OpenAI and Anthropic have placed orders totaling 26 million Google TPUs through 2029, according to top supply chain analyst Ming-Chi Kuo, extending Alphabet's custom chip supply visibility and positioning its TPU business as a credible alternative to Nvidia GPUs amid leadership uncertainty at the latter.
TL;DR
- OpenAI and Anthropic ordered 12M and 14M Google TPUs respectively for 2028–2029, giving Alphabet a multi-year supply runway.
- The orders shift chip procurement power toward Google's TPUs, challenging Nvidia's dominance amid CEO succession uncertainty.
- Analysts estimate 2027–2029 TPU shipments at 8–18M units, with older v8 chips remaining in use alongside newer v9.
What happened
Top supply chain analyst Ming-Chi Kuo reported that OpenAI will seek approximately 12 million TPUs across 2028 and 2029, while Anthropic plans to order about 14 million units during the same period.
OpenAI intends to purchase its TPUs outright, whereas Anthropic will combine leasing with direct purchases, according to Kuo.
The analyst noted that these orders extend Google's TPU shipment visibility to 2029, ahead of Nvidia's GPU and other AI ASIC shipments, which are mostly expected between the fourth quarter of 2027 and the second quarter of 2028.
Kuo also highlighted that the orders will keep Google's older v8 TPU chips in heavy use even as the company ramps up its own expansion toward the newer v9 architecture.
Why it matters
For AI startups and developers, the orders signal a diversification of chip supply chains beyond Nvidia, reducing single-vendor dependency and potentially stabilizing costs and availability.
The long-term commitments from two of the largest AI labs suggest that custom silicon from hyperscalers like Google is becoming a viable alternative, especially as Nvidia faces leadership uncertainty following Jensen Huang's resignation and an ongoing investigation into its ties with Megaspeed.
Investors in AI infrastructure should note that Google's TPU business now has a clearer path to revenue through 2029, which could support sustained capital expenditure in AI hardware and cloud services.
However, the shift also introduces new risks: reliance on a single supplier's custom architecture may create lock-in effects similar to those seen with Nvidia, and the timeline for newer TPU generations (e.g., v9) remains unproven in large-scale deployments.
Key facts
- OpenAI's order covers approximately 12 million TPUs across 2028 and 2029, per analyst Ming-Chi Kuo.
- Anthropic's order totals about 14 million TPUs for the same period, combining leasing and direct purchases.
- Analyst estimates place 2027 TPU shipments at 8–9 million units, rising to 14–15 million in 2028 and 16–18 million in 2029.
- Google's TPU shipment visibility now extends to 2029, ahead of Nvidia's expected GPU and AI ASIC shipments (mostly 4Q27–2Q28).
- OpenAI plans to buy its TPUs outright, while Anthropic will use a mix of leasing and direct purchases.
- Jensen Huang resigned as Nvidia's CEO on February 23, 2026, with CFO Colette Kress named interim chief.
- Nvidia's stock fell 12.8% in the week following the leadership change, its worst weekly performance since March 2025.
- Alphabet holds an estimated 9.9% stake in Anthropic, valued between $126B and $139B as of the latest reports.
Context
The orders underscore a broader trend of AI labs diversifying their chip supply chains to mitigate risks tied to Nvidia's near-monopoly in AI accelerators, which has drawn regulatory and investor scrutiny.
Hyperscalers like Google, Amazon, and Microsoft are increasingly designing and deploying custom AI chips (e.g., Google's TPUs, AWS's Trainium, Microsoft's Maia) to reduce costs and improve performance for their own AI workloads and to offer alternatives to third-party developers.
Nvidia's leadership uncertainty has amplified concerns about supply chain stability, particularly as AI spending is expected to peak in 2028, according to market analysts.
