TPU Shift: Google Cloud Revenue Growth Could Hit 150% by 2027

Woofun AI reports that a significant reallocation of TPU hash rate resources, triggered by leadership restructuring at Google DeepMind, positions Google Cloud as the primary short-term financial beneficiary. This strategic pivot, analyzed by SemiAnalysis, suggests that shifting computational power from internal model training to external commercial sales could drive Google Cloud's revenue growth to approximately 150% by 2027, substantially outpacing the 64% consensus estimate. The core dynamic involves a trade-off between maintaining competitiveness in cutting-edge AI models and accelerating revenue recognition through infrastructure sales.

The leadership restructuring at the heart of this shift involves high-profile personnel changes within Alphabet's AI division. Demis Hassabis, previously the CEO of Google DeepMind, has stepped down from the operational role but retains influence as the chairman of DeepMind and the Chief Scientist of Alphabet. Day-to-day management of DeepMind has been transferred to Koray Kavukcuoglu, the former CTO of DeepMind, who now reports directly to Thomas Kurian, the CEO of Alphabet. This reporting line places the head of Google Cloud in a position of increased influence over the internal competition for scarce TPU resources, potentially prioritizing commercial cloud revenue over pure research autonomy.

Simultaneously, a new AI research entity, Discovery Loop, has been formed by former Google executives Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le. These key figures left Google to establish this independent AI research organization, with Google acting as an initial investor. The relationship is structured around cloud service partnerships, meaning Discovery Loop will likely utilize Google Cloud's infrastructure. This arrangement creates a scenario where former internal researchers become external customers, further incentivizing the allocation of TPU resources toward commercial cloud services rather than internal proprietary model development.

The strategic dilemma centers on the dual use of TPU resources for either supporting the training and inference of Google's own Gemini models or supplying external AI labs and enterprise clients. TPU units can also be sold as complete systems to special purpose entities operating AI data centers. Allocating hash rate to Gemini enhances Google's competitiveness in advanced AI models, whereas directing these resources to commercial clients contributes more quickly to Google Cloud's revenue and backlog orders. SemiAnalysis argues that Thomas Kurian's growing influence is tilting the balance toward external sales, allowing Google Cloud to capitalize on the massive demand for AI infrastructure from third parties.

Despite these market inferences, Alphabet has not officially acknowledged any decline in priority for its Gemini models. During its second-quarter earnings call, the company stated that Gemini 4 is undergoing the most ambitious pre-training to date and that AI services remain constrained by supply issues. Sundar Pichai emphasized that the top priority for TPU allocation remains ensuring Google has the necessary hash rate to compete in the field of advanced AGI. Consequently, the claim that 'Gemini's priority has declined' is an inference made by SemiAnalysis based on personnel changes and resource allocations, rather than a confirmed strategic shift announced by Google.

Financial performance in the second quarter reflects the rapid expansion of Google Cloud, with revenue growing by 82% year-on-year to $24.768 billion. Backlog orders reached a substantial $514 billion, indicating strong future revenue potential.

Additionally, the processing volume of Google's own model APIs increased to about 22 billion tokens per minute, up from 16 billion in the previous quarter. These metrics demonstrate robust demand for both cloud infrastructure and AI services, although the composition of this growth is evolving significantly with the introduction of new revenue streams.

A critical component of this growth is the sale of complete TPU systems to customer data centers, a revenue stream recognized for the first time in the second quarter. SemiAnalysis estimates that revenue recognized from TPU system sales using a gross method is around $35 billion per GW, with related revenue projected to be around $1.2 billion in the second quarter of 2026. Even if this portion of revenue is excluded, Google Cloud's core business still shows a year-on-year growth rate of just over 70%; when TPU system sales are included, the overall growth rate rises to 82%. This indicates that TPU system sales are becoming a major new source of growth, distinct from traditional cloud services.

Looking ahead, SemiAnalysis projects that Google Cloud currently has backlog orders related to TPU systems exceeding $150 billion. If these orders are delivered as planned and recognized by 2027, Google Cloud's revenue growth rate for that year could reach around 150%, compared to the 64% consensus estimate. Alphabet's officially disclosed $514 billion in cloud business backlog orders includes sales of TPU systems, but it does not provide a separate figure for TPU-related orders. Therefore, the figure of over $150 billion comes from SemiAnalysis's modeling. Based on the reported revenue recognition rate of around $35 billion per GW. A single GW-scale project is sufficient to significantly alter Google Cloud's growth trajectory.

Profitability metrics also suggest a positive impact from this strategy. SemiAnalysis estimates that the EBIT margin from TPU system sales is slightly lower than Google Cloud's core cloud business, around 30%, but Google Cloud's overall EBIT margin could still remain between 35% and 39%. Driven by high revenue growth and decent profit margins, these businesses could contribute around $3 to Alphabet's EPS by 2027. This financial chain illustrates how AI department restructuring affects hash rate allocation, leading to increased system sales, higher backlog orders, and ultimately, greater revenue and profits for Alphabet.

Risks and valuation concerns remain tied to the sustainability of this growth model. The growth rate of tokens from Gemini's first-party API has slowed down; in the first quarter of 2026, its processing volume increased from 10 billion tokens per minute to 16 billion, a 60% quarter-on-quarter increase, but it further rose to 22 billion tokens in the second quarter with a growth rate drop to around 38%. While Google still claims strong demand, the reliance on third-party models like Claude from Anthropic may increase if Google falls behind in cutting-edge models. Investors must monitor whether TPU system orders, which have a hardware-like nature, can be delivered on time and whether Google Cloud's profit margin can remain within the 35% to 39% range predicted by SemiAnalysis. If deliveries are smooth, Google Cloud's revenue growth rate in 2027 could approach 150%, providing new momentum for Alphabet's AI business.