JPMorgan: Google TPU v9 Not Delayed, Broadcom AI Revenue Visibility Extended to 2031
Alina Collins
JPMorgan's latest report directly rebuts market rumors, confirming Google's next-gen AI chip TPU v9 remains on track with Broadcom for 2028 mass production; a new five-year deal locks in four generations of TPUs, extending Broadcom's AI revenue visibility to 2031 — a rare degree of long-term certainty in today's AI supply chain.
Has TPU v9 actually been delayed?
JPMorgan is unequivocal: no delay, no cancellation. Broadcom began TPU v9 IP design in early 2025, moved into SoC — system-on-chip, combining multiple functions on one die — development in the second half, and the project has stayed at the company's top priority throughout.
The chip uses a 2nm process and integrates 4 compute dies, 16 HBM stacks, and 400 Gbps SerDes interfaces. This means → it is a next-generation AI training "super-chip," far exceeding current specs.
What's more, early R&D on TPU v10 has already begun. This signals that Google and Broadcom's collaboration is not a one-off engagement but a continuously rolling pipeline.
What exactly does the five-year deal lock in?
In March this year, Google and Broadcom signed a new five-year partnership covering the roadmap for four generations of TPUs — v8 through v11 — with revenue commitments from Google's side.
This means → Broadcom's TPU-related revenue visibility stretches all the way to 2031. In plain terms = most chip companies can only see one or two years of orders ahead; Broadcom has secured what amounts to a six-year contract — virtually unmatched in the AI supply chain.
This reflects a structural dependency, not a short-term convenience — switching suppliers would be too costly and too risky for Google.
Could Google's in-house chip team take Broadcom's place?
Over the past year, Google has expanded its internal chip organization (COT) and brought in additional suppliers — many investors see this as Broadcom's biggest risk. JPMorgan says bluntly: this concern is significantly overstated.
Google's in-house team is still optimizing the existing TPU v8t (codenamed Zebrafish), while Broadcom's TPU v8i (codenamed Sunfish) completed qualification in mid-2025 and is about to enter mass production — putting the gap at 18 months or more in advanced AI chip design.
In plain terms = ASIC — custom chip — competition is not a matter of hiring more engineers and catching up. It spans compute architecture, network design, memory systems, advanced packaging, and accumulated IP. Over 12 years, Broadcom has helped Google deliver 14 of its most advanced chip designs. That track record cannot be replicated quickly.
What other AI contracts does Broadcom hold?
Broadcom currently has the world's largest AI ASIC project pipeline, with clients including Google, Anthropic, OpenAI, and Meta.
It is also engaged in SoftBank/Arm's XPU project, SambaNova's RDU project, and Apple's AI CPU project. This means → Broadcom's AI business is not a single-client bet but a multi-node position across the entire AI ecosystem.
As Broadcom's Singapore advanced substrate factory comes online and advanced packaging capacity ramps in 2028, JPMorgan projects AI revenue will grow 2× to 2.5× year-over-year in 2027 and double again in 2028.
What does this mean for the stock?
JPMorgan reiterates its Overweight rating with a December 2026 price target of $580.
Broadcom shares rose more than 2% pre-market, following a 4.37% decline in the prior session. This reflects a correction of the panic driven by the TPU v9 delay rumors.
In plain terms = JPMorgan's core thesis is that the market has fixated on the risk of Google going in-house while overlooking the revenue certainty the five-year deal provides — and it is that certainty, not the risk, that anchors Broadcom's valuation.
Content is for reference only, not financial advice.