The AMD warrant charge in its third-quarter 2026 results points to a bigger change in the AI chip market: suppliers are competing for long-term capacity commitments, not just individual hardware orders. AMD recorded a $1.097 billion expense from revaluing warrants issued to OpenAI and Meta. The charge did not represent cash paid to either customer, but it does show that the agreements carry a measurable financial cost for AMD.
AMD's November 3 earnings release lists the change in fair value of warrant liabilities at $1.097 billion, or 66 cents per diluted share. The company says the quarterly remeasurement reflects changes in its stock price and other valuation assumptions. The line appeared in the reconciliation between GAAP and non-GAAP results. AMD reported $2.428 billion in GAAP net income and $3.471 billion in non-GAAP net income for the quarter. The warrant revaluation was one of the items excluded from the non-GAAP measure.
Why AMD attached stock warrants to GPU supply deals
A warrant gives its holder the right to buy shares at a set exercise price if the agreement's conditions are met. AMD issued OpenAI and Meta warrants for up to 160 million shares each, at an exercise price of one cent per share. These are not shares handed over at signing. The warrants vest in stages as customers meet GPU purchase milestones, with additional stock-price, technical, and commercial conditions applying before exercise.
The structure connects the customers' potential equity benefit to the scale of their AMD deployments. OpenAI's agreement, announced in October 2025, covers a planned deployment of up to 6 gigawatts of AMD GPUs. Meta's February 2026 agreement also targets up to 6 gigawatts. In both arrangements, the first gigawatt is tied to AMD's Instinct MI450 series.
That design gives AMD a way to compete for large, multi-year infrastructure programs. Instead of relying only on product performance and pricing, it can offer customers an additional financial incentive that becomes valuable only if purchases and other milestones progress. For OpenAI and Meta, the agreements provide another route to secure future accelerator capacity as demand for AI infrastructure grows.
The accounting charge is not the same as cash spending
The $1.097 billion figure is a change in the estimated fair value of a liability. It is not a cash transfer, a $1.097 billion discount on GPUs, or evidence that AMD has already issued the warrant shares. When AMD's share price or other valuation inputs change, the estimated value of outstanding warrants can change too. That movement can affect reported GAAP earnings before the warrants vest or are exercised.
The distinction matters when comparing AMD's headline profit with its adjusted results. Investors should not treat the warrant charge as an ordinary operating expense required to manufacture each GPU. But they should not ignore it either. It reflects the potential economic value AMD has promised as part of customer agreements, and it can make reported earnings more volatile as the share price and valuation assumptions move.
The trade-off for AMD shareholders
If all warrant conditions are eventually met and the customers exercise their rights, AMD could issue up to 320 million shares across the two agreements. The final outcome is not guaranteed: vesting depends on purchase milestones and additional conditions, and the warrants may not all become exercisable. Still, the maximum share count is large enough that existing shareholders should track the terms alongside GPU shipment and revenue growth.
The potential dilution is the price shareholders may pay for a stronger position in the AI accelerator market. Large customers can help AMD justify investment in products, software, manufacturing capacity, and supply-chain commitments. But a purchase commitment is not the same as revenue already earned. AMD must deliver the systems, customers must meet the milestones, and the deployments must perform well enough to support continued buying.
Why the deals matter beyond AMD
The agreements also reflect how AI infrastructure buyers are spreading their dependence across suppliers and capacity providers. Meta is developing a business around selling compute as well as consuming it, as explored in Saganote's coverage of Meta's plan to build a cloud business around AI compute. Its reported discussions to supply large amounts of compute to Anthropic show how infrastructure capacity can become both an internal resource and a commercial product.
That demand is not limited to one supplier or one model company. Saganote has also covered Anthropic's talks to secure up to $10 billion in compute from Meta, as well as Google's decision to restrict Meta's access to Gemini as compute capacity tightened. Together, these developments show why major AI companies are arranging capacity years ahead and why chipmakers are willing to negotiate unusual commercial terms to win those customers.
AMD's own results show that the underlying business is growing: third-quarter revenue reached $13.4 billion, up 45% year over year, while Data Center revenue hit a record $8.5 billion, up 97%. The company expects fourth-quarter revenue of about $17 billion, plus or minus 3%. Those operating figures are the more direct measure of current demand. The warrant charge, by contrast, is a reminder that securing future demand can carry costs that appear in the accounts before all the related GPU sales arrive.
The key question for AMD is therefore not whether the $1.097 billion charge was paid in cash. It was not. The question is whether the OpenAI and Meta deployments turn into sustained GPU shipments, profitable data-center revenue, and durable customer relationships that justify the potential dilution. AMD's third-quarter 2026 results provide the financial baseline; future shipment milestones will show how much of the agreements becomes real business.
