On May 9, 2026, CNBC reported that Nvidia has committed more than $40 billion to AI equity deals in the first five months of the year, led by a $30 billion stake in OpenAI and a fresh wave of public-company investments. The headline is dollar size. The real story is that the company selling the picks and shovels is now also a meaningful shareholder in the people digging the mine, the people building the railroads to the mine, and the company making the rail. That changes how every business should think about its AI vendors.
What Nvidia Has Actually Bought
Nvidia's 2026 investment portfolio is no longer a side activity. CNBC counts at least seven multi-billion-dollar stakes in publicly traded companies and roughly two dozen rounds in private AI companies, on top of the OpenAI deal.
OpenAI: $30 billion, closed late February. Nvidia CEO Jensen Huang said the OpenAI investment "might be the last time" it puts money into the company before its IPO. The check was part of a $110 billion funding round alongside roughly $50 billion from Amazon and $30 billion from SoftBank. Unlike the original deal floated in 2025, this one is not tied to gigawatt-deployment milestones.
Corning: up to $3.2 billion, May 6, 2026. Per Nvidia's announcement and CNBC's reporting, Nvidia receives warrants for up to 15 million Corning shares at $180, plus a pre-funded warrant for 3 million shares worth roughly $500 million. Corning will increase US optical-connectivity manufacturing capacity by ten times and US fiber production by more than 50%, building three new factories in North Carolina and Texas. The strategic point is that Nvidia is shifting AI-rack interconnect from copper to glass fiber for co-packaged optics, which Nvidia says uses up to 20 times less power per bit.
IREN: up to $2.1 billion, May 7, 2026. Nvidia's release details a five-year right to purchase up to 30 million IREN shares at $70 per share. Together, the partners intend to deploy up to 5 gigawatts of Nvidia DSX-aligned infrastructure, anchored by IREN's 2-gigawatt Sweetwater campus in Texas. In a separate but related arrangement reported by CNBC, IREN signed a five-year, $3.4 billion deal to provide Nvidia with managed GPU cloud capacity for Nvidia's own internal AI workloads.
There are also disclosed bets in CoreWeave, Wayve, Mistral, xAI, and others, plus the private-company rounds. The pattern repeats: Nvidia takes equity in a company, and that company commits to buying or deploying Nvidia silicon.
The Circular Deal Question
The structural critique now has a name. According to NBC News and a Bloomberg analysis of AI circular deals, AI is increasingly funded through loops where chip and cloud vendors take equity in AI labs that then buy chips and cloud, while data-center operators take supply commitments from chipmakers and turn around to sell GPU capacity back to those same chipmakers. The IREN agreement is a textbook case: Nvidia gets warrants on IREN, IREN buys Nvidia hardware to fill 5 gigawatts, and IREN sells $3.4 billion of GPU capacity back to Nvidia.
The bear case is that this is vendor financing dressed up as strategy, the same dynamic that inflated the late-1990s telecom bubble. NBC News quoted analysts warning that the cycle "smells like you are pre-funding the purchase of your own GPUs and products." If end-customer demand for AI inference does not grow into the contracted supply, several billion-dollar revenue lines look inflated and several equity stakes lose value at the same time.
The bull case is the one Nvidia and asset managers like Janus Henderson have made publicly: in a market where the most advanced chips are scarce and capital is the binding constraint on data-center buildouts, supplier equity is how you align timing across the chain. Power, capital, and silicon are not pooling here by accident. The stakes are how a fabless chipmaker without its own foundry, fiber plant, or hyperscale data centers ensures the rest of the stack moves on its roadmap.
Our take: The bull and bear cases are both partially right, and that is the problem. Circularity does not need to be fraudulent to matter. The structure increases correlation across vendors who used to be independent, and it makes the system more sensitive to a slowdown in a single revenue line. AI buyers should plan for that without assuming a crash.
Why This Changes AI Vendor Diligence
For most businesses, "AI vendor strategy" still means choosing a model provider, a cloud, and maybe a SaaS layer. After this week, the relevant variable is no longer just which vendor you pick. It is which equity-linked cluster you are buying into.
Before this round of deals, an enterprise running Claude on AWS could reasonably model AWS, Anthropic, and Nvidia as three companies with overlapping interests but distinct cap tables. After Nvidia's $200 billion Google Cloud commitment from Anthropic, our prior analysis on compute concentration, and now the $40 billion Nvidia equity portfolio, the diagram is different. Nvidia has direct equity exposure to OpenAI, exposure through GPU sales to Anthropic, financial exposure to several specialized cloud operators, and a manufacturing partnership underpinning its own roadmap. Your "vendor" is increasingly a node in a network you cannot see from a single procurement screen.
Three concrete shifts in diligence follow.
Map equity, not just contracts. When you evaluate an AI vendor, ask who owns equity in them, who they owe long-term purchase commitments to, and whose capital expansion they are anchoring. Public 8-K filings, S-1 risk factors, and earnings calls are the cheapest sources. The question to answer is whether your vendor's strategic incentives are pointed at your problem or at servicing a financial relationship upstream.
Treat reported demand with skepticism. Headline AI revenue and backlog numbers are increasingly co-mingled with vendor financing. That does not make them fake, but it does mean that consensus growth rates are not a reliable input to your own roadmap. Build pricing scenarios that include a 20% to 30% softening in third-party demand, and verify that your contract terms still hold under those scenarios.
Plan for correlated outages. When suppliers and customers share equity, a stress event hits the whole cluster at once. A demand shock, a regulatory change, or a power-availability problem in West Texas no longer affects only one tier of your stack. Disaster recovery plans built around single-vendor failure modes are now insufficient for AI workloads. The principles in our piece on multi-cloud AI procurement apply with more force this week than last.
Where the Power Is Pooling
Strip away the dollar figures and the pattern is clear. In 2026, the binding constraints on AI are physical: gigawatts, fiber, packaging, and chip allocation. The companies that control any one of those constraints are accumulating financial stakes in the layers above and below them. Nvidia is the cleanest example because its balance sheet is large enough to do this in public, but Microsoft, Amazon, Google, Oracle, and SoftBank are running variations of the same play.
For business buyers, that pooling has two practical consequences. First, the actors with leverage over your AI roadmap are no longer just your direct vendors. They are the silicon, fiber, and power providers two and three steps upstream. Second, the most expensive mistake is treating AI procurement as a software decision. The teams winning in 2026 are negotiating compute, integration, and data architecture as a single bundle, often with help from advisors who can sit between the CFO's contract review and the CTO's architecture and connect the financial structure of the AI stack to the technical roadmap a business actually needs.
What This Does Not Mean
It does not mean Nvidia is a bad vendor. The performance lead is real, the roadmap is well-funded, and the equity strategy locks in supply that smaller buyers cannot lock in for themselves. Walking away from Nvidia silicon to avoid concentration risk usually trades a financial exposure for a capability deficit.
It does not mean OpenAI, Anthropic, or any single AI lab is in trouble. Anthropic's run-rate revenue crossed $30 billion in April, and OpenAI's commercial revenue has grown alongside its compute commitments. Real demand exists. The question is the ratio of organic to financed demand, which is currently unknowable from the outside.
It does not mean you should hold AI investment until the picture clears. It will not clear in 2026. The companies treating ambiguity as a reason to wait are losing ground to the ones treating it as a reason to negotiate harder, diversify deliberately, and own their integration layer. The lessons in our piece on why most AI projects stall between pilot and production apply here as well.
Key Takeaways
- By May 9, 2026, Nvidia had committed more than $40 billion in equity stakes across the AI stack, led by a $30 billion OpenAI investment and including up to $3.2 billion in Corning and up to $2.1 billion in IREN.
- The deals tighten supply for Nvidia's roadmap and align partners on co-packaged optics, gigawatt-scale data centers, and US-based manufacturing.
- Critics call the structure circular financing that risks inflating apparent demand. Defenders call it a virtuous circle locking capital, supply, and demand together in a capacity-constrained market.
- For AI buyers, vendor diligence now needs to include equity relationships, not just contract terms. Headline demand figures should be stress-tested against softer third-party demand scenarios.
- The binding constraints on AI in 2026 are physical, and the companies controlling them are accumulating financial stakes throughout the stack. AI procurement is no longer a software decision.
The businesses that move early on AI vendor strategy will have a meaningful advantage. If you want to be one of them, let's start with a conversation.