Anthropic Unleashes Claude AI on Microsoft’s Azure Foundry Platform
By 813 Staff
NVIDIA just made Anthropic a direct competitor to its own largest cloud partners. An internal announcement from the company’s enterprise division, confirmed by engineers close to the project, reveals that Claude models are now generally available in Foundry, running on NVIDIA’s proprietary DGX infrastructure. The tweet from @nvidia on June 29, 2026, marked the end of a quiet beta period that began in early Q2. The rollout, however, has been anything but smooth.
For the uninitiated, Foundry is NVIDIA’s private cloud service for enterprises that want to run large language models without handing data over to hyperscalers like AWS or Azure. What makes this move seismic is that Claude—Anthropic’s flagship model family—was previously only accessible via Anthropic’s own API or through those same hyperscaler marketplaces. Internal documents show that NVIDIA and Anthropic spent roughly six months optimizing Claude’s inference kernels for H100 and the upcoming B200 “Blackwell” GPUs, achieving what sources describe as a 30% reduction in latency compared to standard deployments. The integration is deep: Foundry customers can now deploy Claude Opus, Sonnet, and Haiku with NVIDIA’s NeMo guardrails baked in, meaning enterprises can enforce content policies and data isolation at the hardware level.
Why it matters beyond the headline: NVIDIA is effectively commoditizing its own largest customers. Every major cloud provider—Amazon, Microsoft, Google—buys NVIDIA GPUs by the tens of thousands to resell AI compute. By offering Claude directly through Foundry, NVIDIA undercuts their margins and captures the full stack value. For enterprise buyers, it means a simpler procurement path: one contract for both the GPU cycle and the model license, with data never leaving NVIDIA’s network. Anthropic benefits by getting Claude onto the most sought-after hardware without competing for cloud capacity.
What happens next remains uncertain. Hyper-scale partners are already privately pushing back, according to industry chatter, threatening to reduce GPU orders in favor of AMD’s MI400 series or custom chips. NVIDIA has not commented on whether additional model providers—including OpenAI or Mistral—will join Foundry. Engineers close to the project say Anthropic secured exclusivity for at least twelve months on frontier models, but that timeline has not been confirmed publicly. For now, the message is clear: NVIDIA is no longer just the pick-and-shovel seller in the AI gold rush. It’s a platform operator, and it’s taking names.
