| Company | Capex figure shown | What they say the money is for |
|---|---|---|
|
Amazon
|
$200B
2026 capex guidance
▲ about 52% vs 2025 cash PP&E purchases ($131.8B)
|
Amazon guided about $200B in 2026 capex across the company.
In the same statement, Amazon tied the step-up to AI, chips, robotics, and low Earth orbit satellites.
AWS footprint context: 39 AWS Regions and 123 Availability Zones. Source: Amazon Q4 2025 earnings release (Feb 5, 2026) |
|
Alphabet
|
$175B
to $185B (2026E)
▲ about 91% to 102% vs 2025 cash PP&E purchases ($91.447B)
|
Alphabet disclosed expected 2026 capex of $175B to $185B.
The filing’s cash flow statement shows $91.447B of 2025 purchases of property and equipment.
Source: Alphabet (Google) SEC filing dated Feb 4, 2026 |
|
Microsoft
|
$37.5B
Q2 FY2026 capex (quarter)
Reported, not full-year 2026 guidance
|
Microsoft reported $37.5B of capex in Q2 FY2026.
CFO commentary said about two-thirds was short-lived assets (servers, including CPUs and GPUs).
Microsoft also said it added nearly 1 gigawatt of capacity in the quarter.
Source: Microsoft FY2026 Q2 earnings call transcript |
|
Meta
|
$115B
to $135B (2026E)
▲ about 59% to 87% vs 2025 capex ($72.22B)
|
Meta guided 2026 capex of $115B to $135B
(capex includes principal payments on finance leases).
Meta attributed the year-over-year increase to investment supporting its Superintelligence Labs efforts and core business.
Source: Meta Q4 2025 earnings release (Jan 28, 2026) |
|
Stargate
OpenAI · SoftBank · Oracle
|
$500B
over 4 years (announced)
Multi-year commitment, not annual capex guidance
|
OpenAI announced Stargate as a plan to invest $500B over four years and begin deploying $100B immediately.
OpenAI later said Stargate is targeting a 10 gigawatt buildout and announced five additional U.S. sites.
Abilene, Texas buildout context has been reported at roughly 4 million square feet. Sources: OpenAI announcement and follow-up site list |
|
xAI
Acquired by SpaceX (Feb 2026)
|
n/a
no public capex guidance
Not comparable to public-company capex guidance
|
Colossus is the world's largest single-site AI training cluster: 200,000+ GPUs operational (150K H100, 50K H200, 30K+ GB200), built in 122 days.
A second facility (Colossus 2) is bringing 555,000 Blackwell GPUs online (~$18B in Nvidia silicon). A third site ("MACROHARDRR") targets 2 GW total capacity and a roadmap to 1 million GPUs.
Separate reporting in early Feb 2026 said SpaceX acquired xAI.
Regulatory: In Jan 2026, EPA closed the "nonroad engine" loophole xAI used to run up to 35 unpermitted gas turbines at the Memphis site. The NAACP and Southern Environmental Law Center had filed intent to sue under the Clean Air Act (June 2025) and appealed the Shelby County Health Department's turbine permit (July 2025). EPA's final rule (40 CFR Part 60, eff. Jan 15, 2026) now requires Clean Air Act permits for site-bound turbines regardless of whether they're trailer-mounted. Sources: xAI/Colossus site; Wikipedia; CNBC (Jan 16, 2026); SELC; Ars Technica; AP; TechCrunch |