One Deal Built Innventure's ~$540 Million Valuation. Its Absence From the 10-Q Is the Invoice.

On the evening of Aug. 13, Innventure (NASDAQ: INV) reported a quarter whose revenue was $953,000 — under $1 million for a company Wall Street had recently valued near $540 million. The bigger news sat in the 10-Q filed the same day, in the language of a project just crossed off a list: the deployment site behind its flagship deal was "no longer available," and the deal had been "removed from internal bookings." By the next close the stock was down 55%, and the lead plaintiff in a securities class action now has until October 27 to step forward. This is not a story the reader needs to take on anyone's word. It is a story about one number that had to hold an entire valuation, and the paper record of how it came apart.

The claim, and the number that resisted it

Innventure describes itself as a technology commercialization company: it licenses undeveloped technologies from large corporations and builds standalone businesses around them. Its most valuable subsidiary, Accelsius Holdings, makes two-phase, direct-to-chip liquid cooling for AI data centers. On November 17, 2025, Accelsius announced that DarkNX, which the company called a "global digital infrastructure company," had agreed to deploy Accelsius' NeuCool cooling across a new 300-megawatt AI data center campus in Ontario — a deal management billed as the largest two-phase, direct-to-chip deployment to date.

That single counterparty became the engine of the bull case. On the Q1 call in May, management projected Accelsius would exit 2026 at an annual revenue run rate around $100 million and reach cash-flow positivity by year-end. One analyst estimated the Accelsius stake was worth roughly 77% of Innventure's approximately $540 million market value. The stock, and the financing it enabled, rested on a story in which one customer would spend hundreds of millions on a technology that had produced about $1.5 million of 2025 revenue, mostly from demonstration units.

The claim and the counter-evidence never moved at the same speed. In Q2 2026, total consolidated revenue was $1 million, of which Accelsius supplied $0.9 million — 96%. Analysts had been looking for about $10 million. The net loss widened to $34.9 million for the quarter. The gap between the $100-million-run-rate narrative and a quarter that booked under a million dollars of actual sales was the number the story could not reconcile; the August filing stopped pretending it could.

The counterparty on paper

The question the whole case turns on is whether DarkNX was ever in a position to build anything. What is documented, from a short-seller report published May 28, is that the counterparty looked thin on every independent dimension a detective would check first.

DarkNX, according to the Morpheus Research report, was incorporated in November 2024 — a year before the deal was announced — from a single-family home in a Toronto suburb. Its nine LinkedIn-listed employees showed no data center experience, and several appeared to hold other full-time jobs. The alleged site in Mississauga was occupied by a trucking company, Day & Ross, which denied sharing the address with DarkNX or having heard of it. Dell, Schneider Electric, and Supermicro were each reported to have denied any partnership with DarkNX. Morpheus, which disclosed a short position, quoted a former Innventure executive saying management used "false information" and revenue projections that were "pure fiction" to solicit investment, and a former Accelsius employee saying, "There's no data center."

Now the discipline the case requires: none of that is proof. A short seller has a financial interest in the stock falling, and its report is an allegation, not a finding. On the other side of the ledger stands genuine evidence that the underlying technology is real — Accelsius raised $65 million in a Series B in January led by industrial giant Johnson Controls at a $665 million post-money valuation, and it has independent validation that its cooling keeps GPUs meaningfully cooler using far less coolant. A technology can be worthwhile while the single contract in the forecast is not.

What changed the weight of the short report was the company's own filing. Innventure had waved off the allegations in May but did not reconcile them. On Aug. 13, its 10-Q did the reconciling for it: the site was gone, the deal was out of the bookings, and management suspended its 2026 revenue and cash-flow targets for Accelsius and withdrew its 2028 consolidated cash-flow-positive target. CEO Bill Haskell attributed the shortfall to "evolving dynamics in the AI infrastructure market" and called 2026 revenue an "imprecise reflection" of progress — a framing that answered the disappointment but not the arithmetic. It also, notably, did not address the specific DarkNX facts the skeptic had raised.

Here is the shareholder invoice

If the lead isn't a fraud case, it is still an expensive one to have been carrying, and the cost shows up in two places.

First, dilution. By June 2026 the common share count had climbed roughly 41% over 15 months, to about 84 million shares, with more than 18 million warrants outstanding on top. Operating cash burn ran about $59.5 million in the first half of 2026, and cash on hand fell to $41.5 million from $60.4 million at the end of 2025. To fund that burn, management sold equity — including about $13 million drawn in Q2 through a standby equity purchase agreement at an average price of about $6.21 a share. Shares now trade near 85 cents, within touching distance of a 52-week low of $0.83 and down roughly four-fifths from the start of the year. Every dollar raised at $6.21 while the DarkNX story was live was effectively raising capital at a price the subsequent filing did not support.

Second, the compensation line. The Morpheus report alleges that in April, executives received roughly two million "earnout" shares valued near $9.2 million, a payout tied to Accelsius entering "binding contracts" providing more than $15 million of revenue — a threshold the report says was met through the DarkNX purchase order before that same order was stripped from bookings. The exact trigger language and what management knew is precisely what the litigation will test; the alignment question — that the executives who promoted the deal were also the ones who stood to be paid when it was booked — does not disappear if the court later clears them.

What the lawsuit does and doesn't establish

A securities class action was filed in the Southern District of New York on behalf of investors who bought between November 17, 2025, and August 13, 2026, alleging violations of Sections 10(b) and 20(a) of the Exchange Act. The complaint claims Innventure's 2026 revenue and cash-flow targets were overstated because the DarkNX deal was unlikely to materialize. That is an allegation in a pending suit; nobody has been found liable, and the company has not admitted wrongdoing. The one fact the company itself has placed on the record is narrow but real: the deal is gone from the books.

Priced against three endings, the position sorts itself. In the benign-but-persistent case, the cooling technology is real and the $65 million of third-party money in Accelsius is a genuine asset, but the parent is years from positive cash flow and now must fund a widening burn at far lower share prices — a slow and costly combination. In the persistent-but-lawful case, the deal was a bad booking that failed for market reasons, and the company still faces the dilution and the unwind. In the worst case — if the projections were knowingly unsupported and tied to compensation — the remedy is years away and the invoice includes not just dilution but the class's claimed losses. All three roads run through one fact the filing established: the single number that carried the valuation is no longer in the forecast.

The next document that moves this case one level is the one due when the class's lead plaintiff is chosen — or the company's own answer, if it ever confronts the DarkNX specifics directly rather than the weather metaphor. Until then, the investor's job is to notice what was true before anyone called it a fraud: a stock valued near $540 million on a counterparty whose claim to an Ontario data center could not be located on a map. That is the anomaly the lawsuit is now asking a court to characterize — and the reason a ninety-percent-plus decline, whatever the court decides, was not an accident.