AMD–OpenAI Deal: 6 GW of GPUs and a 160M-Share Warrant

The warrant mechanics

AMD and OpenAI announced a supply agreement for up to six gigawatts of AMD Instinct GPUs across several product generations, with the first gigawatt of MI450 systems arriving in the second half of 2026. Alongside it, AMD issued OpenAI a warrant to purchase up to 160 million shares of AMD common stock, roughly a tenth of the company, at an exercise price of one cent per share. Buying the full 160 million shares at that price costs OpenAI about $1.6 million; the value of the shares themselves depends entirely on what AMD trades for when they vest. [1][2]

Vesting is a double gate, not a calendar. AMD's own quarterly filing describes it as tied to both AMD Instinct GPU purchase milestones and stock-performance thresholds, meaning OpenAI has to actually take delivery of gigawatts of GPUs and AMD's share price has to climb before any tranche unlocks. The first tranche vests on delivery of the initial one-gigawatt MI450 deployment; later tranches vest as purchases scale toward the full six gigawatts. Independent analysis of the filing puts the top of the stock-price ladder at $600 a share for the final tranche, a level AMD's stock would have to roughly triple from its pre-deal price to reach. The warrant expires October 5, 2030. [2][7]

None of it has paid out yet. AMD's 10-Q for the quarter ended March 28, 2026, states plainly that as of that date, and as of the prior quarter's close on December 27, 2025, no shares under the OpenAI warrant had vested or become exercisable. The instrument that made headlines as a ten-percent stake giveaway was, five months in, still just a set of conditions nobody had cleared. [2]

Equity instead of a purchase order

Atlas interpretation: A warrant is a strange instrument to attach to a hardware order. AMD did not simply sell OpenAI GPUs on payment terms; it gave OpenAI a claim on AMD's own stock price, contingent on OpenAI buying more of those GPUs. The effect is to make OpenAI a shareholder in its own supplier, with a direct financial stake in AMD executing well and AMD's stock rising, on top of whatever it already wants the chips for. AMD's CFO described the arithmetic from the company's side: the partnership was expected to deliver tens of billions of dollars in revenue and be highly accretive to non-GAAP earnings per share, which is the language of a company deciding a large customer relationship is worth diluting existing shareholders to lock in. [1]

Atlas interpretation: The instrument is cheap for OpenAI to hold and expensive for AMD to have promised. A maximum exercise cost of $1.6 million against a warrant that outside analysts valued in the tens of billions of dollars at full vesting is not a hedge OpenAI is paying much for; it is AMD handing over a slice of future upside in exchange for a buyer committing to specific, verifiable purchase volumes. That trade only makes sense if AMD's alternative, selling the same GPUs to OpenAI on ordinary commercial terms without the warrant, was less certain to happen at the scale AMD wanted booked. [7]

How Wall Street reacted

AMD's stock moved immediately and sharply. Shares jumped as much as 30% or more in early trading on October 6 before settling to a close reported around 23% to 24% higher on the day, adding tens of billions of dollars to AMD's market capitalization in a single session. CNBC characterized the move as the market pricing in OpenAI effectively becoming one of AMD's largest shareholders; Fortune's coverage presented it explicitly as evidence that a major AI lab was moving to reduce its dependence on a single chip supplier. [3][4]

Atlas interpretation: The size of the reaction says as much about Nvidia's position as it does about AMD's. A single multi-gigawatt commitment from OpenAI, structured with equity attached, was large enough news to move tens of billions of dollars of AMD's valuation in hours, which is only remarkable because AMD had spent years being treated by the market as a distant second choice for frontier AI training and inference silicon. The deal did not make AMD Nvidia's equal; it made AMD a large enough alternative that OpenAI's spending could visibly reprice AMD stock, which is a lower bar than displacing Nvidia's installed base but a real one. [3][4]

The circular deal question

The AMD warrant landed two weeks after Nvidia's own announcement that it would invest up to $100 billion in OpenAI to fund OpenAI's purchases of Nvidia systems. Bloomberg grouped the two deals, plus OpenAI's cloud commitments to Microsoft and Oracle, under a single label that stuck through the rest of 2025: circular deals, in which a chip or cloud vendor's own revenue and, in AMD's and Nvidia's cases, stock price depend on the continued spending of a customer the vendor has also taken a financial stake in. [5]

The skepticism was specific, not vague unease. D.A. Davidson analyst Gil Luria described the pattern as an unhealthy layer of related-party transactions capable of artificially inflating valuations, and warned of deflating activity if investors came to see the ties that way. Oxford Economics cautioned that the scale of the commitments showed the companies involved were taking on significant risk, and that a sharp correction in tech stocks, with knock-on effects on the broader economy, would be likely if the productivity gains from AI spending fell short. Peter Boockvar of OnePoint BFG Wealth Partners put the exposure in blunter terms: OpenAI would need to generate huge revenues and profits to pay for the obligations it was signing, or the losses would be severe. [6]

Atlas interpretation: The revenue-to-commitment gap the skeptics pointed to was real and large. A month after the AMD deal, Sam Altman told investors OpenAI would end 2025 above $20 billion in annualized revenue, on its way to a projected hundreds of billions by 2030. Set against that, AMD's own estimate put the eventual value of the GPU purchases and associated ripple effects above $100 billion, and the AMD deal was only one line item alongside the Nvidia investment and OpenAI's separate cloud-compute commitments. None of that made the AMD warrant fraudulent or the demand fake; GPU capacity was genuinely scarce in late 2025, which is its own argument for locking in supply with financing attached. It did mean that a meaningful share of the revenue AMD and Nvidia were about to report, and the stock gains that followed, traced back to one customer's projected future income rather than to a broad base of independent buyers. [8][5]