TPU-aaS could lift Alphabet's 2028 profit by 15%—Barclays

Barclays released a research note arguing that Alphabet is poised to launch a TPU-as-a-service (TPU-aaS) model as one of its most significant new AI infrastructure initiatives. Under this model, Google would sell its Tensor Processing Units directly rather than only through Google Cloud, initially supporting Anthropic and potentially expanding to other AI developers. The brokerage contends that the market is underestimating Alphabet's long-term earnings potential from this shift.

According to Barclays' analysis, the TPU-aaS initiative could lift Alphabet's 2028 gross profit by about 15% while significantly expanding operating income. The firm estimates Google could generate roughly $250 billion（約40兆円） in TPU-aaS revenue by 2028 if deployments scale as expected, with external TPU sales reaching more than 11 gigawatts of capacity. This revenue level would represent more than one-third upside to current 2028 revenue consensus, suggesting significant upside to street expectations. The strategy is expected to broaden AI compute capacity across the industry while improving Alphabet's capital efficiency by shifting toward a more asset-light infrastructure model.

Beyond Google's financials, Barclays identifies ecosystem-wide benefits. AI labs such as Anthropic would gain greater control over their computing infrastructure and reduce dependence on hyperscaler cloud agreements. Partners including Broadcom, Blackstone, and infrastructure provider Fluidstack stand to benefit from expanded demand for AI hardware, financing, and data center management as the TPU-aaS model scales. Barclays maintained its Overweight rating and $425 price target on Alphabet, arguing the TPU-aaS opportunity is not fully reflected in investor expectations. The firm significantly raised its Google Cloud revenue and operating income forecasts after incorporating the new business model into its financial estimates.

Barclays' research note on Alphabet's TPU-aaS initiative highlights a structural shift in how the company monetizes its AI infrastructure. Rather than selling compute capacity exclusively through Google Cloud's subscription model, Alphabet would sell its Tensor Processing Units directly to AI developers. This shift addresses a key tension in the AI ecosystem: AI labs like Anthropic need specialized compute power but also want independence from the major cloud providers. By offering TPU-aaS, Alphabet captures revenue while reducing customer lock-in concerns that might otherwise push developers toward competitors.

The financial forecasts are striking. Barclays' $250 billion（約40兆円） TPU-aaS revenue estimate by 2028, paired with a projected 15% boost to 2028 gross profit, suggests the brokerage sees this as a material contributor to Alphabet's growth—particularly as it positions Google Cloud for long-term expansion. The firm's reasoning centers on capital efficiency: an asset-light model (where Alphabet sells the hardware rather than owning and operating all the infrastructure itself) can generate higher margins with lower capital intensity than traditional cloud models. The spillover benefits—expanded demand for Broadcom's chips, Blackstone's investment in data centers, and Fluidstack's infrastructure management—indicate Barclays views this as an industry-wide opportunity, not just a Google story.

No comments yet. Be the first to share your thoughts!