BofA raised capex estimates for hyperscale internet cloud operators, and AI data center capacity monetization became the key to valuation upgrades

Overview

This is an industry and company research report on capex, capacity expansion, and monetization potential for AI data centers of US internet and e-commerce hyperscale cloud operators. The report focuses on Alphabet, Amazon AWS, and Meta, and in a market environment where there is concern that high capex is pressuring free cash flow and long-term margins, BofA raised capex estimates and used installed GW capacity, revenue per GW, and implied capacity valuation to assess the potential returns of AI infrastructure investment.

Core views

Core views include: first, AI capex return remains the market's biggest dispute, but constrained capacity supply and rising new AI workload demand support hyperscale cloud operators prioritizing further expansion. Second, Amazon is expected to add the most capacity in 2026-2027 and has lower incremental GW cost, benefiting from AWS scale, internal chips, and more balanced cloud workloads. Third, Google Cloud shows higher revenue per GW, reflecting differences in its cloud portfolio and TPU/Gemini infrastructure. Fourth, Meta's AI capacity has very low value implied by its current stock price, and valuation reset potential is largest if enterprise AI, subscriptions, or external compute leasing generate visible revenue.

Analysis framework

The report first raises capex estimates based on company capex guidance, capital market financing, supply-chain orders, management commentary, and memory prices; it then converts capex into estimated data-center GW capacity; next, it adjusts monetizable cloud/AI capacity by internal-use share and calculates revenue per GW; finally, it strips out valuation for core businesses such as advertising and retail to estimate the implied valuation of AI/cloud capacity assets.

Methodology notes

• Capex-to-capacity estimationCapex-to-GW capacity triangulation

  Estimate installed capacity using capex, disclosed capacity, and capex per GW of build cost

  The report uses Alphabet, Meta, and Amazon AWS capex forecasts as a starting point and combines Amazon's disclosed GW, industry build-cost estimates, and company-specific chip/facility assumptions to estimate installed data-center capacity for 2026 to 2030.

• Per-capacity monetizationRevenue per GW analysis

  Calculate annual revenue per GW based on adjusted cloud/AI capacity

  The report assumes Google and Amazon each have 30% of capacity allocated to core businesses and Meta has 60% used by core advertising, while the remainder is for cloud or external AI capacity monetization, and estimates potential revenue per GW for AWS, Google Cloud, and Meta.

• Valuation splitImplied value per GW capacity valuation

  Strip out core business valuation first, then calculate per-GW capacity value from residual market value

  Using 2027 core business revenue and 2028 GW capacity as a base, the report values advertising, retail, and subscription businesses using revenue multiples or implied P/E valuations, then allocates the residual enterprise value to AI/cloud capacity and compares pricing gaps across Alphabet, Amazon, and Meta.

Asset mapping & comparison

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

• ALPHABET INC / GOOGL.US

  One of the core coverage names, representing Google Cloud and TPU/Gemini AI infrastructure

  Strengths

  Higher revenue per GW, with a Google Cloud portfolio and self-developed TPU/Gemini architecture supporting high-value AI workloads.

  Weaknesses

  Capex was raised materially; stock performance since financing has been pressured, and the market still doubts AI investment returns.

  Comparison

  The report estimates implied AI capacity valuation of about $110 billion/GW, higher than Amazon and Meta.

  Risks

  Data-center buildout pace, cloud revenue conversion, depreciation and free-cash-flow pressure, and AI capacity utilization below expectations.

• Amazon.com / AMZN.US

  One of the core coverage names, representing AWS cloud and AI data-center capacity expansion

  Strengths

  Expected to add the most capacity in 2026-2027, lowest incremental GW cost, with AWS scale, internal chips, and mixed-workload cloud mix creating cost advantages.

  Weaknesses

  Revenue per GW is lower than Google Cloud, and the market may view AI infrastructure and LLM differentiation as weaker.

  Comparison

  The report estimates Amazon AI capacity valuation at about $59 billion/GW, below Alphabet but far above Meta.

  Risks

  Server supply chain, cloud pricing, AI workload mix, AWS margins, and capex efficiency.

• META PLATFORMS INC / META.US

  One of the core coverage names, representing the shift from advertising platform to AI capacity assets and enterprise AI monetization

  Strengths

  The stock currently assigns a very low implied value to AI capacity, so if enterprise AI, AI subscriptions, Business Agent, or external compute leasing generate revenue, valuation upside is large.

  Weaknesses

  No mature enterprise cloud business yet, higher capex cost per GW, and greater dependence on external GPUs and upstream facility investment.

  Comparison

  The report estimates Meta AI capacity valuation at only about $4 billion/GW, significantly below Alphabet and Amazon.

  Risks

  Enterprise AI commercialization below expectations, excessive ad-capacity use for core business, drag from projects like Reality Labs, and inability to effectively monetize capacity externally.

Key data

• Alphabet capex forecast2026: $195 billion; 2027: $290 billion; 2028: $330 billionThese are above the prior 2026, 2027, and 2028 forecasts of $187 billion, $257 billion, and $310 billion, respectively.
• Meta capex forecast2026: $145 billion; 2027: $185 billion; 2028: $210 billionThese are above the prior 2026, 2027, and 2028 forecasts of $130 billion, $157 billion, and $171 billion, respectively.
• Amazon AWS capex forecast2026: $159 billion; 2027: $230 billion; 2028: $276 billionUnchanged in 2026; higher than prior estimates of $196 billion in 2027 and $221 billion in 2028.
• Capacity scale of the three internet giantsAbout 27GW at end-2025, 39GW in 2026, 57GW in 2027The report estimates that total installed capacity of Amazon, Alphabet, and Meta is growing rapidly.
• Incremental capacity added in 2026-2027Amazon about 15GW, Google about 9GW, Meta about 6GWAmazon adds the most capacity.
• 2026 incremental GW costAmazon about $25 billion/GW, Google about $37 billion/GW, Meta about $45 billion/GWDifferences stem from scale, internal chips, build timing of facilities, and share of AI-dedicated capacity.
• 2030 installed capacity forecastAmazon 58.1GW, Alphabet 32.4GW, Meta 22.8GWReflects continued expansion of AI and cloud infrastructure.
• 2026 cloud revenue per GWAWS about $10.6 billion/GW, Google Cloud about $16 billion/GWBelow levels implied by some dedicated AI capacity transactions.
• 2030 revenue potentialAWS $409 billion, Alphabet $387 billion, Meta $110 billionThe report states that AI capacity of hyperscale internet cloud operators could support nearly $900 billion in revenue by 2030.
• Implied capacity valuationAlphabet about $110 billion/GW, Amazon about $59 billion/GW, Meta about $4 billion/GWMeta has the lowest implied valuation for AI capacity.

Impact & implications

For investment implications, the report argues that the market may continue to worry in the near term that high capex is weighing on free cash flow, but if cloud margin improvement, capacity leasing, or enterprise AI revenue is validated, investor confidence in AI capex returns could rise. Amazon benefits from the largest incremental capacity and lower unit cost; Alphabet benefits from higher revenue per GW and TPU/Gemini differentiation; Meta has the largest potential for rerating because current implied AI capacity valuation is extremely low once external monetization evidence appears.

Risks

• AI capex returns fall short of expectations, pressuring free cash flow, EPS, and long-term margins.
• Power availability may constrain data-center expansion; the report cites U.S. Department of Energy estimates that data centers may consume 9% of U.S. electricity generation by 2030.
• Rising construction costs, GPU and memory prices, cooling, land, and power infrastructure costs could increase cost per GW.
• Capacity ramp-up lags; capex occurs first while revenue and depreciation effects lag, potentially creating short-term financial metric pressure.
• Meta's external AI capacity sales and enterprise AI business remain early-stage with limited revenue visibility.
• The valuation framework relies on many assumptions, including core business valuation multiples, internal-use allocation of capacity, and 2028 capacity scale.

What to watch

• 2Q cloud revenue and cloud margins, especially margin improvement quarter over quarter at AWS and Google Cloud.
• Amazon cloud pricing, Bedrock demand, and changes in server supply-chain orders.
• Revenue realization from high-value AI workloads tied to Google Cloud, Gemini, and TPU.
• Commercial progress of Meta Enterprise Solutions, Meta AI subscriptions, and Meta Business Agent.
• Pricing and volume in capacity leasing or dedicated AI data-center deals from Anthropic, Google, SpaceX, Crusoe, and others.
• Changes in DRAM, GPU, networking equipment, and power infrastructure prices.
• Actual impact of high capex on depreciation, free cash flow, EPS, and valuation multiples.