Alphabet expands TPU edge in AI compute race
Alphabet's custom Tensor Processing Units (TPUs) now underpin both its own Gemini models and a fast-growing external compute business, CNBC reported June 27, 2026, with Google Cloud revenue projected by FactSet to rise about 64 percent this year to $96 billion. The clearest evidence of that shift is a $5 billion joint venture Google launched with Blackstone on May 18, 2026, which will bring 500 megawatts of TPU capacity online by 2027 as a compute-as-a-service offering separate from standard Google Cloud subscriptions. Anthropic is among the customers renting TPU capacity, and some enterprise customers can now buy TPUs for their own data centers, giving Alphabet a hardware-vendor role that puts it in more direct competition with Nvidia rather than just Nvidia's cloud customers.
The Blackstone joint venture is the clearest signal yet that Alphabet is turning a decade of internal chip investment into an external hardware business, not just a cost advantage inside Google Cloud. That shift, from TPUs as an internal efficiency tool to TPUs as a product Alphabet sells or rents to outside companies including a major AI lab, is the part of this story practitioners should track, more than the revenue projections themselves.
What happened
CNBC reported June 27, 2026 that Alphabet's in-house TPUs power its Gemini chatbot and form a growing part of Google Cloud's AI compute business. Customers including Anthropic rent TPU access on Google Cloud, and some enterprise customers can now buy TPUs for installation in their own data centers. CNBC cited FactSet projections of Google Cloud revenue rising roughly 64 percent this year to $96 billion, with analysts modeling growth above 50 percent in 2027; Google Cloud already reported more than $20 billion in Q1 2026 revenue, up 63 percent year over year. CNBC quoted analyst Brad Gastwirth calling Google "probably the most underappreciated competitor of Nvidia."
Industry context
Blackstone and Google announced their joint venture in official releases on May 18, 2026: Blackstone committed $5 billion in initial equity from its managed funds to bring 500 MW of TPU compute capacity online by 2027, with plans to scale further. Google veteran Benjamin Treynor Sloss, who spent 22 years in Google infrastructure and site-reliability engineering, was named CEO of the new company. Blackstone President and COO Jon Gray called it "a generational opportunity to invest capital at scale building AI infrastructure," and Google Cloud CEO Thomas Kurian said the venture "helps meet growing demand for TPUs, which are optimized specifically for efficiency and performance in the AI era" (Blackstone press release; Google Cloud blog; CIO Dive, May 19, 2026). The venture follows an earlier Google-Broadcom agreement with Anthropic to add multiple gigawatts of TPU capacity starting in 2027, part of Anthropic's stated multi-platform strategy that also includes AWS Trainium and Nvidia GPUs.
Technical context
Google split its eighth-generation TPU line at Google Cloud Next in April 2026 into TPU 8t for training and TPU 8i for inference, following the seventh-generation Ironwood chips that reached general availability in late 2025. Custom accelerators like these optimize the matrix-multiply and memory-access patterns central to large-model training and inference, typically improving performance-per-dollar and energy efficiency compared with general-purpose GPUs when workloads are well matched to the hardware. For practitioners, the Blackstone venture adds a third access path, alongside standard Google Cloud subscriptions and direct on-premises purchase, that changes cost modeling and procurement options for teams evaluating TPUs against Nvidia GPU alternatives.
What to watch
- •Independent, workload-matched benchmarks comparing TPU price-performance against Nvidia GPU alternatives, since current claims are largely vendor- or analyst-sourced.
- •Whether the Blackstone joint venture's 500 MW, 2027 target stays on schedule as the first concrete test of TPUs as an externally sold compute product.
- •Updated Google Cloud revenue figures against the FactSet-modeled 64 percent 2026 growth rate as Q2 2026 results are reported.
- •Broader framework and tooling support for TPUs beyond early adopters like Anthropic, which would signal the ecosystem maturing past a small set of anchor customers.
Key Points
- 1Alphabet's TPUs now power Gemini internally and are rented or sold externally, turning years of custom-silicon investment into a standalone hardware revenue stream.
- 2The $5 billion Google-Blackstone joint venture targets 500 megawatts of TPU capacity by 2027, adding a third access path beyond Cloud subscriptions or on-prem purchase.
- 3Anthropic's TPU rental and Google Cloud's projected 64 percent revenue growth position Alphabet as a more direct Nvidia competitor, reshaping AI infrastructure procurement choices.
Scoring Rationale
Confirmed via primary sources (Blackstone and Google official announcements) plus CNBC and CIO Dive: a $5 billion joint venture, confirmed TPU rental and on-prem sales to enterprise customers including Anthropic, and strong Google Cloud revenue growth projections. This is a notable infrastructure and competitive-landscape development affecting procurement decisions and the Nvidia rivalry, but it is an extension of an already-announced venture rather than a new regulatory or frontier-model milestone, keeping it in the Notable tier.
Sources
Primary source and supporting public references used for this report.
Practice with real Ad Tech data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Ad Tech problems
