Groq Halves in Value; Big Tech's $3T Hidden AI Tab

Groq’s new funding round puts a spotlight on how quickly AI chip startup valuations can swing, while a fresh accounting analysis reveals that Big Tech’s real AI spending obligations dwarf what appears on any balance sheet.

Key points

• Groq has raised $350 million at a $3.5 billion valuation, led by Dallas-based Disruptive, with Nvidia also participating. That valuation is roughly half what the company was worth before Nvidia licensed its technology and hired away much of its talent.
• Nine major tech companies, including Alphabet, Meta, Amazon, Microsoft, Oracle, and Nvidia, carry roughly $3 trillion in off-balance-sheet AI commitments, according to a Wall Street Journal analysis. Traditional capex from the same group totalled about $600 billion over the past year.
• Anthropic’s upcoming IPO valuation hinges on a $190-200 billion revenue forecast for 2028, with Wall Street benchmarking it against Palantir and SpaceX multiples.
• Thrive Capital’s 2022 early-stage fund, which made early bets on OpenAI, SpaceX, and Anduril, has grown sevenfold to $3.7 billion as of end of June.
• The UK’s AI chip shortage is contributing to inflationary pressure, with July inflation expected to hit 2.9%, the first acceleration in four months.

What happened to Groq?

Founded in 2016, Groq built a reputation as one of the more credible challengers in AI inference chips. Its latest round tells a complicated story. The $3.5 billion valuation is roughly half the company’s previous peak, a direct consequence of Nvidia licensing its technology and, according to the Yahoo Finance report, hiring away much of its talent.

The twist is that Nvidia is now also investing in the very round that reflects Groq’s diminished standing. That structure, where the acquirer of a startup’s IP and people then backstops its fundraise, is unusual. It suggests Nvidia sees value in keeping Groq operational, perhaps as a distribution channel or to maintain competitive optics in inference, rather than absorbing it outright.

For investors tracking the AI chip ecosystem beyond Nvidia and AMD, Groq’s situation is a reminder that second-tier chip startups face an asymmetric fight. Talent and technology can be licensed or hired away faster than capital can be raised to replace them.

How big is Big Tech’s real AI bill?

The off-balance-sheet analysis from the Wall Street Journal deserves serious attention. The nine companies examined reported roughly $600 billion in traditional capex over the past year. Their actual forward obligations tied to AI infrastructure, covering data centre leases, chip purchase commitments, and energy agreements, total approximately $3 trillion.

The breakdown matters. Uncommenced lease obligations alone hit $1.2 trillion, roughly four times the level from a year earlier. Chip and hardware purchase commitments across the group stood at $1.9 trillion. Alphabet’s commitments reached $811 billion as of June, more than double the $332 billion reported just three months prior, spanning deals that in some cases run to 2054.

Under current accounting rules, these obligations stay off the balance sheet until leases begin or hardware is delivered. Morgan Stanley accounting analysts flagged in April that the growing complexity of these arrangements makes it increasingly difficult for investors to gauge true leverage. That concern is not abstract. Alphabet and Amazon have both recently reported negative free cash flow, meaning operating cash is already insufficient to cover spending before these additional trillions are factored in.

The commitments are largely non-cancellable. If AI demand growth disappoints, these companies cannot simply walk away. That asymmetry is what makes the $3 trillion figure worth tracking, not just as a sign of confidence in AI, but as a structural risk that does not show up in standard valuation screens.

Does the Anthropic IPO math hold up?

Wall Street is sizing Anthropic’s IPO using forward revenue multiples anchored to a $190-200 billion revenue projection for 2028. Palantir and SpaceX are being used as comparable benchmarks. That framing raises an immediate question: both of those companies trade at multiples that reflect either defense-sector pricing power or private scarcity premiums that may not translate cleanly to a public AI software company competing against OpenAI, Google DeepMind, and others with vastly deeper capital bases.

The Thrive Capital data point adds context here. Thrive’s 2022 fund grew sevenfold largely on early positions in OpenAI and SpaceX, companies that benefited from private market scarcity and concentrated access. Once Anthropic is public, that scarcity disappears, and the revenue multiples will need to be justified by actual numbers, not forecasts anchored to 2028.

Nothing in this article is investment advice. This site is independent and not affiliated with any company or broker mentioned.