Nvidia has scaled back the size of its planned investment in OpenAI after pushback from its own shareholders, according to The Decoder — a retreat that lands in the same week Anthropic put out growth numbers investors are reading as a rebuttal to AI bubble warnings.

The two data points sit awkwardly next to each other. One of the largest capital commitments in tech history — Nvidia's compute-tied pledge to OpenAI — is getting trimmed because shareholders don't like its risk profile. Meanwhile Anthropic, which built its balance sheet on strategic backers rather than vendor-financed compute deals, is putting up numbers strong enough to complicate the "this is all a bubble" narrative that has dominated AI commentary through 2025 and into 2026.

For teams building on top of either lab, the interesting part isn't the headline tension — it's what each company's trajectory signals about where capital discipline is showing up first, and where the market still isn't pricing risk correctly.

Why Nvidia's board got cold feet on OpenAI

Nvidia's OpenAI commitment was structured around compute: capital that flows toward OpenAI's data center buildout, a large share of which ultimately gets spent on Nvidia GPUs. That loop — a supplier investing in its own biggest customer, whose main use for the money is buying more from the supplier — is exactly the structure that critics of AI-sector accounting have flagged as circular financing. It lets both companies show impressive growth numbers without a matching increase in outside demand.

Investors pushing back on that structure tend to raise a consistent set of objections:

• Concentration risk: a huge bet on a single customer's future compute needs, regardless of that customer's size or reputation
• Circularity: revenue that partly reflects Nvidia's own capital going back out the door, not new external demand
• Governance exposure: Nvidia carrying OpenAI-specific risk on its balance sheet in a business it doesn't control

Shrinking the bet doesn't mean Nvidia is walking away from OpenAI as a customer — it means the company is under pressure to decouple "we sell chips to OpenAI" from "we underwrite OpenAI's growth with our own capital."

Anthropic's numbers cut the other way

The contrast, per the same report, is Anthropic. Its growth has been driven largely by enterprise API usage and coding tools — revenue that comes from customers paying for output, not from a chip vendor's balance sheet. That distinction matters to the bubble debate specifically because it isolates demand from financing structure: if usage-based revenue keeps climbing without a vendor recycling its own capital into the numbers, it's harder to argue the whole sector is propped up by circular deals.

None of this proves AI spending overall is rational or that every lab's valuation is justified. It shows one specific thing: Anthropic's growth is legible in a way that doesn't require unpacking who financed whom.

What builders should take from this

• Watch compute-availability assumptions. If Nvidia is tightening the scale of its OpenAI-linked buildout, don't assume the largest lab's compute roadmap is fixed — plan for possible delays or reallocation.
• Read vendor-investment deals as a risk signal, not a validation signal. A chip maker or cloud provider investing in the company you depend on can mean subsidized capacity today and dependency risk tomorrow.
• Favor revenue-per-customer over headline valuation when judging which platform to build on long-term — it's a better proxy for whether a vendor will still be pricing sanely in two years.
• Diversify infrastructure commitments rather than betting a roadmap on any single mega-deal continuing at its announced scale.

AiiN's takeaway

This isn't proof the AI bubble has burst, and it isn't proof it never existed. It's a sign that capital markets are starting to price the difference between demand-driven revenue and reciprocal-investment revenue — and that Nvidia's shareholders, at least, are no longer willing to treat the two as equivalent. In our estimation, expect more of these vendor-financing structures across the sector to face similar scrutiny in the coming quarters if usage-based revenue at labs like Anthropic keeps outpacing capex-linked commitments elsewhere.