Broadcom CEO: OpenAI Chip Mass Production by Year-End, Betting on Google-Anthropic Iron Triangle
Claire Weston
Broadcom's Q2 revenue topped $22 billion for the first time with AI semiconductor sales up nearly 80%, but a Q3 AI guide of $16 billion — below some analysts' $17 billion-plus forecasts — sent shares down over 10% after hours; CEO Hock Tan responded with specific timelines and client data, dismissing the selloff as a 'surreal AI environment.'
Record quarter — so why did the stock tank?
Broadcom posted Q2 net revenue above $22 billion, up roughly 48% year-over-year; AI semiconductor revenue rose nearly 80%.
But Q3 AI semiconductor guidance of $16 billion fell short of some analysts' $17 billion-plus expectations, triggering an after-hours drop of more than 10%.
This means → the market isn't rejecting Broadcom's fundamentals — it's punishing "not fast enough," a classic expectations-gap reaction.
CEO Hock Tan was blunt: "Focus on fundamentals, focus on creating enormous value, and stop thinking about the stock price. Watching it only brings grief."
Where exactly does the OpenAI custom chip stand?
Tan disclosed for the first time that Broadcom has six custom AI accelerator clients at various stages, with three in deep collaboration for over two years.
Google is furthest along; OpenAI is also a partner. The OpenAI chip "runs very well in their labs and data centers" and will go into production late this year.
On rumors that Broadcom needs a Microsoft agreement to sell a share of chips, Tan was categorical: "Not at all."
In plain terms = OpenAI's custom chip is past the lab stage and headed for real workloads by year-end — and the deal has no strings attached to Microsoft.
Could Google build its own chips and drop Broadcom?
Tan offered a counterintuitive frame: Google's real competitor is Nvidia's GPUs, not Broadcom.
"As long as Nvidia keeps rolling out generation after generation of superb technology, Google has to create equally good technology to compete — and that's where we come in."
This means → Broadcom's moat isn't "Google can't leave me." It's "Google can't compete with Nvidia without me" — the fiercer the GPU race, the more indispensable Broadcom becomes.
He acknowledged Google is also working with Marvell and others on in-house toolchains, but called those efforts "supplementary and guided" — the core contest remains a head-on technology fight.
What is the Broadcom-Google-Anthropic "iron triangle"?
About a year ago, Broadcom and Google jointly designed TPUs — Google's custom AI chips — to supply compute for Anthropic.
Tan called the partnership a "leap of faith" — a bet not just on Anthropic as a company, but on generative AI scaling across enterprise customers.
In plain terms = Broadcom is wagering that enterprise-scale AI adoption is inevitable, and Anthropic is on that exact path — so it placed the bet early.
One engineer, one week, doing what ten people took three months to do — really?
Tan revealed that Broadcom engineers are heavily using Anthropic's Opus 4.7 and other large-model tools for chip design and code assistance.
His ROI data point: "One skilled senior engineer can now finish an application design in a week; previously it took ten engineers earning $300,000 a year three months to do the same."
Broadcom has placed no cap on tool usage — ROI is already compelling, and engineers are still early on the learning curve, meaning gains will compound.
This reflects that Broadcom isn't just an AI-chip seller — it's a heavy AI-tool buyer, validating the "AI boosts productivity" narrative with its own spending.
Revenue doubling in two years — so why stop doing deals?
Broadcom built its empire on aggressive M&A, but Tan said the company has entered a phase where "acquisitions aren't needed to drive growth."
His math is simple: "Revenue will double between 2024 and 2026, exceeding $50 billion a year. I look around — what can I buy that matches that scale? Nothing."
He positioned Broadcom as the "picks and shovels" supplier of the AI compute wave, calling generative AI's demand for compute "insatiable" — organic growth alone is stunning enough.
This means → M&A was once Broadcom's core growth engine; organic growth has replaced it — which is itself a mirror of how explosive AI demand has become.
Content is for reference only, not financial advice.