AMD and Anthropic Replicate NVIDIA's Circular Vendor-Financing Playbook
The strategic alliance between Advanced Micro Devices, Inc. (AMD) and Anthropic PBC has emerged as a cornerstone of AMD's strategy to challenge NVIDIA's dominance in the AI chip market. In its Q2 2026 earnings report on August 4, 2026, AMD announced record quarterly revenue of $11.54 billion (+50.1% YoY), driven by a Data Center segment that more than doubled year-over-year. Central to this growth is AMD's newly launched Helios rack-scale AI platform, which integrates its EPYC Venice CPUs, MI450 series GPUs, and Pensando networking.
The 2-Gigawatt Anthropic Deal and Helios Ramp
During the Q2 earnings call, AMD CEO Lisa Su confirmed that Helios has entered full production, with initial shipments scheduled to begin in Q3 2026 and a substantial ramp projected for Q4 2026 and 2027. The centerpiece of the Helios launch is a massive strategic partnership with Anthropic:
- 2 GW Deployment: Anthropic has committed to deploy up to 2 gigawatts (GW) of AMD Instinct MI450 series GPUs in Helios racks.
- Timeline: Deployment of the first gigawatt is scheduled to begin in the first half of 2027.
- Hyperscaler Validation: Microsoft has also committed to deploy Helios at scale on Azure for frontier model inferencing across its AI services and customers.
CEO Lisa Su highlighted the competitive performance of the platform:
"Across a broad range of inferencing workloads, Helios delivers up to 15% more throughput at the same rack power and up to 30% more tokens per dollar than the competition... In addition to our multi-generation gigawatt scale deployments with OpenAI and Meta, we announced a new strategic partnership with Anthropic."
Anthropic's Dual-Track Strategy and Financial Strain
The 2 GW AMD deal reveals a highly complex, dual-track strategy by Anthropic to mitigate hardware vendor lock-in. Anthropic is currently paying SpaceX $1.25 billion per month ($15 billion per year) to lease Nvidia GPUs at the Colossus 1 data center. By simultaneously partnering with AMD for a massive 2 GW Helios deployment in 2027, Anthropic is attempting to build long-term leverage over its infrastructure providers.
However, this dual-track strategy places immense financial strain on Anthropic. Paying $15 billion annually to SpaceX (representing roughly half of Anthropic's current annual run rate) while committing to fund a multi-gigawatt hardware buildout with AMD in 2027 creates an incredibly high-risk capital structure that is highly dependent on continuous funding from its backers1, including Google and Amazon.
The SpaceX-Nvidia Exclusivity Blow
Despite reporting record earnings and the massive Anthropic contract, AMD's stock plummeted up to 10% (closing 7% lower on Wednesday, August 5) due to a major competitive setback. On the same night as AMD's earnings, Elon Musk announced on SpaceX's first public earnings call that SpaceX would build its AI infrastructure exclusively on Nvidia's next-generation Vera Rubin architecture.
This decision represents a direct reversal of Musk's prior stance of buying from both Nvidia and AMD. Because SpaceX was AMD's prior chip supplier, this exclusive commitment undercuts AMD's pipeline and deprives its MI450 chips of a high-profile validation win, illustrating how quickly hardware demand can shift in a highly concentrated buyer market.
-
An instance of Bypassing hyperscaler provisioning queues forces frontier developers to accept existential balance-sheet risk. — To bypass provisioning queues and secure compute, Anthropic is taking on massive financial commitments that place extreme strain on its balance sheet. ↩︎