Nvidia pulls the plug on OpenAI and Anthropic checks

Being the biggest backer of the two most dominant AI labs in the world sounds like the safest bet in tech, but Jensen Huang is ready to close the checkbook.

Speaking at the Morgan Stanley Tech, Media and Telecom conference in San Francisco on Wednesday, the Nvidia CEO revealed that the company's recent investments in OpenAI and Anthropic are likely its last. His stated logic was pretty simple: both startups are expected to go public later this year, and once an IPO happens, the window for this kind of strategic check officially shuts.

If you follow late-stage venture capital, that explanation does not really hold up. Tech giants routinely buy into pre-IPO rounds right up to the bell to capture quick public upside. Nvidia is already minting billions selling the very chips that keep both companies running, so it hardly needs to squeeze extra financial returns out of their equity.

The shrinking checks and circular math

Behind the scenes, the mechanics of these deals have started looking increasingly messy.

Back in September, Nvidia made headlines by pledging up to $100 billion to OpenAI. MIT Sloan professor Michael Cusumano famously called the deal kind of a wash in the Financial Times, pointing out the obvious circle: Nvidia gives OpenAI $100 billion in cash, and OpenAI hands right back $100 billion to buy Nvidia silicon.

Amid mounting scrutiny over round-tripping capital and broader AI bubble fears, that giant check shrank dramatically. When OpenAI finalized its $110 billion round last week, Nvidia's actual check came in at $30 billion. Huang publicly dismissed talk of bad blood between the two firms as nonsense, but the math spoke for itself.

Davos drama and Pentagon warfare

Meanwhile, the relationship with Anthropic has become an outright headache.

Just two months after Nvidia put $10 billion into Anthropic in November, Anthropic CEO Dario Amodei went on stage at Davos. Without naming Nvidia directly, he compared American chipmakers selling top-tier processors to approved Chinese buyers to selling nuclear weapons to North Korea.

Things escalated further this week. The Trump administration placed Anthropic on a federal blacklist, banning government agencies and defense contractors from using its models after the startup refused to allow its tech in autonomous weapons or domestic surveillance.

Within hours of the ban, OpenAI pounced, signing its own defense deal with the Pentagon. Anthropic blasted the move as mendacious, but the public backlash was swift. Within 24 hours, Claude skyrocketed from outside the top 100 in late January to the number one free app on the US iOS App Store, leapfrogging ChatGPT.

Nvidia suddenly found itself holding massive equity stakes in two bitter rivals that are weaponizing geopolitics against each other.

Why it matters

We think Huang's IPO excuse is just convenient cover for a hasty tactical retreat from a dumpster fire.

For enterprise teams and developers, this spat proves that model selection is no longer just about benchmarks or context windows. Anthropic's principled refusal of military contracts triggered a consumer surge for Claude, while OpenAI's instant Pentagon deal won defense cash at the cost of public trust. Choosing an ecosystem now requires navigating ideological and regulatory crossfire.

For the wider AI market, Nvidia's retreat signals the end of the circular financing era. The strategy of pumping vendor capital into startups just to guarantee GPU orders is drawing too much regulatory heat and creating unsustainable valuation bubbles.

Walking away right now is a masterclass in risk management by Huang. By freezing further checks before their respective IPOs, Nvidia secures its earlier ecosystem gains, avoids getting subpoenaed over military or China entanglements, and goes back to doing what it does best: selling pickaxes to everyone without caring who digs where.