Cathie Wood’s ARK Holds a Quarter of the AI Firms She Says Will ‘Win’

MakeCryippreferred on

Cathie Wood said on Aug. 9 that open-weight AI models, the free, downloadable kind like DeepSeek or GLM, are “ironically” going to push the “vast majority of model-driven AI revenue” toward OpenAI, Anthropic, and SpaceXAI, the AI arm folded into SpaceX. Her fund, ARK Invest, holds sizable positions in all three.

Wood argues open-weight models make enterprises easier to attack, which forces them to keep paying frontier labs for defense. ARK’s own portfolio disclosure shows meaningful stakes in SpaceX, OpenAI, and Anthropic, the three companies she says will win. The claim arrives after weeks of ARK buying SpaceX stock through a share-price slump, and days before two of the three companies head toward their own IPOs.

The Hidden Stake

Wood’s post doesn’t mention that ARK Venture Fund (ARKVX) is a direct holder in all three companies she names. ARK’s own portfolio disclosure puts SpaceX, OpenAI, and Anthropic at roughly a quarter of the fund combined, about 26%, though the fund doesn’t break out exact weights by name on that page.

Talking your own book isn’t illegal, and plenty of fund managers do it openly. But it does mean the tweet isn’t a neutral third-party read; it’s an investor describing why her own largest bets should pay off.

The Security Argument

The argument traces back to ARK Chief Futurist Brett Winton, who framed it two days earlier:

“People using open weight models to try to steal from and exploit enterprises are going to force companies to continually spend at the frontier to protect themselves.”

There’s real data behind that. The UK’s AI Security Institute reported on July 17 that GLM-5.2, an open-weight model released in June, now matches the cyber capability of frontier closed models from four to seven months earlier, a gap AISI says has been narrowing from a wider stretch in 2025, though the precise year-ago baseline wasn’t something this piece could pin down directly. If open models keep closing that gap, enterprises have less room to trust them with sensitive systems and more reason to pay a premium for whichever lab can promise better guardrails.

Karp’s Opposite Bet

Palantir CEO Alex Karp makes close to the opposite argument. He argues enterprises renting frontier models from OpenAI and Anthropic are paying a “wealth tax,” buying “tokens that create no value” while handing over their own data and business logic to a third party.

Karp’s pitch is that the real value sits with whoever controls deployment, not whoever built the model. That happens to be exactly what Palantir sells.

Anthropic itself lands somewhere in between: it won’t back an outright ban on open-weight models, but its own July 27 position statement warns they “potentially present a higher risk than closed models, because it is very difficult to apply guardrails to them.” Anthropic was also a notable no-show on Nvidia’s Open Weights letter, an industry push that ultimately drew roughly 50 signatories, including Microsoft, Meta, and, after an initial round without them, OpenAI and Google.

A Timeline of Escalating Bets

This wasn’t Wood’s first bullish call on these names, and it landed at a specific moment in ARK’s own trading record.

SpaceX shares have fallen 45% from their record high and were trading only fractionally below the stock’s $135 IPO price as of late July, a slide that started weeks earlier: ARK was already buying SpaceX shares below their $135 IPO price on July 18. ARK’s own trade disclosures, not a secondhand recap of them, show it also bought 229,651 SpaceX shares on July 27 and 28, part of a $47.3 million purchase that also included Tesla stock. Around July 22, Wood had called SpaceX “the most important company in history” while ARK was “buying the 38% dip.”

The Aug. 9 tweet arrived 12 days after that late-July purchase and roughly 18 days after the “most important company” line. It’s the latest entry in a string of increasingly bullish public statements made while the stock was struggling, not recovering.

It’s also not the first time Wood has framed a related deal in maximalist terms. In May, she called xAI’s compute deal with Anthropic a “brilliant” move that would turn “massive losses” into “significant profitability” for what she then called SpaceXAI. That post also drew attention that week for mistakenly tagging actress Gwyneth Paltrow instead of SpaceX president Gwynne Shotwell.

The maximalist framing isn’t limited to AI holdings, either. Around the same stretch, Wood was making an equally bold case on the other side of ARK’s book, arguing that global instability could ignite Bitcoin’s next major rally, a separate thesis running on the same confident register as the one now being applied to SpaceX, OpenAI, and Anthropic.

Two IPOs Will Test the Thesis

Two of the three companies in Wood’s tweet have their own reckonings coming. Anthropic is said to have filed for an IPO near a $1 trillion valuation, and OpenAI is said to be targeting a September listing, though neither number traces to a filing or company statement this piece could check directly.

If Wood’s thesis holds, both would need to price at multiples that validate exactly the “frontier defense” premium she’s describing. If it’s mostly narrative support for a fund that needed one, the tell will be whether ARK’s public framing of these three names keeps escalating specifically during price weakness, or holds steady once SpaceX stock and the two pending IPOs stop being a live concern.