imaginable. All the more surprising the footnote in the distribution chapter: Kirby's Distribution & Services segment serves its customers through "62 branch locations across 16 states and Colombia, South America". Add to that distribution rights for EMD engines in Mexico, Central America, northern South America and the Caribbean, acquired in 2025.

The annual report describes the consequence right along with it: because of the foreign business, Kirby is subject to the Foreign Corrupt Practices Act (FCPA), the U.S. anti-corruption law for business abroad — and trains its entire workforce in corruption prevention. A tank barge operator from the Mississippi that drills its whole staff against foreign bribery: that is the kind of detail only the mandatory filing brings to light.

You would expect it otherwise from a 105-year-old infrastructure company with stable cash flows — but there it stands, black on white, in the annual report: "Since 1989, the Company has not paid any dividends on its common stock." 37 years of nothing — and a fixed dividend policy expressly does not exist.

The money flows into buybacks of its own shares instead: in 2025 alone, Kirby repurchased 3.7 million shares for $354.2 million (average price $96.27) — nearly the entire annual profit of $354.6 million. The share count has fallen by a good tenth since the end of 2022, which additionally lifts earnings per share. For income investors the stock is structurally uninteresting; for everyone else it is one of the most consistent buyback-instead-of-dividend policies in the U.S. industrial sector.

On June 2, 2025, a shareholder filed a class action in the U.S. District Court for the Southern District of New York — Lucas Deslande v. Fortrea Holdings Inc. et al. — against the company and current and former officers. The allegation: omissions and misrepresentations toward investors in violation of U.S. securities law. On September 3, 2025 the court appointed two pension funds as lead plaintiffs: the Construction Industry Laborers Pension Fund and the City of Pontiac Reestablished General Employees Retirement System.

The amended complaint followed on November 10, 2025, Fortrea's motion to dismiss on January 28, 2026 — the proceedings are thus at the very beginning, and by its own account the company cannot yet estimate a possible loss. For investors the timeline is the most revealing part: the lawsuit fell into the same spring as the share-price slide below $5, the goodwill impairments of $797.9 million — and nine days later the board's poison pill.

Fortrea has sold $300 million of customer receivables and derecognized them from its balance sheet — through a securitization program that has been running since May 2024 and spares the cash balance accordingly. On February 24, 2026, two days before the annual report was published, the program was extended through February 2029. The same amendment contains a detail that is easy to read past: it grants the administrative agent special rights as soon as one of two named rating agencies downgrades Fortrea's creditworthiness.

Translated: part of the liquidity supply hangs on the credit rating — at exactly the spot where it would hurt, because a downgrade would typically come when the business is already struggling. The receivables sale also has running costs: $4.7 million in the first quarter of 2026 alone, booked in administrative expenses. The program is a legitimate financing tool — but anyone valuing Fortrea's cash of $147.5 million (March 31, 2026) should know that $300 million of future payment inflows have already been sold ahead of it.

Deep in Note 16 of the annual report sits an episode that throws a spotlight on the balance of power in the contract research business: a customer's clinical trial ran into a problem caused — the report says so expressly — by a third-party provider unaffiliated with Fortrea. Fortrea nevertheless granted the customer concessions, discounts and other benefits worth a total of $12.5 million "as part of a multi-party resolution": $8.7 million reduced 2023 revenue, another $3.8 million reduced 2024 revenue.

Legally, by its own account, Fortrea owed nothing — yet the company paid anyway, to keep the trials running. Translated: when a major customer calls, the service provider will, if need be, pick up the bill for someone else's mistakes. At a company whose ten largest customers account for 57 percent of revenue, that is not an anecdote but business logic.

On June 11, 2025 — the share price had at times stood at $4.94 in the June quarter of 2025, and a shareholder class action had been filed nine days earlier — Fortrea's board adopted a "poison pill" (stockholder rights plan) of limited duration: every share received a right to buy one one-thousandth of a Series A preferred share at $50.00. The mechanism arms itself as soon as an investor accumulates 10 percent or more of the shares — at which point all other shareholders may buy in at a steep discount, diluting the attacker's stake.

The threshold is what stands out: comparable defense plans — such as Sabre's of March 2026 — often kick in only at 15 percent. Fortrea laid its tripwires tighter, in a shareholder register that as of March 31, 2026 showed BlackRock at about 16 percent and hedge funds such as Corvex Management and Sessa Capital at about 5 percent each (source: fundamental data). Anyone speculating on a takeover premium at a bombed-out price should know: the board has put a weapon in the cabinet against exactly that scenario.

In March 2024 a shareholder filed a so-called derivative suit with the Delaware Court of Chancery — Hanna v. Paradise et al. — formally in the company's name against current and former officers, directors and major shareholders of Skillz. The accusation: in the secondary stock offering of March 2021 — near the all-time high of the SPAC euphoria — insiders sold shares while in possession of material non-public information, enriching themselves at the company's expense.

The suit is alive: in July 2025 the court converted the defendants' motion to dismiss into a motion for summary judgment and ordered limited discovery on the independence of a former director. For investors, the potential damages matter less (they would flow to the company itself) than the spotlight: the reckoning with the SPAC era of 2021 is still occupying the courts in 2026 — and the defendants carry the names of those who control Firy to this day.

Firy's balance sheet carries a position of $52.8 million under "non-marketable equity securities" — stakes in companies that are not publicly listed, carried at cost. That equals about 40 percent of Firy's entire market value (about $131 million, data as of July 8, 2026). Remarkable: the annual report (10-K) for 2025 names neither the name of the investee company nor its business — only that in 2025 there were no indications of impairment or observable price changes.

The value has sat unchanged in the books since at least the end of 2024. Whether a hidden treasure or dead capital lies behind it cannot be judged from the mandatory filings — and exactly that makes the position a genuine find: at a company of this size, a single unnamed balance-sheet item decides a substantial part of the substance.

Skillz grew up in San Francisco — but the group moved to Las Vegas. What remained of the old headquarters was a running lease. In fiscal year 2025 the company agreed with the landlord on an early termination: against a one-time payment of $14.0 million.

For scale: that equals roughly a fifth of the operating cash outflow of the entire year 2025 ($68.9 million) — so a substantial part of the cash burn was not ongoing business but the settling of a legacy from the era when Skillz was still valued in the billions and resided accordingly. Anyone extrapolating the 2025 burn rate into the future should know about this one-off effect.

Firy (then still Skillz) has been waging a remarkable campaign for years: the company sues competitors that advertise their money-gaming apps as fair contests between real players while, according to Skillz's account, computer bots actually compete against paying humans — steering tournament outcomes in the operator's favor. Against AviaGames, the campaign ended in April 2024 with a settlement worth $80 million: $50 million flowed immediately, plus $7.5 million per year over four years as a patent license fee.

The war goes on: a suit against Papaya Gaming has been running since March 2024, one against Voodoo SAS ("Blitz Win Cash") since July 2024 — both before the federal district court for the Southern District of New York, both over false "fairness" advertising. Papaya is now countering with counterclaims that in turn accuse Skillz of bots and reputational damage. For investors this is doubly remarkable: the litigation wins genuinely prop up the income statement ($7.5 million per year) — and at the same time the company's own business model lives on customers still believing the industry's fair-play promise at all.

Anyone who in 2026 reads pages about AI-assisted drug discovery at a research-driven oncology company will be surprised by Exelixis. In all six evaluated SEC filings (two annual reports 10-K, four quarterly reports 10-Q), artificial intelligence appears substantively only a single time — and not as a growth story, but as a cyber risk: "AI software is increasingly being used in the biopharmaceutical industry, including, in limited instances, by us."

Not a word about selling AI as a product, no quantified efficiency story, no AI revenue. The only further mention concerns the U.S. drug agency FDA, which has itself begun to use AI in reviewing marketing applications — for Exelixis a source of uncertainty about the duration of review procedures, not a business model of its own. In our company-specific AI classification, Exelixis is therefore rated "Neutral": the value of the company is decided by molecules and patents, not by algorithms.

Besides the patent clock, a second, quieter clock ticks at Exelixis — that of the U.S. drug-pricing law (the Inflation Reduction Act, IRA). Since 2022 the state health insurer Medicare has been allowed to negotiate a "Maximum Fair Price" for certain high-revenue drugs, that is, to enforce a capped price. For small biotech firms there is a temporary exemption — and Exelixis hangs precisely on it. The annual report states that the company received the small-biotech exemption for its cabozantinib franchise only through the price year (IPAY) 2027 and had to reapply for 2028.

In plain terms: cabozantinib could enter Medicare price negotiation from the end of the decade — on top of the generic pressure. For a product that accounts for practically the entire group revenue and for which older patients with kidney cancer are an important target group (that is, a large Medicare share), that is no side issue. Two state-timed risks — patent expiry and price cap — converge here on the same horizon, and both stand in the same report, just a few chapters apart.

Exelixis buys back its own shares on a large scale. The board authorized, in three steps, buyback programs totaling $1.75 billion: $500 million in August 2024, another $500 million in February 2025 and once more $750 million in October 2025. Through December 31, 2025 the company had already bought back 30.2 million shares for $1,159.7 million — at an average price of $38.39 per share. About $590 million of the most recent program is still open (through the end of 2026).

The other side of the same coin: cash including securities fell from $1.75 billion (end of 2024) to $1.66 billion (end of 2025) — despite a record profit of $782.6 million. The operating business brought money in, but a large part flowed straight back into its own shares. That makes earnings per share look prettier and signals confidence. But it is also a bet: whoever, in a one-product business with a patent timetable, thins out the war chest instead of hoarding it for the successor zanzalintinib or for acquisitions is relying on the transition succeeding and on the supply from the ongoing business never running dry.

That Exelixis earns almost everything with a single molecule is written large in every analysis. Less known is a second concentration risk that sits one level deeper — in distribution. The annual report soberly lists which individual customers account for more than ten percent of total revenue: affiliates of Cencora, Inc. with 22 percent and affiliates of McKesson Corporation with 19 percent in 2025. Together that is 41 percent of revenue over just two addresses — and the share has risen over the years (2024: 18 and 16 percent; 2023: 17 and 17 percent).

For investors this is no reason to panic: Cencora and McKesson are pharmaceutical wholesalers, not end customers — they distribute the drug to pharmacies and clinics, and demand comes from the cancer patients behind them, not from the wholesalers themselves. But the concentration means bargaining power on the other side and an operational risk: if one of these distribution channels stalls — through a payment dispute, logistics problems or a change in inventory policy — it hits a substantial part of revenue at a stroke. A one-product business that additionally flows out through two channels has two bottlenecks instead of one.

In November 2014 a whistleblower (in U.S. law: a "relator") filed a so-called qui tam suit in Illinois against Eli Lilly and Takeda. The allegation sounds like accountant trivia but concerns hard cash: certain credits to distributors should have been treated as retroactive price increases and included in the "Average Manufacturer Price" — the metric on which manufacturers' Medicaid rebates hang. Whoever sets the average price too low pays the government too little rebate.

In August 2022 — eight years later — a jury found for the plaintiff. In September 2025 the federal appeals court (Seventh Circuit) upheld the verdict, and Lilly booked a charge for it; in December 2025 the petition for rehearing failed as well. Eleven years from suit to checkout: a lesson in how long-lived price-calculation disputes are in the U.S. health system — and that in the end they land in the income statement even at a trillion-dollar company.

Between factory investments and dividends, the annual report hides a side job: as of December 31, 2025, Eli Lilly had about $902 million in not-yet-called commitments to venture capital funds outstanding — payable over up to ten years (still about $850 million as of March 31, 2026). The drugmaker is thus, on the side, a sizable venture-capital investor.

Strategically that is no accident: whoever sits early in biotech funds sees the takeover candidates of the day after tomorrow first. Together with the ongoing acquisitions — as of the quarterly reporting date, pending acquisitions of up to about $12 billion were already headed for closing in 2026 — the footnote shows how industrially Lilly refills its pipeline from outside. For investors that means: part of the celebrated research pipeline is not invented in the company's own lab but bought — and this supply reliably costs billions.

In the legal proceedings of the annual report, a chapter of company history lives on that has nothing to do with GLP-1 glamour: in the Brazilian town of Cosmópolis, Eli Lilly operated a factory from 1977 to 2003. Since March 2008 — for 18 years now — the labor prosecutor of the state of São Paulo has been litigating against the Brazilian subsidiary: employees were allegedly exposed to soil and groundwater contamination.

In 2014 the labor court ordered Lilly Brasil to take remediation and compensation measures, among them health care for an entire group of former employees and certain children plus a quantified fine; in 2019 the court even froze real estate of the subsidiary. Only in December 2025 did the highest labor court (TST) reduce the fine significantly — further appeals remain possible, and former employees are suing individually alongside. A lesson in how long environmental legacies can hibernate in the footnotes of a global company: longer than some product generations.

Deep in the accounting section of the annual report sits a table that explains the American drug-pricing system in two lines: for the most important U.S. programs alone (managed care, Medicare, Medicaid, chargebacks, patient assistance programs), Eli Lilly deducted $62.1 billion in rebates, discounts and returns from gross revenue in 2025 — after $41.5 billion the year before. For comparison: total reported group revenue was $65.2 billion, U.S. net revenue $43.5 billion.

Translated: the label on U.S. medicines shows roughly double what actually reaches Lilly — the list price is a shop-window price around which drugmaker, insurance middlemen and the government perform a complex rebate ballet. The accrued rebate liability grew to $15.1 billion by the end of 2025. Whoever reads about the moon prices of American medicines should know this footnote: between list price and net lies half a group revenue.

SEC filings have been submitted machine-readable for years — every number in the document carries invisible labels in the so-called iXBRL format so that regulators and databases can read them out automatically. For dollars, shares or percent there are standard units. But whoever looks into the source code of Deckers' annual report for fiscal year 2026 finds a self-defined unit of measurement named "deck:tannery" — the tannery as an official counting unit of a stock-exchange filing.

The reason is the dependence disclosed in the report on two tanneries in China that process the sheepskin for the UGG products: for that "2" to be machine-readable, Deckers had to give it its own unit — just as other firms define "number of aircraft" or "number of drilling rigs". It is a miniature at the margin, but a revealing one: when a global company has to create its own unit of measurement for its vulnerability, the mandatory filing has served its purpose. The article on the stock deals with the concentration risk itself — this technical detail behind it stayed out.

It is stated verbatim in the annual report: "We have not declared or paid any cash dividends on our common stock since our inception." — Since the company's founding, Deckers has never paid a cash dividend. All the surplus money flows into share buybacks — and their price history is a lesson on market timing inside one's own house.

The buyback table in Note 11 reads like this: in fiscal year 2024 Deckers bought its own shares at an average of $96.74 (split-adjusted), in fiscal year 2025 — at the peak of its flight — $567 million worth at an average of $149.21, and in fiscal year 2026, after the halving of the stock, $1.08 billion worth at an average of $102.43. Translated: even its own management did not see the summit coming and bought most expensively near the high. To its credit: instead of ducking after the crash, the board topped up the authorization on May 20, 2026 by $3.5 billion to about $4.84 billion — roughly a third of the market value. Anyone holding the stock should know: this company's "distribution" happens exclusively through the buyback button — at prices that were sometimes clever and sometimes expensive.

A remarkable footnote of the 2026 reporting season: Deckers has been paying additional import duties on shoes from Southeast Asia since the tariff wave of 2025 — and the annual report records that "certain tariffs imposed under the International Emergency Economic Powers Act have been invalidated by a recent US Supreme Court decision". The U.S. customs authority has even announced a staged process for refund claims.

And yet: "As of March 31, 2026, we have not recognized any amounts related to potential tariff refunds or other recoveries" — as of March 31, 2026 Deckers has recognized no amounts whatsoever for possible tariff refunds, because availability, timing and size are open. For investors this means: a potential one-off tailwind of unknown size slumbers in the balance sheet, which the company prudently values at zero — the rare kind of surprise that is more likely to turn out positive than negative. You should not count on it: the report warns in the same breath that future tariffs can just as well be newly imposed as refunded.

UGG sells sheepskin boots all over the world — yet the hide takes an astonishingly narrow path: it comes, per the annual report, "primarily from Australia" and is "processed largely by two tanneries in China" that meet Deckers' quality, volume and animal-welfare standards. The report calls the child by its name: "This geographic and supplier concentration exposes us to supply disruption risk."

For scale: the UGG brand generated $2.74 billion in fiscal year 2026 (ended March 31, 2026) — roughly half of group revenue. A substantial part of these products thus hangs on two processing plants in a country with which the United States regularly fights trade conflicts. Deckers hedges with fixed purchase contracts and itself writes that sheepskin prices have recently been stable — but a concentration risk that fits into two addresses is rarely documented as clearly as in this mandatory filing.

Anyone reading Cytokinetics' pipeline stumbles over a remarkable pattern: almost everything revolves around a single principle — the mechanics of the heart muscle. aficamten (MYQORZO) dampens an overly forceful heart muscle in oHCM; the sister candidate ulacamten targets a related form of heart failure (HFpEF); and omecamtiv mecarbil does the exact opposite — it strengthens a heart muscle that is too weak (HFrEF).

That is scientifically elegant and at the same time a concentration risk in slow motion: over two decades Cytokinetics has become the specialist for muscle mechanics, but precisely this focus means that a fundamental setback in the underlying biology could hit several programs at once. A company that bets everything on the same mechanism wins in depth — and loses in spread.

A detail that easily drowns in the celebration around the approval: while the price marked new highs, the company's own executives filed strikingly one-sidedly. Our data set (as of July 8, 2026) counts 20 insider sale filings (Form 4) and not a single purchase in the most recent period; the chief executive, too, stands in the statistics with "Sell".

By itself that is no scandal — many sales by biotech executives run automatically through pre-arranged plans, and after years without price gains, a partial sale after the big jump is humanly understandable. But it is a sober counterpoint to the euphoria: of all people, those who know the business best used the record prices to hand over shares — and not one of them bought more at this level.

Why did Cytokinetics lose a full $785 million in 2025 — considerably more than the operating loss of $612 million? A large chunk is not an operating expense but financial accounting: a "debt conversion expense" of $121.2 million. It arose because the company exchanged part of its convertible notes early for new paper and shares and had to book the difference as an expense.

For investors this is an instructive footnote: part of the red ink is produced not in the lab but on the capital market — through the contortions with which a perpetually loss-making research house pushes its maturities into the future. Such one-off items inflate the reported loss but say little about the underlying business. Anyone who looks only at the $785 million headline confuses refinancing bookkeeping with cash outflow.

When MYQORZO was approved in December 2025, part of its future revenue had already been given away — years in advance. Through the "RP Aficamten Royalty Purchase Agreement", the drug-royalty specialist Royalty Pharma had secured, against payments of up to $150 million (three times $50 million, tied to study starts), the right to 4.5 percent of worldwide annual aficamten revenue up to $5 billion (1 percent above that).

That is an elegant form of financing without classic dilution — and at the same time a silent permanent guest in the till: Cytokinetics carries an "RP Aficamten Liability" on its balance sheet for it, because the company itself has to generate the revenue stream out of which the participation is paid. Translated: of every dollar MYQORZO brings in going forward, four and a half cents go to an investor who never developed a single molecule. Anyone reading the revenue forecasts should factor in this diversion.

Bitdeer's power map of the world has a center of gravity hardly any investor would guess: the Kingdom of Bhutan. In Gedu (100 megawatts) and Jigmeling (500 megawatts), Bitdeer operates hydropower-fed mining data centers — together 600 megawatts, roughly a fifth of the entire 3-gigawatt portfolio. The land was originally leased from Druk Holding and Investments — the sovereign wealth fund of the Himalayan kingdom.

The arrangement is elegant for both sides: Bhutan turns surplus hydropower into foreign currency, Bitdeer gets cheap, low-carbon energy. But it couples a Nasdaq company to the politics of an absolute hereditary monarchy of 800,000 people — a location risk that shows up in no metric and disappears in the risk chapter of the annual report under "operations in multiple jurisdictions".

In the first quarter of 2026 Bitdeer lost $159.5 million on the bottom line, and the till was refilled with fresh convertible notes of $568.3 million. In the same quarterly accounts, tucked away under "other net losses", appears an item you do not expect at a company with negative operating cash flow: a donation of $5.0 million.

Who received the money, the report does not say — only that, alongside losses from derivatives and from redeeming convertible notes, it ranked among the largest single items of the $17.8 million of "other net losses". Already in the annual report for 2025, Bitdeer had specifically carved a one-off donation out of its adjusted metrics. Generosity is honorable — but whoever pays for growth with borrowed money ends up donating borrowed money, too.

SEC filings are normally written in the most cautious legalese. All the more remarkable what Bitdeer writes in its interim report for the first quarter of 2026 about its biggest pipeline site, Clarington in Ohio: the 570 megawatts are secured by contract with the local utility, but timeline and construction could be impaired by ongoing lawsuits from a neighboring firm — and then it gets personal: the neighboring firm American Heavy Plate Solutions is, per the filing, under far-reaching influence of MHR, a New York private-equity firm founded by Mark H. Rachesky.

That a company names the presumed string-puller of a neighborhood dispute in a mandatory filing is highly unusual — and shows how much hangs on this site for Bitdeer: at 570 megawatts, Clarington is the largest single item of the 1,259.5-megawatt pipeline and earmarked for the higher-margin colocation business.

In the footnotes of Bitdeer's annual report for 2025 stands a construction you have to read twice: "substantially all" of the group's cryptocurrencies sat in custody at BIT Group (named "Matrixport" until March 2026) in 2023, 2024 and 2025, and purchases and sales also ran "primarily from and to BIT Group". BIT Group is not a neutral bank but, per the report, a firm over which Bitdeer's controlling person has significant influence — Jihan Wu, Bitdeer's founder and chairman, is at the same time co-founder and chairman of BIT Group.

The same BIT Group is also Bitdeer's house bank for emergencies: a secured credit line of up to $400 million (8.35 percent interest, bitcoin as collateral), another of $200 million — and, since February 2026, a bitcoin borrowing raised within weeks from 800 to 6,000 bitcoin. The annual report itself calls the bundle by its name: a concentrated counterparty risk — if BIT Group fails, cash, loans and crypto holdings all hang on the same hook.

Bit Digital's board of directors has included Brock Pierce since October 2021 — a co-founder of Tether, Block.One and Blockchain Capital, and one of the more colorful figures in the crypto world. The proxy statement (DEF 14A) lists him as an independent director on the audit, compensation and nominating committees.

In July 2025, Bit Digital swapped ETH worth $19.1 million into 4,719 LsETH (a liquid-staking token from Liquid Collective). The token price fell, and the company wrote down $6.0 million — then swapped the entire position back into ETH in November 2025. An expensive four-month detour that sits in the 2025 income statement as "Impairment on digital intangible assets."

Bit Digital used to be called Golden Bull Limited and brokered peer-to-peer loans in China; from February 2020 to June 2021 the company also mined bitcoin in the People's Republic. The annual report (10-K) for 2025 still carries a risk factor to this day noting that Chinese authorities could impose fines for the legacy business — the statute of limitations could extend from two to five years, the report states, should the former mining operation be classified as a threat to financial security.

Whoever takes Aurinia for a thoroughly American company is only partly right. On the cover page of the annual report, the state of incorporation is not Delaware or California but Alberta, Canada — with a registered address in Edmonton. The commercial heart, by contrast, beats in Rockville in the U.S. state of Maryland, from where Aurinia sells LUPKYNIS through its own U.S. subsidiary (a Delaware corporation).

And a third place joins in: for manufacturing, Aurinia maintains, per the report, its own facility in Visp, Switzerland. This three-country constellation — Canadian corporate seat, U.S. distribution, Swiss production, plus partner Otsuka for Europe and Japan — is more than a footnote for investors: it means currency, tax and regulatory borders running straight through the value chain of a company that outwardly presents itself as a pure U.S. Nasdaq stock.

A biotech that has only just become sustainably profitable buys back its own shares on a grand scale? At Aurinia that is exactly the case. In February 2024 the board approved a buyback program of $150 million. On July 31, 2025 it added another $150 million — in total, then, $300 million for repurchasing its own shares.

For a company with about $398 million in cash and investments (end of 2025), that is a self-confident capital decision: instead of hoarding the whole cushion for the pipeline (the BAFF/APRIL inhibitor aritinercept) or for building a second leg to stand on, a substantial part flows back to shareholders. That makes earnings per share look prettier and signals confidence — but it can also mean that management currently sees no better use for the half billion of capital it raised than its own stock. For a one-product business with a patent expiry in 2027, that is a bet that the till stays amply filled even after the expiry date.

The headline for fiscal year 2025 sounded brilliant: $287.2 million in net income — almost fifty times the prior year ($5.8 million). But whoever reads the income statement one line higher finds the sober amount: pre-tax income stood at $114.2 million. The rest — a tax benefit of $173.0 million — is not a drug sold but an accounting entry.

It arose because, after years of losses, Aurinia released the valuation allowance on its deferred tax assets: a company that has long written red numbers may recognize the future tax advantages arising from them on its balance sheet only once profits become probable. Exactly that happened in 2025 — and the catch-up effect landed in the profit in one stroke as income. Remarkable compared with many another biotech: at Aurinia the operating core is genuinely profitable regardless (operating income of $104.9 million). But for the price-to-earnings ratio, the headline is what counts: a value filter that bluntly looks at the last reported earnings therefore considers the stock "cheaper" than the day-to-day business supports. The bonus flows exactly once — in the first quarter of 2026, Aurinia was already paying normal taxes again.

In the legal-proceedings section of the annual report stands a date that means everything for a one-product company. In February and March 2025, Aurinia received a so-called paragraph IV notice from no fewer than eight generic drugmakers — the formal announcement that they have asked the U.S. drug regulator FDA to approve a copycat version of LUPKYNIS (an ANDA). The names read like a who's who of the generics industry: Hikma, Lotus, Galenicum, Zydus, Teva, Dr. Reddy's, DifGen and Sandoz.

Aurinia has filed a patent-infringement suit against each of these applications within the deadline. Under U.S. drug law (Hatch-Waxman), that triggers an automatic stay: the FDA may clear the copycats at the earliest 7.5 years after the original LUPKYNIS approval — unless a court invalidates the patents sooner. For investors this is a double signal: first, somebody considers LUPKYNIS lucrative enough to want to copy it eight times over. Second, the race against the expiry date is officially open — and the core patent on the active ingredient voclosporin runs only until October 2027 anyway.

The collaboration with Gilead is a construction kit of payments whose sizes astonish. The annual report lists, among others: an option fee of $150 million per program, should Gilead want to add another Arcus program to the collaboration before the deadline; $45 million per program on exercise of the license option after completion of certain preliminary studies; and an option-extension payment of $100 million that flowed in the third quarter of 2024.

Such numbers show how valuable the mere right to decide later is. For Arcus they are oxygen that fills the till without a drug having to be sold. But for the investor they are also a reminder: a substantial part of the revenue is not product demand, but the pricing of options by a single partner.

In the cash statement of the annual report hides a sentence easily overlooked: to conserve funds, the company "paused" the further development of etrumadenant in third-line metastatic colorectal cancer, in order to concentrate resources on the late-stage portfolio. Etrumadenant was once one of the hopefuls of the early Arcus pipeline (a so-called adenosine antagonist).

For investors that is a useful reminder of how biotech really works: not every molecule from the pipeline graphic reaches the finish line — many get cleared away along the way, because the money is redirected to the most promising candidates. A broad pipeline is a stock of options, not a stock of certainties.

That a pharma giant takes a stake in a smaller biotech is common. How deeply Gilead Sciences is anchored at Arcus is surprising nonetheless: as of December 31, 2025, Gilead holds about 25.1 percent of the outstanding Arcus shares and, on the basis of an "Investor Rights Agreement", has sent three of its own designees to the board of directors. Together with executives and other major shareholders, insiders and block shareholders control, per the annual report, about 39.5 percent of the votes.

The report says itself what that means: these shareholders could, "acting together", exert "significant influence over all matters that require approval by our stockholders, including the election of directors". For free-float holders that means: on the company's fundamental course-settings, the most important partner sits at the same time on the longer lever — an alignment of interests that can turn into a conflict of interest the moment Gilead and Arcus once want different things.

The legal proceedings chapter of Apple's annual report for 2025 records an event that practically never happens at a company of this size: in the Epic Games case, the federal district court in Northern California found on April 30, 2025 not only that Apple had violated the court's 2021 injunction — it also referred the matter to the U.S. Attorney to determine whether criminal contempt proceedings are appropriate. In the original wording: "referred the Company to the U.S. Attorney for the Northern District of California for a determination whether criminal contempt proceedings are appropriate."

At issue was Apple's handling of the requirement to let developers point users to payment options outside the app. The appeals court partially defused the matter in December 2025 — Apple may, in principle, again charge a commission on "link-out" purchases, but must allow equal treatment of the payment options. The criminal referral itself remains a remarkable footnote find nonetheless: a court considered it possible that the most valuable company in the world had not merely misunderstood a judicial order, but defied it.

The quarterly report as of March 28, 2026 contains a sentence that made hardly any headlines: on February 20, 2026 the U.S. Supreme Court declared certain tariffs invalid that had previously been imposed on the basis of the emergency statute IEEPA (International Emergency Economic Powers Act of 1977) — and Apple is "seeking a refund of tariffs paid" under the procedures of U.S. Customs and Border Protection (CBP).

How much money that is, the filing does not reveal — but the annual report for 2025 had still explicitly named tariff costs as a drag on the product margin. The punch line: while trade policy keeps rolling out new instruments (the filing lists possible measures under Section 232, Section 122 and Section 301 in whole rows), the largest importer of consumer electronics may be getting part of its already-paid tariffs back from the government. A rare case in which the fine print hides good news — with the warning in the same paragraph that new tariffs can follow at any time.

A remarkable jump hides in the balance sheet details of Apple's quarterly report as of March 28, 2026: intangible assets (gross) grew within six months from $24.95 to $37.77 billion — a plus of $12.8 billion or a good 50 percent, without any acquisition being reported and without a corresponding outflow in the cash flow statement. In parallel, the "other purchase obligations" (among other things for licensed intellectual property and content) nearly doubled — from $14.8 to $30.4 billion — and other non-current liabilities rose from $41.5 to $55.5 billion.

Translated: within half a year, Apple booked usage rights on the scale of a DAX corporation — payment comes later, names come never. Neither counterparties nor purpose appear in the filing; the note discloses only the bare sums. In a half-year in which Apple's research spending jumped by a third because of "infrastructure costs" and the company added a standalone AI risk factor for the first time, that is a strikingly large, strikingly quiet position. What exactly was licensed will likely only show in later filings.

The balance sheet of Apple's annual report for 2025 contains a line you would not expect at a company with $112 billion in annual profit: "Accumulated deficit" — a shortfall of $14.3 billion as of September 27, 2025 (a year earlier: even $19.2 billion). Where other corporations show the retained earnings of decades, Apple shows a minus.

The reason is not a loss but the buyback machine: over the years, Apple has paid out more to its shareholders than it earned — from fiscal years 2022 through 2025 alone, roughly $413 billion flowed into buybacks and dividends, against about $403 billion in net income over the same period. Equity shrank as a result to $57 billion at times — against $365 billion in total assets. Only the record run of fiscal year 2026 turned the line positive again ($12.4 billion as of March 28, 2026). A curiosity on the side: it is precisely this depleted equity that makes metrics like return on equity (over 140 percent) look more spectacular than the business could ever be.

In February 2026 the quarterly report announced one of the largest funding rounds in startup history: the robotaxi subsidiary Waymo received $16.0 billion in fresh capital. The decisive half-sentence follows right after: "the significant majority of which was funded by Alphabet" — the far greater part was paid by Alphabet itself. The group essentially raises money from itself, and the valuation round simultaneously sets the benchmark at which the stake is carried on the books.

The bet on the future is expensive either way: the "Other Bets" segment, which includes Waymo, lost $7.5 billion operationally in 2025 on just $1.5 billion in revenue — the loss also grew because a valuation-linked compensation component for Waymo employees became more expensive: the higher the Waymo valuation, the higher the personnel expense. In the first quarter of 2026 alone the Other Bets loss added another $2.1 billion. Robotaxis are by now genuinely driving through American cities; per the mandatory filings, though, the segment is still not profitable.

Google Fiber was once the project meant to teach the U.S. telecom giants fear — fiber internet from the search engine company. The latest quarterly report now records the quiet farewell: in March 2026 Alphabet agreed to contribute its GFiber stake to a newly formed company. At closing the group receives $1.5 billion in cash, a $2.0 billion receivable — and keeps only 49.99 percent. As early as March 31, 2026 GFiber was reclassified as "held for sale"; roughly $6.8 billion in property and equipment is affected.

The timing is remarkable: in the very year Alphabet is building more infrastructure than ever before, it parts with the infrastructure bet of the first generation. The message between the lines: capital flows to where the AI return is presumed — and a consumer fiber network apparently no longer belongs there. For "Other Bets" watchers it is the second big signal of portfolio pruning in the race for AI capital.

Hidden in the annual report's chapter on lease obligations is a remarkable arrangement: in January 2026 Alphabet signed a power purchase agreement that is to be booked as a lease and, depending on the contract terms, triggers payments of $9.9 billion between 2027 and 2047. The twist is in the subordinate clause: if certain project conditions are not met, Alphabet instead pays a one-time sum of roughly $3.5 billion — and takes ownership of the power generation assets.

A search engine company with a contractually built-in power plant takeover: it is hard to state more clearly on a balance sheet that electricity has become a strategic raw material in the AI age. Fittingly, in December 2025 Alphabet acquired the data center and energy infrastructure provider Intersect for $4.8 billion. The days when Big Tech simply bought electricity from the socket are over — now the power plant is secured along with it.

In the derivatives note of the latest quarterly report sits a business you would not expect at an advertising company: Alphabet guarantees the lease and loan obligations of third-party data center operators — booked as credit derivatives. At the end of 2025 the maximum payment obligation from these guarantees stood at $16.9 billion; by March 31, 2026 it was already $28.4 billion — plus $9.0 billion in financial guarantees for energy infrastructure companies. And it keeps going: in April 2026, per the filing, new data center guarantees of roughly $15.3 billion were added. Terms: up to 15 years.

Economically this means: the biggest tenant of the AI boom insures its own landlords' creditworthiness — so that third parties can build the very data centers that will ultimately be needed for AI compute load (including Alphabet's own). In the updated risk chapter the company names the flip side itself: in the event of defaults or an industry crisis it would face additional liabilities and "excess capacity that we cannot easily redeploy." Anyone who wants to understand the circuits of AI financing will find one of its quietest and largest arteries here.

Almost every large U.S. public company is legally domiciled in Delaware — the state is the default for corporate law. Affirm quietly left that behind: effective July 1, 2025, the company reincorporated from Delaware to Nevada (10-K, fiscal year 2025). The operating headquarters stays in San Francisco; this is purely about which corporate law applies.

Nevada is considered more management-friendly: its law limits the liability of executives and directors more tightly and tends to give shareholders less leverage to challenge corporate decisions than the case law developed in Delaware. Combined with the dual-class structure through which founder Max Levchin controls about 44.4 percent of the votes, the move shifts the balance of power a bit further toward management — a detail that rarely makes headlines but touches the rights of outside shareholders.

Tucked into the annual meeting proxy statement (DEF 14A of October 24, 2025) is an unusual pay package: founder and CEO Max Levchin receives a "Value Creation Award" tied exclusively to the share price. The ten tranches only vest once the stock clears staggered hurdles — from $65.66 up to $371.91 per share (measured against the $49 IPO price). As of June 30, 2025, the first four of ten tranches had been "earned."

That's a notably double-edged design. On one hand, Levchin gets nothing unless the price rises substantially — a clear alignment with shareholder interests. On the other hand, the package is built to push a founder who is already highly influential (about 44.4 percent of the votes via shares carrying 15 votes each) even harder toward driving the price up — and every hurdle cleared creates new shares, meaning dilution for everyone else. An incentive that fuels exactly the growth premium the market is already paying a rich price for.

How tightly Affirm is bound to its biggest partner shows up in an easy-to-miss footnote of the quarterly report (10-Q as of March 31, 2026): in November 2025, Affirm lowered the strike price of the warrants granted to Amazon from $100 to $63.06 per share. Warrants are the promise of being allowed to buy shares later at a fixed price; when that price drops, they become more valuable for the holder — at the expense of everyone else's shares. Affirm booked $37.7 million in additional expense for the remeasurement.

You can read that as the routine upkeep of an important business relationship. But you can also ask what it says about the balance of power when a partner responsible for about 22 percent of gross merchandise volume gets better terms on its stock options mid-contract. For investors, it's a quiet signal about who holds the longer end of the lever in this partnership.