Bank of America raised internet capex estimates, highlighting long-term monetization potential for AI data center capacity
AI summary card
Bank of America raised internet capex estimates, highlighting long-term monetization potential for AI data center capacity
The report raises Alphabet, Meta, and Amazon AWS capex to a higher capacity buildout path and points to Meta's AI capacity being significantly underappreciated by the market through GW capacity, revenue per GW, and implied valuation analysis.
- Bank of America believes the core debate for hyperscale internet cloud providers remains AI capex returns, with near-term free cash flow pressure likely to rise, but the monetization indicators for capacity are improving.
- The report raises 2027-2028 capex forecasts for Alphabet, Meta, and Amazon AWS, and expects combined installed capacity across the three companies to rise from about 27GW at the end of 2025 to about 57GW in 2027.
- Amazon is expected to add the most capacity in 2026-2027, about 15GW, and its incremental capacity cost in 2026 is about $25bn/GW, below Google at about $37bn/GW and Meta at about $45bn/GW.
- The report estimates 2026 AWS revenue per Cloud GW at about $10.6bn and Google Cloud at about $16bn, below the recent level of over $40bn/GW in dedicated AI capacity transactions.
- Valuation analysis shows implied AI/Cloud capacity values of about $110bn/GW, $59bn/GW, and $4bn/GW for Google, Amazon, and Meta, respectively, with Meta’s AI capacity asset receiving the lowest market value.
Report interpretation
Overview
This report focuses on AI capital expenditure, data center capacity expansion, and capacity monetization potential for hyperscale cloud providers such as Alphabet, Amazon, and Meta in the U.S. internet and e-commerce segment. Bank of America argues that investors are currently most concerned that higher capex will weigh on free cash flow and margins, but recent cloud pricing, capacity leasing, LLM demand, and enterprise AI product signals show that AI data center capacity still may translate into meaningful revenue.
Core views
Core views include: first, hyperscale cloud providers will prioritize capacity needed for AI training, inference, cloud workloads, and internal AI product deployment over short-term free cash flow optimization; second, Amazon is expected to add the most capacity in 2026-2027 with the lowest unit cost, giving it the greatest capacity-driven incremental revenue potential; third, Google Cloud has higher revenue per GW, benefiting from TPU/Gemini and higher-value AI workloads; fourth, although Meta has not yet built a mature enterprise cloud business, its AI capacity is reflected at only about $4bn/GW in stock valuation, and if enterprise AI, subscriptions, and external compute leasing materialize, revaluation upside is substantial.
Analysis framework
The report first raises capex forecasts, then converts capex into installed data center capacity in GW, estimates revenue per GW based on each company’s AI/Cloud capacity mix, and finally strips out ad, retail, and other core business valuation to compute the implied AI/Cloud capacity value per GW in the stock.
Methodology notes
Estimate data center capacity from company capex, per-GW build cost, and capacity ramp timing.
The report cross-validates company capex forecasts, Amazon-disclosed GW clues, and industry estimates of AI data center build cost, assuming that data center capacity build cost in 2026 is about $25bn-$45bn per 1GW.
Divide cloud revenue or potential external AI capacity revenue by adjusted AI/Cloud capacity.
The report assumes 30% of Google’s and Amazon’s capacity is used for core businesses and 60% of Meta’s capacity is used for core advertising, then estimates revenue yield from the remaining AI/Cloud capacity.
First strip out valuation of core advertising, retail, and subscription businesses, then compare remaining enterprise value to 2028 capacity.
The report uses 2027 core business revenue and different revenue multiples to estimate core business value; the residual value is allocated to AI/Cloud capacity assets to compare implied per-GW valuation across Google, Amazon, and Meta.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- ALPHABET INC / GOOGL.USAI infrastructure and Google Cloud capacity expansion beneficiary.
- Strengths
- Google Cloud has higher revenue per GW, supported by TPU/Gemini and higher-value AI workloads that support a premium valuation.
- Weaknesses
- Raised capex and financing arrangements may intensify investor concerns around free cash flow and returns.
- Comparison
- Implied AI capacity valuation is about $110bn/GW, above Amazon and Meta.
- Risks
- Higher per-unit build cost, uncertain capacity ramp timing, and added capex pressure on margins and cash flow.
- Amazon.com / AMZN.USBeneficiary of AWS capacity expansion and cloud revenue growth.
- Strengths
- Expected to add the most capacity in 2026-2027, with unit incremental GW cost in 2026 around $25bn, below Google and Meta.
- Weaknesses
- Revenue per GW is lower than Google Cloud, and the market may view Amazon’s AI infrastructure and LLM differentiation as weaker than Google’s.
- Comparison
- Implied AI capacity valuation is about $59bn/GW, below Google but materially above Meta.
- Risks
- Server supply, energy, cloud pricing, and changes in AI workload mix could affect capital returns.
- META PLATFORMS INC / META.USAI data center capacity, enterprise AI, and external compute monetization are candidates for potential re-rating.
- Strengths
- The report believes the market assigns very little value to Meta’s capacity buildout; enterprise AI, Meta AI subscriptions, and Business Agent could provide incremental revenue sources.
- Weaknesses
- Meta lacks a mature enterprise cloud business, so the path for realizing capacity revenue is more uncertain than for Amazon and Google.
- Comparison
- Implied AI capacity valuation is about $4bn/GW, far below Google and Amazon, implying the largest re-rating potential.
- Risks
- High capex, GPU dependence, uncertain demand from external customers, and risks from core advertising capacity usage share and margin.
Key data
- Alphabet capex forecast2026 $195bn, 2027 $290bn, 2028 $330bnPreviously forecast $187bn, $257bn, and $310bn for 2026, 2027, and 2028, respectively.
- Meta capex forecast2026 $145bn, 2027 $185bn, 2028 $210bnPreviously forecast $130bn, $157bn, and $171bn for 2026, 2027, and 2028, respectively.
- Amazon AWS capex forecast2026 $159bn, 2027 $230bn, 2028 $276bnUnchanged for 2026, raised from $196bn and $221bn in 2027 and 2028.
- Three internet majors' capacityabout 27GW at end-2025, 39GW in 2026, 57GW in 2027Amazon is expected to add 15GW in 2026-2027, Google 9GW, and Meta about 6GW.
- 2030 capacity forecastAlphabet 32.4GW, Amazon 58.1GW, Meta 22.8GWThe report believes capacity growth would support long-term AI and cloud revenue upside.
- 2026 incremental capacity cost per unitAmazon about $25bn/GW, Google about $37bn/GW, Meta about $45bn/GWAmazon benefits from scale, in-house chips, and a more balanced cloud workload mix, while Meta’s higher cost reflects land, construction, and GPU-related upfront investment.
- 2026 revenue per GWAWS about $10.6bn/GW, Google Cloud about $16bn/GW, Meta potential external capacity about $12bn/GWRecent capacity transaction estimates with Anthropic and Google’s deal with SpaceX and others exceeded $40bn/GW, indicating upside.
- 2030 revenue potentialAWS $409bn, Alphabet $387bn, Meta potential $110bnTotaling close to around $900bn in AI/Cloud capacity-related revenue potential.
- Implied AI/Cloud capacity valuationGoogle about $110bn/GW, Amazon about $59bn/GW, Meta about $4bn/GWMeta’s AI capacity asset receives the lowest implied value in current stock valuation.
Impact & implications
The investment implication of the report is that the market may be placing too much emphasis on the pressure higher AI capex creates on near-term free cash flow and margins, while underestimating the long-term monetization from capacity scarcity, improving cloud pricing, external compute leasing, and enterprise AI products. Relative to peers, Amazon has the largest incremental capacity and a lower-cost advantage, Google has higher revenue per GW and AI infrastructure differentiation, while Meta has the most pronounced implied valuation discount and potential for re-rating.
Risks
- AI capex returns are lower than expected, putting pressure on free cash flow and valuation.
- Data center power supply shortages constrain capacity build and ramp pace.
- AI servers, GPUs, memory, networking, and power equipment costs are higher than expected.
- Demand for cloud and AI capacity comes in below expectations, leading to low utilization or revenue per GW below model assumptions.
- Meta’s external AI capacity sales, enterprise AI, or subscription business fails to scale into meaningful revenue.
- The report’s capacity, cost, and valuation analysis depends on multiple assumptions, and company-level disclosure is limited.
What to watch
- Quarterly changes in Amazon and Google cloud revenue growth and cloud margins.
- Further revisions to Alphabet, Meta, and Amazon capex guidance for 2026-2028.
- Power, land, construction cycle, and equipment supply constraints for AI data centers.
- Pricing, adoption, and revenue disclosure for AWS, Google Cloud, and Meta enterprise AI-related products.
- Verification of specialist AI compute pricing through capacity transactions by players such as Anthropic, SpaceX, and Crusoe.
- Commercialization progress of Meta Enterprise Solutions, Meta AI subscriptions, and Business Agent.