Anthropic’s reported multichip strategy routes workloads across suppliers, but the October account of its S-1 describes options rather than a quantified shift away from Nvidia. The filing, published October 26, 2026, names Amazon, Google, AMD, and Nvidia as strategic partners whose chips software can select according to productivity. NeoTeo’s earlier coverage of Anthropic’s possible IPO valuation provides background on the listing story.
On this page
- How Anthropic routes work across chip suppliers
- Why training is harder to move between chips
- What Anthropic’s Samsung chip exploration involved
- What Anthropic’s reported billions measure
How Anthropic routes work across chip suppliers
Business Insider España reported on October 29 that Anthropic described its compute, the processing capacity used to train and run AI models, as a shared pool from multiple strategic partners. Software can direct workloads to different chips according to where they are more productive. The account says developing that flexibility took years and significant investment.
Nvidia remains one of the suppliers named in the strategy, alongside Amazon, Google, and AMD. The reported arrangement offers Anthropic more sourcing options; it does not amount to a measured change in Nvidia’s share of Anthropic’s workload.
Why training is harder to move between chips
Switching chip platforms is reported to be particularly difficult for training. IDC analyst Ashish Nadkarni said using multiple suppliers could make Anthropic less vulnerable to Nvidia supply fluctuations. Brad Gastwirth of Circular Technology said those options do not erase Nvidia’s advantages in performance, networking, and software.
That distinction matters because supplier variety and a shift in actual workloads are different things. The strategy describes how Anthropic can route work; the figures do not assign a share of that work to any one supplier.
What Anthropic’s Samsung chip exploration involved
A report published July 2, 2026, described early-stage work on a custom Anthropic chip and discussions with Samsung Electronics about its 2-nanometer process and advanced packaging. The report said detailed design and manufacturing had not begun at that time. It also attributed to Anthropic that AWS Trainium, Google TPUs, and Nvidia GPUs remained central to scaling compute.
What Anthropic’s reported billions measure
Forbes’ October 30 video report described $587 billion in chip and data-center commitments as of June 2026, followed by $266 billion in new or amended agreements over the next four months. It also cited $20 billion in expected AMD capacity and an estimated $125 billion to $150 billion for Google-designed chips, and characterized possible obligations through 2045 as around $1 trillion. The figures describe a mix of commitments, expected capacity, and estimates.
Forbes separately reported $19.2 billion in infrastructure spending and $21.3 billion in revenue from January 2023 through June 2026. Of that revenue, $16.3 billion was recorded in the first half of 2026.
An earlier partnership account, dated November 18, 2025, described a $30 billion Anthropic commitment to Azure compute and additional capacity of up to 1 gigawatt. The initial capacity was associated with Nvidia Grace Blackwell and Vera Rubin systems.
The Forbes report also discusses supplier relationships and financial risks around Anthropic’s infrastructure commitments.
