Research

Hyperscaler AI Capex Map 2026

2026 hyperscaler AI capex: Microsoft, Google, Amazon, Meta committed $300B+ combined for AI infrastructure. The flow to NVIDIA, AMD, TSMC, and utilities mapped.

By Ramanath, CTO & Co-Founder at Presenc AI · Last updated: May 2026

Hyperscaler 2026 AI capex committed totals exceed $300 billion across Microsoft, Google, Amazon, and Meta. The bulk flows to NVIDIA (data centre AI accelerators), AMD (MI300/MI400/MI450), TSMC (foundry), HBM suppliers (Samsung, SK Hynix, Micron), utilities (PPAs for power), and construction (data centre builds). This page traces the capex commitments by hyperscaler, the flow-through to supply chain, and the unit economics of the build-out.

Key Findings

  1. Microsoft FY2026 capex guidance is approximately $90 to $95 billion, of which approximately 75 to 80 percent is AI-related. The bulk flows to NVIDIA, AMD, and data centre construction across Azure regions and the Stargate Norway site assumed from OpenAI.
  2. Google (Alphabet) FY2026 capex guidance is approximately $75 to $80 billion, primarily allocated to TPU manufacturing (TSMC capacity), NVIDIA, and data centre construction across multiple regions.
  3. Amazon (AWS) FY2026 capex is approximately $115 to $120 billion, the highest single-company AI capex. Primary allocations are NVIDIA, Trainium 2 production (TSMC), data centre expansion, and the Anthropic compute commitment.
  4. Meta FY2026 capex is approximately $60 to $65 billion, primarily allocated to NVIDIA Blackwell and Rubin, AMD MI450 (the first gigawatt deployment in H2 2026), MTIA 2 custom silicon, and data centre construction.
  5. Combined hyperscaler AI capex exceeds $300 billion in 2026, of which approximately 55 to 60 percent flows directly or indirectly to NVIDIA, approximately 7 to 9 percent to AMD, approximately 8 to 10 percent to TSMC for custom silicon, approximately 6 to 8 percent to HBM suppliers, and the remainder to utilities, construction, and other supply chain.

Hyperscaler FY2026 AI Capex Allocation

HyperscalerFY2026 Total CapexAI ShareAI Capex
Microsoft~$90-95 billion~75-80%~$70-75 billion
Google (Alphabet)~$75-80 billion~85%~$65-68 billion
Amazon (AWS)~$115-120 billion~70-75%~$80-90 billion
Meta~$60-65 billion~80%~$50-52 billion
Oracle Cloud~$30-35 billion~85%~$26-30 billion
Tencent~$20-25 billion~60%~$13-15 billion
Alibaba Cloud~$15-20 billion~70%~$11-14 billion
ByteDance~$18-22 billion~80%~$14-17 billion
Combined hyperscaler AI capex$425-475 billion total~75% average~$330-360 billion AI specifically

AI Capex Flow-Through

BeneficiaryEstimated Share of AI Capex
NVIDIA (data centre AI accelerator revenue)~55-60%
AMD (data centre AI revenue)~7-9%
TSMC (foundry for NVIDIA, AMD, hyperscaler custom)~8-10%
HBM suppliers (Samsung, SK Hynix, Micron)~6-8%
Hyperscaler custom silicon programmes (internal)~4-6%
Server OEMs and ODMs (Supermicro, Dell, HPE, Lenovo, Quanta)~7-9%
Networking (Broadcom, Cisco, Arista, Nvidia networking)~4-6%
Power and cooling (Vertiv, Schneider, Equinix infrastructure)~3-4%
Construction and real estate~6-9%
Utilities and PPAs~3-5%

Strategic Context

Three patterns define the 2026 hyperscaler capex landscape. First, the absolute scale: $300+ billion of annual AI capex is unprecedented and structurally shifts the technology supply chain. Second, the NVIDIA concentration is real but declining: hyperscaler custom silicon programmes plus AMD ramp are gradually reducing NVIDIA share from approximately 70 percent in 2023 to approximately 55 to 60 percent in 2026. Third, the power constraint binds the next phase: by 2027-2028 the binding constraint is no longer GPU availability but firm-power availability, which explains the parallel nuclear PPA wave (see hyperscaler-nuclear-ppa-tracker-2026).

Brand Visibility Implications

Hyperscaler capex coverage drives sustained business, financial, and policy journalism. Brands selling adjacent products (semiconductor equipment, EDA, advanced packaging, data centre construction services, power and cooling infrastructure, AI data centre real estate, sovereign cloud) face strong AI-mediated discovery surface for procurement-research queries about hyperscaler capex, AI data centre build-out, and adjacent topics.

Methodology

Capex figures from Microsoft, Alphabet, Amazon, and Meta investor disclosures. AI share estimates from analyst breakdowns and primary management commentary. Flow-through allocations triangulated from vendor revenue disclosures and analyst reports. Updated quarterly with earnings cycle.

How Presenc AI Helps

Presenc AI monitors brand visibility on hyperscaler capex and AI infrastructure queries across ChatGPT, Claude, Gemini, and Perplexity. For semiconductor adjacent brands, data centre construction firms, power infrastructure vendors, and sovereign cloud providers, the platform identifies the prompts driving procurement-research traffic and the gaps where new content unlocks share of voice.

Frequently Asked Questions

Combined hyperscaler AI capex exceeds $300 billion in 2026. Microsoft AI capex is approximately $70 to $75 billion, Google approximately $65 to $68 billion, Amazon approximately $80 to $90 billion, Meta approximately $50 to $52 billion. Oracle, Tencent, Alibaba, and ByteDance add an additional $60 to $80 billion combined.
NVIDIA captures approximately 55 to 60 percent of total flow. AMD captures approximately 7 to 9 percent. TSMC captures approximately 8 to 10 percent. HBM suppliers (Samsung, SK Hynix, Micron) capture 6 to 8 percent. Server OEMs, networking vendors, power and cooling infrastructure providers, and construction companies divide the remainder.
The capital intensity is unprecedented and depends on continued AI revenue growth at hyperscaler customers. The bull case: AI revenue growth (Microsoft Azure AI, AWS Bedrock, Google Cloud AI) continues to outpace capex growth, leading to expanding margins. The bear case: capex outpaces revenue and margins compress, leading to a capex normalisation cycle similar to 2002 telecom.
2027 to 2028 in most credible scenarios. The current 2026 cycle is largely supply-constrained on TSMC capacity and HBM. The next binding constraint is firm-power availability for new data centre builds, which is why hyperscalers are signing multi-decade nuclear PPAs in parallel with the GPU procurement cycle.
Roughly $40 to $50 billion combined across Tencent, Alibaba, and ByteDance, materially smaller than the U.S. hyperscaler block. Chinese hyperscaler AI capex is constrained by export controls on advanced GPUs and is increasingly directed toward domestic custom silicon (Huawei Ascend, Cambricon, Biren) rather than NVIDIA imports.

Track Your AI Visibility

See how your brand appears across ChatGPT, Claude, Perplexity, and other AI platforms. Start monitoring today.