Barclays maintains Overweight on GOOGL, believes TPU-aaS is underestimated by the market

Overview

This is a company research report by Barclays on Alphabet Inc., focused on the TPU-as-a-service business model that Google is advancing. The report believes Google is selling TPU inventory to external customers or SPVs in a merchant supplier model, allowing AI labs to gain more direct control over computing power outside GCP; this model could become one of Alphabet's most important new businesses over the next few years.

Core views

The core view is that the scale and financial impact of TPU-aaS may be underestimated by the market. The report notes that the 20GW-scale AI XPV platform, the Blackstone-Google JV, the USD 811bn purchase commitments disclosed in Google's 10-Q, the USD 275bn+ increase in backlog over the past 6 months, and Broadcom's commitment to 20+GW of capacity all suggest that external AI compute demand could be extremely large. Barclays therefore significantly raises its Google Cloud revenue and operating profit estimates, while still maintaining its USD 425.00 target price.

Analysis framework

The report analyzes from two angles: ecosystem structure and unit economics. It first explains how AI labs can purchase Google TPUs through SPVs, with deployment by data centers and hosting service providers such as Fluidstack, and lease capacity to customers such as Anthropic; it then estimates the number of TPUs per GW, the selling price per TPU, gross margin, the ramp-up of external TPU capacity, and Google Cloud's revenue and gross profit contribution, finally incorporating these estimates into Alphabet's 2026E-2028E financial model and valuation framework.

Methodology notes

• Company fundamentalsSegment revenue and gross profit reconstruction

  Google Cloud model reconstruction

  Barclays incorporates TPU-aaS revenue and profit into the Google Cloud segment and reassesses revenue, gross profit, operating profit, and cash flow for 2026E-2028E.

• Unit economicsTPU-aaS unit economics estimation

  Revenue and gross profit estimation per GW

  The report assumes that about 700k TPUs can be deployed per GW in 2026, that Google sells them to SPVs at about USD 20k per unit, and earns a gross margin of about 25%, corresponding to about USD 14bn of revenue per GW and about USD 5k gross profit per TPU.

• Valuation methodsP/E and EV/EBITDA valuation framework

  2027 valuation benchmark

  The report maintains its USD 425.00 target price, but rolls back the valuation framework to 2027, using an average of 25x EPS and 15x EBITDA to more prudently reflect TPU-aaS contribution and capex uncertainty.

Asset mapping & comparison

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

• ALPHABET INC (GOOGL.US)

  Core beneficiary and covered company in the report

  Strengths

  Owns TPU design, Google Cloud customer base, and AI infrastructure capabilities; TPU-aaS could bring a new revenue pool, gross profit contribution, and improved capital efficiency.

  Weaknesses

  AI infrastructure still requires massive capital investment, and the report also notes that 2028 capex estimates could be as high as USD 500bn.

  Comparison

  Compared with the traditional GCP leasing model, TPU-aaS allows Google to sell TPUs into external SPV structures, with different revenue recognition and capital burden.

  Risks

  Capacity ramp below expectations, weaker-than-expected unit economics, insufficient customer conversion, capex pressure, and regulatory or contractual constraints.

• Broadcom Inc. (AVGO)

  Co-designer of TPUs and key semiconductor partner for the AI XPV platform

  Strengths

  The 20GW-scale AI XPV platform opens a new semiconductor revenue pool and may expand business to more AI labs and custom XPU customers.

  Weaknesses

  Part of the 20GW capacity includes non-Google TPUs and other custom ASICs, so the actual share attributable to Google TPUs is uncertain.

  Comparison

  Broadcom refers to the platform as XPU, while the report refers to it as TPU-as-a-service from the perspective of GOOGL shareholders.

  Risks

  Platform execution, customer chip design progress, and supply chain and capacity delivery risks.

• Anthropic

  Potential major TPU-aaS customer and AI lab demand-side participant

  Strengths

  Can gain deeper infrastructure control, customization capability, and vertical integration flexibility, while reducing part of its dependence on traditional hyperscaler service agreements.

  Weaknesses

  Owning or controlling compute capacity through SPVs may bring more complex financing, operations, and long-term commitments.

  Comparison

  Compared with directly leasing TPUs from GCP, the TPU-aaS structure provides more control, but with a different cost structure and greater execution complexity.

  Risks

  Financing costs, utilization rates, changes in model training demand, and service agreement risks.

• Blackstone / Apollo / SPV structure

  AI infrastructure financing and asset holding platform

  Strengths

  Purchases compute assets such as TPUs through equity and debt financing, supporting large-scale AI capacity buildout.

  Weaknesses

  Highly dependent on long-term customer leases, asset residual value, and financing market conditions.

  Comparison

  SPVs partially separate the balance sheet for compute assets from Google's own data center investments.

  Risks

  Deterioration in financing conditions, customer default, insufficient asset utilization, and delays in data center delivery.

• Fluidstack

  Hosting service and colo managed service provider

  Strengths

  Can manage TPU clusters outside Google's data center walls, expanding its AI infrastructure hosting footprint.

  Weaknesses

  Needs to prove large-scale TPU cluster operations capability and service reliability.

  Comparison

  Its role is closer to an external hosting and managed Kubernetes service provider than to a traditional cloud vendor.

  Risks

  Execution capability, service levels, supply chain coordination, and customer concentration risks.

Key data

• Target priceUSD 425.00Maintained unchanged.
• Current priceUSD 317.69As of 23-Jul-2026.
• Potential upsideapproximately 33.8%Calculated based on target price of USD 425.00 and current price of USD 317.69.
• RatingOverweightGOOGL rating remains unchanged.
• Industry viewPositiveU.S. Internet industry view remains unchanged.
• External TPU revenue2026E USD 20.0bn; 2027E USD 67.7bn; 2028E USD 252.7bnThe report estimates the incremental contribution of TPU-aaS to Alphabet's revenue.
• External TPU gross profit2026E USD 5.0bn; 2027E USD 16.9bn; 2028E USD 63.2bnEstimated based on a gross margin of about 25%.
• 2028E revenue upsideapproximately 36%Relative to current consensus Alphabet total revenue.
• 2028E gross profit upsideapproximately 15%Relative to current consensus Alphabet gross profit.
• Potential external TPU capacity2027E 3.7GW; 2028E 11.5GWThe report estimates this based on announced platform and JV capacity ramp-up.
• Disclosed purchase commitmentsUSD 811bnThe report says this figure comes from this week's 10-Q disclosure.
• Estimated TPU backlogapproximately USD 88bnThe report estimates Google TPU revenue backlog as of 2Q26.

Impact & implications

If TPU-aaS is successfully rolled out, Alphabet could expand from simply renting compute capacity within GCP to selling TPUs within the external AI infrastructure ecosystem and sharing in hardware gross profit. This could increase Google Cloud's revenue scale, broaden its influence in the AI infrastructure industry, and ease some future capex pressure through a more asset-light structure. However, the model is highly sensitive to capacity deployment, customer adoption, SPV financing, supply chain execution, and the durability of AI compute demand.

Risks

• TPU-aaS capacity ramp-up, data center construction, or supply chain delivery may fall short of the report's estimates.
• The pace or scale of TPU-aaS adoption by AI labs may be below expectations.
• Changes in SPV financing costs, debt availability, or asset collateral structures may affect project economics.
• Google Cloud revenue recognition, gross margin, or accounting treatment may differ from the report's assumptions.
• Continued increases in future capital expenditures may offset the capital efficiency gains from a more asset-light structure.
• Uncertainty related to hyperscaler service agreements, acceptable use policies, regulatory disclosures, or customer contracts.
• A slowdown in Alphabet's core advertising business or a shift toward lower-margin businesses may pressure overall valuation multiples.

What to watch

• The progress of the first 0.5GW launch of the Blackstone-Google JV in 2027 and subsequent expansion plans.
• The actual deployment pace of the Broadcom/Blackstone/Apollo AI XPV platform toward 20GW of capacity by 2028.
• The actual share of TPUs within the 20GW platform, and the diversion of capacity by non-TPU custom ASICs such as those for OpenAI.
• Further disclosures on Google Cloud revenue, operating profit, inventory, and purchase commitments.
• Whether Anthropic and other AI labs sign larger-scale long-term TPU-aaS leasing or purchase agreements.
• Whether assumptions on TPU price per unit, deployment density per GW, and roughly 25% gross margin can be validated by future financial data.
• Whether Alphabet's 2027/2028 valuation framework moves upward as TPU-aaS visibility improves.