AVGO jumps on Anthropic demand

Broadcom and Amazon are moving because Anthropic’s compute appetite stopped looking theoretical and started looking booked. That is the core of this trade. In April, Anthropic signed one expansion with Amazon for up to 5 gigawatts of capacity and another with Google and Broadcom tied to about 3.5 gigawatts of compute. ### Why did AVGO jump? Because Broadcom sits in the part of the AI stack investors increasingly care about now — not just merchant GPUs, but custom accelerators, packaging, interconnect, and the plumbing that lets hyperscalers scale giant clusters. Broadcom has been pushing that exact story, from custom XPU design to advanced 3.5D packaging and cluster networking for gigawatt-scale AI systems. ### What does Anthropic have to do with that? (anthropic.com) Anthropic basically validated the demand side. Its April announcement said it trains and runs Claude across AWS Trainium, Google TPUs, and Nvidia GPUs, while calling Amazon its primary cloud provider and training partner. That matters because Broadcom is deeply tied to Google’s TPU roadmap, and Anthropic’s new Google/Broadcom arrangement makes that linkage much more concrete for investors. (broadcom.com) ### Why is Amazon in the trade too? Because Amazon is not just an investor in Anthropic — it is the main cloud and training partner, and Anthropic committed to spending more than $100 billion on AWS technologies over 10 years as part of the expanded deal. Amazon also agreed to invest up to $25 billion more in Anthropic on top of the $8 billion already committed in prior rounds. That turns “cloud credits” chatter into something much bigger — a long-duration infrastructure customer. (anthropic.com) ### What is the number everyone is keying on? Gigawatts. Not GPU counts. Anthropic said the Amazon agreement secures up to 5 GW, with nearly 1 GW of Trainium2 and Trainium3 capacity online by the end of 2026. The Google/Broadcom side adds roughly 3.5 GW more. At that scale, investors stop thinking about one model launch and start thinking about power, packaging yield, networking, and who captures each layer of spend. (anthropic.com) ### Why does this help Broadcom more than a random chip name? Because Broadcom is one of the few companies selling the picks-and-shovels for custom AI silicon at hyperscaler scale. Its pitch is that next-wave AI clusters will mix custom XPUs, advanced packaging, Ethernet fabrics, optics, DSPs, and PCIe connectivity. If Anthropic is spreading workloads across Trainium, TPUs, and GPUs, the winners are not just chip vendors — they are the companies enabling heterogeneous systems to work at absurd scale. (anthropic.com) ### Is the stock move just vibes? Not entirely. Broadcom already jumped after the April 6 disclosure of expanded Google and Anthropic deals, and analysts immediately framed it as evidence Broadcom could outrun its long-term revenue targets. As of May 8, AVGO closed around $430, near fresh highs, while Amazon closed around $272, also close to its 52-week high. The market is clearly capitalizing these announcements into future demand. (broadcom.com) ### What changed versus the old AI trade? The old trade was mostly “buy Nvidia.” The newer one is broader — buy the companies that finance, package, connect, and manufacture model-scale compute. Anthropic’s revenue run rate topping $30 billion gave that shift extra credibility, because it suggests this infrastructure is being pulled by real commercial demand, not just speculative capex. (cnbc.com) ### So what is the bottom line? Anthropic’s deals turned AI demand into named counterparties, hard capacity numbers, and multi-year spending commitments. That is why Broadcom jumped — and why Amazon keeps getting pulled into the same conversation. The market is starting to price AI as an infrastructure system, not a single-chip story. (anthropic.com) (bloomberg.com)