Claude is now generally available in Microsoft Foundry on Azure
If you run anything on Azure, this is the change worth acting on today. Claude has moved from preview to general availability inside Microsoft Foundry, which means you can now call Claude Opus 4.8 and Claude Haiku 4.5 through the Messages API using the same Azure identity, billing and governance your team already relies on. The practical win is that you no longer need a separate Anthropic account or a parallel procurement path to put Claude into production.
Data residency is handled: you can keep inference inside a US data zone where that matters. Prompt caching and extended thinking are both supported, so coding agents and longer reasoning workloads behave the way they do on Anthropic's own API. Billing arrives as a single Claude Consumption Units line on your Azure invoice, making it straightforward to fold Claude spend into an existing Microsoft commitment. For most enterprise teams, this removes the last administrative reason to keep Claude at arm's length. Both Anthropic and Microsoft announced this through their own channels before mainstream press picked it up.
Claude on GB300 Blackwell Ultra: hardware choice now affects how your agents feel in production
The same Azure GA launch matters for a second reason worth separating out: the hardware underneath it. Claude in Microsoft Foundry runs on Nvidia GB300 Blackwell Ultra GPUs, and the choice does real work rather than sitting in a spec sheet. The GB300 roughly doubles the high-bandwidth memory of the GB200 and links up to 72 GPUs in a single NVLink fabric, which is what allows the largest Claude Opus model to run without the throughput penalty you would normally expect at that scale.
Microsoft's early figures point to materially faster token generation for Claude Sonnet on GB300 compared with older H100 nodes. For anyone who is latency-sensitive, the platform you pick now changes how your agents feel in production. This is the first time hardware generation has been promoted as a user-facing feature in a Claude deployment rather than an infrastructure footnote, which signals that the gap between hardware tiers has become large enough to matter to everyday users, not just ops teams.
Today's themes
Two themes stand out from an otherwise quiet day. First, the centre of gravity for frontier models is shifting towards the major cloud platforms rather than the model makers' own front doors: Claude reaching general availability inside Microsoft Foundry means the buying decision increasingly happens where your governance and billing already live. Second, hardware is becoming a visible part of the product story rather than a back-office detail, with GB300 Blackwell Ultra hosting promoted as a user-facing performance feature. Both points surfaced first through Anthropic's, Microsoft's and Nvidia's own channels before mainstream press picked them up. For anyone building on Claude, the practical takeaway is that your platform choice now affects access, cost structure and latency in a single decision.