AMD's Data Centre Revenue Has Doubled, but the OpenAI Gigawatt It Promised Has Not Shipped

AMD reported second quarter revenue of $11.5 billion on 4 August 2026, a company record, with data centre revenue of $6.7 billion, up 107 per cent year on year and accounting for 58 per cent of total sales, according to AMD's results release. Chair and chief executive Lisa Su said data centre revenue more than doubled year on year and pointed to the ramp of the company's Helios rack systems in the second half. AMD guided third quarter revenue to between $12.7 billion and $13.3 billion.

The first gigawatt of AMD hardware for OpenAI is not in those numbers. It remains scheduled for the second half of 2026 and has not been deployed.

The agreement, announced on 6 October 2025, commits OpenAI to deploy up to six gigawatts of AMD Instinct GPUs across multiple generations, beginning with the MI450 series and an initial one gigawatt in the second half of 2026, according to AMD's and OpenAI's announcements. AMD issued OpenAI a warrant for up to 160 million AMD shares at one cent each, roughly ten per cent of the company, vesting in tranches: the first on the initial gigawatt deployment, the last requiring AMD stock to reach $600 as well as OpenAI meeting technical and commercial milestones.

The announcement moved the share price hard. Forbes recorded a gain of more than 30 per cent after the market opened on 6 October 2025, and Reuters reported a premarket surge of about 34.85 per cent. The projection of more than $100 billion in new revenue over four years from OpenAI and other customers was AMD's own, made by executives including data centre head Forrest Norrod and reported by Reuters and NBC News, rather than an independent analyst estimate. On a deal of this structure, where the buyer holds warrants over the seller's stock, the distinction between a vendor forecast and an outside one is not decorative.

What has shipped is the platform. At its Advancing AI event on 22 and 23 July 2026, AMD launched Helios, its first full rack scale system, built on the Instinct MI455X. AMD says a Helios rack links 72 MI455X GPUs into a single scale-up domain with 31 terabytes of unified HBM4 memory, 2.9 exaflops of peak FP4 compute and 260 terabytes per second of scale-up bandwidth. Su said the system is in full production with first shipments in September and a ramp through the fourth quarter, as reported by Fierce Network and Moor Insights and Strategy.

The customer list has widened well beyond OpenAI. On 24 February 2026, AMD and Meta announced a multi-year, multi-generation agreement covering up to six gigawatts of AMD Instinct GPUs, with shipments supporting the first gigawatt beginning in the second half of 2026 on a custom MI450-based GPU paired with sixth generation EPYC processors codenamed Venice. AMD says the Helios rack architecture was developed jointly with Meta through the Open Compute Project.

On 22 July 2026, AMD and Anthropic announced a partnership covering up to two gigawatts of MI450 series GPUs, with the first gigawatt coming online in the first half of 2027, alongside an AMD equity investment of up to $5 billion in Anthropic and an engineering collaboration using Claude to speed AMD's ROCm software work, per AMD's release and CNBC. Oracle has said it will deploy 50,000 MI450 GPUs from the third quarter of 2026.

Not all of AMD is expanding. Gaming revenue fell 31 per cent year on year in the June quarter, and Su attributed the decline to prices weighing on consumer demand, Tom's Hardware reported.

AMD has the platform in production and three hyperscale customers committed to fourteen gigawatts between them, but the OpenAI deployment that triggers the first warrant tranche has not happened. OpenAI's compute lead has said the company expects to begin deploying Helios racks later in 2026 and to accelerate through 2027. If that slips into next year, the first tranche slips with it, and the deal's headline structure, the one that moved the share price by a third in a day, remains unexercised. Whether AMD stock reaches the $600 the final tranche requires is a separate question that nothing in the current results answers.