Barclays: Alphabet's TPU-aaS could become an underappreciated new growth curve

Overview

This Barclays report focuses on Alphabet Inc.'s new TPU-as-a-service (TPU-aaS) business model. The core view is that, through partnerships with Anthropic, Broadcom, Blackstone/Apollo, Fluidstack, and others, Google is expanding TPU compute from the traditional internal GCP leasing model into sales to external entities and SPV structures, thereby opening up an AI infrastructure revenue pool that may be underestimated by consumer internet investors.

Core views

The report argues that TPU-aaS could become one of Alphabet's most important new business models over the coming years. Google can sell TPUs directly to SPVs and earn hardware sales revenue at roughly a 25% gross margin, while AI labs gain deeper customization and vertical integration through more direct control of compute assets. Barclays therefore materially raises its Google Cloud revenue and operating profit forecasts, but due to uncertainties around capital expenditure and valuation, it maintains the $425 target price rather than shifting to a more aggressive 2028 valuation framework.

Analysis framework

The report uses a bottom-up approach based on capacity, unit economics, and segment-level financial reconstruction: it first estimates the number of TPUs supported per GW, TPU pricing, unit cost, and gross profit, then combines the potential deployment pace of the Broadcom/Blackstone/Apollo AI XPV Platform and the Blackstone-Google TPU Cloud JV to derive external TPU sales revenue, gross profit, Google Cloud segment revenue, and the impact on group earnings.

Methodology notes

• Capacity and unit economicsTPU-aaS unit economics

  Derive revenue and gross profit from TPU capacity per GW, unit ASP, unit cost, and gross margin.

  Barclays assumes that in 2026E each TPU sells for about $20,000, with unit cost of about $15,000, gross profit of about $5,000, and gross margin of about 25%; each GW corresponds to about 714,286 TPUs, or about $14.3 billion of revenue.

• Financial forecastingSegment estimate rebuild

  Rebuild Google Cloud segment revenue and operating profit forecasts.

  The report incorporates TPU-aaS revenue and profit into the Google Cloud model and compares new versus old forecasts, showing that 2028E Google Cloud revenue and operating profit are raised by 128.4% and 70.5%, respectively, versus the prior forecast.

• Valuation methodsEPS and EBITDA multiple valuation

  Maintain the target price using 2027 EPS and EBITDA multiples.

  The report maintains the $425 target price using an average valuation framework of 25x 2027 EPS and 15x EBITDA, rather than the previous blended 2027/2028 framework.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• GOOGL.US

  Core covered security

  Strengths

  Owns in-house TPUs, Google Cloud distribution, and an AI infrastructure partner ecosystem; TPU-aaS could materially expand the revenue pool and improve capital efficiency.

  Weaknesses

  The scale of model upgrades is large, and uncertainty remains around execution, supply, demand, and the timing of financial recognition.

  Comparison

  Compared with the traditional GCP leasing model, TPU-aaS allows Google to sell TPUs directly as a merchant supplier and could drive faster external capacity expansion.

  Risks

  Rising capital expenditures, conservative valuation framework, weaker-than-expected AI compute demand, and delays in partner project execution.

• Anthropic

  Potential TPU-aaS customer and ecosystem beneficiary

  Strengths

  Can gain more direct control over compute, greater customization, and more room for vertical integration.

  Weaknesses

  Needs to bear more complex infrastructure arrangements and economic trade-off analysis.

  Comparison

  Compared with simply leasing compute through GCP or AWS, the SPV/TPU-aaS structure offers greater control.

  Risks

  Cost, contract terms, utilization, and technology migration risks.

• Broadcom

  Beneficiary in the TPU/XPU ecosystem supply chain

  Strengths

  The AI XPV Platform opens a new large-scale revenue pool and may diversify customer and project sources.

  Weaknesses

  Not all of the 20GW capacity belongs to Google TPU, and it also involves other XPU/ASIC projects.

  Comparison

  The report views Broadcom's disclosure on 20GW of AI XPV as an important clue for assessing total TPU-aaS capacity.

  Risks

  Project mix, delivery timing, and changes in customer chip design.

• Blackstone/Apollo

  AI infrastructure financiers and SPV organizers

  Strengths

  Can scale investment in AI compute infrastructure through SPV, debt, and equity structures.

  Weaknesses

  Need to rely on long-term customer demand, asset financing conditions, and data center delivery.

  Comparison

  Compared with hyperscalers' self-build model, the SPV model involving financial institutions is more of an asset-financing platform.

  Risks

  Financing costs, asset utilization, data center delivery, and customer concentration.

• Fluidstack

  Managed service and colo operator

  Strengths

  May expand its footprint as a managed service provider or TPU neocloud.

  Weaknesses

  Business expansion depends on execution of large customers and infrastructure projects.

  Comparison

  In this structure, it plays the hosting and management role outside Google's data center walls.

  Risks

  Operational delivery, scaling, and customer dependency risks.

Key data

• RatingOverweightMaintained unchanged.
• Target priceUSD 425.00The report maintains the target price.
• Current priceUSD 317.69Price date is 2026-07-23.
• Potential upside+33.8%Source is the estimate on the report cover page.
• 2028E external TPU GW11.5GWIncludes the Broadcom-Apollo-Blackstone AI XPV Platform and the Blackstone-Google TPU Cloud JV.
• 2028E GOOGL external TPU revenue$252.7BEstimated under the report's external TPU sales framework.
• 2028E GOOGL external TPU gross profit$63.2BBased on roughly 25% gross margin.
• 2028E upside to consensus revenue36%The report says external TPU revenue represents potential upside versus current consensus for total Alphabet revenue.
• 2028E upside to consensus gross profit15%The report says external TPU gross profit represents potential upside versus current consensus for Alphabet gross profit.
• 2Q26 TPU backlog estimate约$88BThe report estimates Google had about $88 billion of TPU revenue backlog as of 2Q26.
• Purchase commitments$811BThe report cites purchase commitments disclosed in the 10-Q.

Impact & implications

If Barclays' estimates are realized, TPU-aaS would reshape the market's understanding of Alphabet's ability to commercialize AI infrastructure, the growth slope of Google Cloud, and capital efficiency. For GOOGL shareholders, this could mean higher cloud revenue and operating profit; for the AI ecosystem, TPU could evolve from internal Google infrastructure into a broader industry standard.

Risks

• TPU-aaS capacity, deployment, and revenue recognition may come in below the report's assumptions.
• The report acknowledges that the magnitude of estimate increases is large, and actual results over the next three years remain highly uncertain.
• The capital expenditure outlook is elevated, which may weigh on valuation or cash flow expectations.
• Differences in customer economics between TPU-aaS and the GCP leasing model may affect the pace of adoption by AI labs.
• Coordination among SPVs, data centers, chip suppliers, financiers, and managed service providers creates execution risk.
• After the large upward revision to Google Cloud revenue and profit, delays in projects or insufficient demand could create significant downside to expectations.
• Barclays discloses that it may have business relationships with covered companies, and investors should consider potential conflicts of interest.

What to watch

• Progress of the first 0.5GW coming online in 2027 for the Blackstone-Google TPU Cloud JV.
• The actual mix of TPU and other XPU within the Broadcom/Blackstone/Apollo AI XPV Platform's 20GW target by 2028.
• Further disclosures in Google Cloud earnings on TPU-aaS revenue, inventory, purchase commitments, and backlog.
• Whether AI labs such as Anthropic, OpenAI, and Meta adopt similar external TPU/XPU structures.
• Whether external TPU sales approach the estimates of 11.5GW and $252.7 billion of revenue by 2028.
• Whether Alphabet reduces future capital expenditure growth through a more asset-light AI infrastructure model.
• Whether Barclays shifts its valuation framework from 2027 to 2028 in future reports.