On October 27, 2026, Alphabet and Oracle announced a TPU partnership. This FAQ breaks down the Wells Fargo Q3 estimate, TD Cowen's 2027 forecasts, the roughly 20 percent external-sales margin, and the Nvidia impact.
What exactly changed in Google's cloud strategy?
According to a Cailian Press report on October 28, 2026, Alphabet announced on Tuesday, October 27 local time that it will offer TPUs on Oracle Cloud Infrastructure (OCI), making Oracle the first hyperscaler besides Google to provide TPU capacity. The related compute is expected to start deploying in 2027. Previously TPUs were mainly used internally or delivered through Google Cloud itself, so the deal marks the first time Google sells the chips through partner cloud channels.
How did Wells Fargo arrive at its estimate of about 9 billion dollars in Q3 TPU sales?
Citing institutional forecasts, Cailian Press reported that Wells Fargo estimates Google's Q3 2026 TPU sales at about 9 billion dollars, rising to 99 billion dollars in 2027. Alphabet CFO Anat Ashkenazi said on the July Q2 earnings call that only a small portion of revenue under existing TPU sales agreements is recognized in 2026, with the vast majority expected to be recognized in 2027, which is why banks are placing the revenue ramp in 2027.
What lies behind the 2027 TPU revenue range of 62.5 to 99 billion dollars, and why do estimates diverge so much?
Cailian Press reported that TD Cowen projects Google's H2 2026 TPU sales at 9.4 billion dollars, growing to 62.5 billion dollars in 2027, while Wells Fargo's 2027 forecast stands at 99 billion dollars, a gap of about 36.5 billion between the two. The divergence comes from assumptions about customer coverage (the Oracle channel only, or more cloud partners) and the recognition pace of long-term deals with large buyers such as Anthropic.
What does the roughly 20 percent external-sales margin mean next to Google Cloud's 35.6 percent operating margin?
TD Cowen estimates the operating margin of the TPU external-sales business at about 20 percent, below Google Cloud's actual Q2 2026 operating margin of 35.6 percent, according to Cailian Press. Selling chips through partners means earning as a chip supplier rather than as a full-stack cloud provider. Citing financial reports, Cailian Press noted Google Cloud posted 25 billion dollars of revenue in Q2 2026, below AWS at 42 billion and Azure at 29 billion, though it grew the fastest of the three.
How will TPU external sales hit Nvidia, and is the cost advantage real?
Citing industry research firm SemiAnalysis, Cailian Press reported that Google's latest Ironwood TPU has about a 44 percent lower total cost of ownership than Nvidia's GB200 servers. Google Cloud CEO Thomas Kurian said at a Goldman Sachs conference in September that the payback period for Google's custom-chip servers is about half that of ordinary AI servers, which take less than two years. If the TCO advantage holds, selling TPUs via OCI and other channels lets customers access compute without locking into Nvidia's CUDA ecosystem.
What do the Anthropic million-TPU deal and OpenAI's purchasing plans mean for this business?
Cailian Press reported that Anthropic signed a tens-of-billions-of-dollars agreement with Google in October 2025 to use up to 1 million Google TPUs; Mizuho Securities said the deal will bring nearly 1 gigawatt of new compute capacity in 2026. Such long-term agreements underpin both banks' projections of a 2027 revenue ramp: even at roughly 20 percent margin, million-chip volumes translate into tens of billions of dollars in revenue, a key lever for Google to close the scale gap with AWS and Azure.