Anthropic discloses a multi-platform compute strategy in its S-1, integrating Amazon Trainium, Google TPU, NVIDIA and AMD…

NVIDIA GPUs Are No Longer the Only Choice Source: Semiconductor Horizons Published: 2026-10-30 15:40 NVIDIA AMD Generate Poster Anthropic's multi-platform compute strategy disclosed in its S-1 filing integrates Amazon Trainium, Google TPU, NVIDIA GPUs and AMD GPUs, showing an attempt to break dependence on a single chip platform.

For a long time, NVIDIA GPUs have held an important position in the AI compute market thanks to their mature software ecosystem, powerful computing performance and complete system solutions. But as demand for AI model training and inference continues to grow, leading companies have begun exploring more diversified ways of supplying compute. With the multi-platform resource pool strategy newly disclosed by Anthropic, by integrating chip resources from different vendors, companies can not only expand their compute supply channels but also begin to have more flexible resource scheduling and procurement options.

From purchasing GPUs to multi-platform compute scheduling

Anthropic disclosed in its S-1 filing that it adopts a multi-platform resource pool strategy, bringing Amazon Trainium, Google Tensor Processing Units (TPUs), NVIDIA GPUs and AMD GPUs into a unified resource pool, integrating the compute resources of different partners through software and allocating computing workloads across different chip platforms according to task requirements.

In the past, AI companies expanded compute mainly by increasing GPU purchases, building computing clusters and upgrading interconnect capabilities. Today, leading companies have begun to bring chips of different architectures into the same resource management system, dynamically adjusting compute allocation based on cost, supply conditions, workload type and platform efficiency. In other words, compute strategy is no longer just about "how many cards were bought," but also about "how many kinds of chips can be scheduled" and "how to assign tasks to the most suitable platform." However, the multi-platform layout is mainly intended to expand compute options, and does not mean AI companies will immediately shift to a route of fully breaking away from NVIDIA.

For NVIDIA, short-term order demand has not disappeared, but as customers gradually gain cross-platform scheduling capabilities, GPU purchases are no longer the only choice, and its pricing power, supply negotiation advantages and the customer stickiness brought by its ecosystem will all face new variables.

According to Business Insider citing the view of Patrick Moorhead, founder and CEO of research firm Moor Insights & Strategy, what is most likely to be affected for NVIDIA is its pricing power. Compared with customers that rely on a single chip platform, companies able to flexibly allocate workloads across four platforms clearly have greater bargaining room in procurement negotiations. IDC analyst Ashish Nadkarni also said that a multi-chip strategy can reduce Anthropic's dependence on NVIDIA's compute supply, giving it greater room to adjust when chip supply fluctuates.

For hyperscale AI companies, the stability and continuity of compute supply directly affect the speed of model updates and iteration. Therefore, Anthropic's strategy has a dual meaning: on the one hand, it diversifies supply chain risk; on the other, it strengthens its initiative in negotiations over price and supply terms.

This change reflects that the supply-demand relationship in the AI compute market is shifting. For a long time, NVIDIA GPUs have established a significant market advantage through powerful computing performance, as well as the CUDA software ecosystem, network interconnect solutions and system-level delivery capabilities. If large AI companies could only build computing clusters around GPUs, their room to adjust in the face of supply cycles and price fluctuations would be relatively limited.

The focus of Anthropic's disclosure this time is not just which chips it purchased, but the demonstration of a cross-platform way of organizing and managing compute: integrating computing resources from different suppliers into a unified resource pool, and then, through software scheduling, assigning training or inference tasks to more suitable hardware platforms. From an industry perspective, this means competition in AI infrastructure is gradually shifting from a contest of single-hardware performance to system-level competition combining hardware diversity and software scheduling capabilities.

Cross-platform scheduling still faces technical barriers

Multi-platform compute scheduling is attractive, but actually implementing it still faces relatively high technical requirements. Different chip platforms differ markedly in instruction sets, compilation tools, software stacks and parallel computing strategies, and migrating workloads is not as simple as swapping hardware. Brad Gastwirth, head of global research and market intelligence at Circular Technology, believes Anthropic's strategy will not pose a major threat to NVIDIA in the short term. Even if AI companies have more chip choices, they cannot directly eliminate the advantages NVIDIA has already established in computing performance, network interconnect and software ecosystem.

Competition among AI computing clusters is not simply a contest of the performance of a single chip, but a whole-system competition covering chips, switching networks, interconnect protocols, storage systems and scheduling software. NVIDIA's competitive moat is built not only on the GPU itself, but also on the complete system-level ecosystem formed around the GPU. More importantly, currently only the largest AI companies have the resources and capabilities needed to implement cross-platform compute scheduling. Managing compute across platforms requires professional technical teams and sustained software investment, and also requires companies to have computing tasks of sufficient scale to spread the related costs. Therefore, it is difficult for small and medium-sized AI companies to directly replicate this model.

Overall, Anthropic's multi-platform compute strategy is expected to first affect the procurement decisions and infrastructure planning of leading large AI customers, and then gradually transmit to the broader industrial supply chain. At the same time, NVIDIA is also continuing to accelerate its market layout, and the competitive landscape of the AI compute market will further deepen.

This article is reproduced from media reports or online platforms and represents the author's personal position or views. Our reproduction is solely for sharing and does not mean we approve of or agree with it. If the source attribution is incorrect or your lawful rights and interests are infringed, please contact customer service in a timely manner, and as a neutral platform service provider we will promptly correct, delete or handle it in accordance with the law. Comments No user comments yet Latest News More Related Recommendations

Samsung bets on Exynos 2700 to capture a higher market share 2026-10-30 15:41

Wolfspeed commercializes and launches transparent silicon carbide technology 2026-10-30 15:11

Murata plans to invest US$6.4 billion over the next five years to expand MLCC production, expects fiscal 2030 sales to double those of fiscal 2025 2026-10-30 15:11