Anthropic confirmed at Google Cloud Next 2026 that it will deploy multiple gigawatts of next-generation TPU capacity starting in 2027, built through the Alphabet-Broadcom TPU joint venture. The commitment extends the April 22 Ironwood 1M-TPU deal into the 8th-generation silicon roadmap.
What was committed
- Scale: multiple gigawatts (exact capacity undisclosed).
- Timeline: deployment begins 2027, ramping through 2028-2029.
- Silicon substrate: Alphabet-Broadcom co-designed next-gen TPUs, including TPU 8t/8i unveiled Day 2.
- Carryover: stacks on top of the earlier 1M Ironwood (TPU v7) commitment.
Combined with the $25B Amazon-Anthropic pact (up to 5 GW on Trainium2/3) and the April 7 3.5 GW Google TPU capacity deal, Anthropic now has well over 10 GW of multi-cloud AI compute committed for Claude through 2030.
Who benefits financially
| Party | Upside |
|---|---|
| Alphabet (Google Cloud) | Multi-gigawatt external TPU demand, Q4 2025 Cloud revenue +48% YoY before this deal |
| Broadcom | Custom-AI-silicon division at $8.4B quarterly revenue (+106% YoY); company guides to $100B+ annual AI-silicon revenue by 2027 |
| Anthropic | Long-cycle compute floor priced competitively vs cash purchases of Nvidia Blackwell |
Strategic read
Anthropic has publicly anchored itself on three compute substrates:
- AWS Trainium2/3 (via the Amazon pact): up to 5 GW.
- Google Ironwood + 8th-gen TPUs (via this commitment): multi-GW from 2027.
- Nvidia GPUs (legacy and hybrid workloads): undisclosed, declining share.
The multi-cloud hedging is structural. If any one silicon program slips, the others pick up the load. The downside for Anthropic is complexity: three distinct inference stacks, three sets of tooling, three partner relationships to manage.
For Claude users, the near-term impact is capacity. Rate limits and queue times on Opus 4.7 and Sonnet 4.6 should continue to ease as 2027-2028 capacity lands.
For Nvidia, the competitive impact is incremental. Anthropic never was a majority-Nvidia customer at the scale of OpenAI’s Azure deployment. The message here is more about what happens if OpenAI ever diversifies similarly.
Open questions
- Exact gigawatt figure was not disclosed.
- Pricing terms vs the 2023-2026 Anthropic-Google TPU deal remain confidential.
- Which 8th-gen generation (TPU 8i, TPU 8t, or both) anchors the commitment.
- Interaction with the Anthropic IPO narrative (recent reports of $800B valuation offers) if and when Anthropic goes public.
Related
- Google unveils Ironwood TPU at Cloud Next 2026
- TPU 8t/8i unveiled on Cloud Next Day 2
- Amazon commits up to $25B more to Anthropic
- Anthropic locks in 3.5GW Google TPU capacity
Sources
Primary and corroborating references used for this news item.
Spotted an error or want to share your experience with Anthropic commits multi-gigawatt capacity on next-gen Alphabet-Broadcom TPUs starting 2027?
Every tool page is re-verified on a recurring cycle, and corrections land faster when readers flag them directly. If you spot a stale fact, a missing capability, or have used Anthropic commits multi-gigawatt capacity on next-gen Alphabet-Broadcom TPUs starting 2027 and want to share what worked or didn't, the editorial desk reviews every message sent through this form.
Email editorial@aipedia.wiki