AI infrastructure

Amazon (~$200B), Alphabet (guidance raised to $180-190B from $175-185B), Microsoft (tracking toward roughly $190B by some 2026 estimates), and Meta (guidance raised mid-year to $125-145B) guided to a combined ~$725B of 2026 capex, up ~77% on 2025's ~$410B (CNBC reporting on Q1-Q2 2026 earnings calls). The IEA separately verified that capex at the top 5 tech firms surged past $400B in 2025 and would rise a further ~75% in 2026. Beyond the hyperscalers, the GCC region (Saudi Arabia, UAE, Qatar) is layering on an estimated USD 5-7B of AI data-centre investment in 2026 alone (Analysys Mason), part of a stated USD 30B+ pipeline of GCC AI data-centre capacity through 2030.

Mixed. Headwind: the US Commerce Department in late May 2026 moved to close a loophole that let Chinese-owned entities acquire Nvidia's most advanced chips (Blackwell, Rubin) and AMD's MI350X via overseas subsidiaries, extending export-licence requirements to Chinese-headquartered buyers wherever located -- reversing a brief December 2025 easing under a case-by-case licensing regime. Tailwind: national governments are fast-tracking sovereign AI infrastructure as industrial policy -- Saudi Arabia's HUMAIN and the USD 8-10B "Stargate UAE" project (OpenAI, Oracle, Nvidia, Cisco, G42, SoftBank) both carry direct state backing and expedited permitting, with first phases of 200MW-1GW targeted for 2026-2027.

Indeed Hiring Lab (July 14, 2026) finds data-center job postings have more than doubled over two years: roughly 6 of every 1,000 US job postings are now data-center-related, up from 2 per 1,000 in May 2023, with hiring for the AI-infrastructure buildout spreading beyond hyperscaler-HQ metros into smaller markets (e.g. Columbus, OH; Reno, NV).

WIPO's PCT Yearly Review 2025 recorded 275,900 international patent applications, up 0.7% year-on-year; semiconductors were among the fastest-growing of the top 10 technology fields (+6.1%), alongside digital communication. Patent-analytics vendor PatSnap separately estimates AI-inference-chip patent filings rose from roughly 11 in 2017 to about 335 in 2025 (a ~30x increase) -- flagged here as a vendor estimate with an acknowledged ~18-month publication lag, directional rather than exact.

A distinct "neocloud" category of GPU-specialist infrastructure providers has scaled rapidly alongside the hyperscalers: CoreWeave ended 2025 with a $66.8B contracted backlog; Nebius raised 2026 ARR guidance to $7-9B on the strength of a $17.4-19.4B multi-year Microsoft deal and a $3B Meta contract; Crusoe was in talks as of July 2026 to raise at roughly a $30B valuation, about 3x its October 2025 mark, built on stranded-gas-powered data centers.

Enterprise and hyperscaler demand is outrunning supply rather than lagging it: CoreWeave's $66.8B backlog and Nebius's multi-billion-dollar Microsoft/Meta compute contracts both point to committed, contracted forward demand rather than speculative capacity-building, and Alphabet management has pointed to AI-driven Cloud backlog growth as evidence of monetized (not just anticipated) demand.

The binding constraint has shifted from chip supply to power delivery. The IEA's April 2026 update found data-centre electricity demand rose 17% in 2025 (AI-focused data centres alone +50%) against 3% global electricity-demand growth, and projects data-centre consumption roughly doubling from 485 TWh (2025) to 950 TWh (2030), with AI-focused demand tripling over that window. Industry trackers report high-voltage substation and transformer lead times of 3-5 years against 12-24 month IT-hardware cycles, driving a shift toward "bring your own power" (on-site gas turbines, in some cases small modular reactors). Physically, the buildout still concentrates in a small number of power-rich clusters -- Northern Virginia ("Data Center Alley"), Texas, and the GCC -- though city-level entities for these clusters are not yet published on this site, so they are described here in prose rather than linked.