In 2025, four technology companies will collectively spend more than $320 billion on AI infrastructure. That number is not a projection — it is confirmed guidance from each company's CFO.
To put it in context: $320B is larger than the entire GDP of Portugal. It is roughly five times the total venture capital deployed in the US in 2024. And it is being committed not over a decade but in a single fiscal year.
The big tech AI capex supercycle is real, it is accelerating, and it is reshaping every layer of the technology stack — from GPU supply chains to energy markets to the competitive dynamics for AI startups. Here is what the data actually shows and what it means.
The Four Companies Driving the Big Tech AI Capex Supercycle
These are not estimates. Each figure reflects publicly disclosed or confirmed capital expenditure guidance for fiscal year 2025, as reported in earnings calls and investor filings. Track real-time updates on the AI Spending Dashboard.
| Company | 2025 Capex Guidance | YoY Change | Primary AI Focus |
|---|---|---|---|
| Amazon (AWS) | $100B+ | +35% | Trainium2 clusters, AWS regions, Bedrock infrastructure |
| Microsoft | ~$80B | +60% | Azure AI compute, OpenAI partnership, Copilot infrastructure |
| Google (Alphabet) | ~$75B | +55% | TPU v5 clusters, Gemini training, data center expansion |
| Meta | $60–72B | +67% | Llama training compute, NVIDIA clusters, US data centers |
| Total | $315–332B | ~+52% | Largest coordinated infrastructure spend in tech history |
Where the Money Is Actually Going
Most coverage frames the AI capex supercycle as "buying NVIDIA GPUs." That is accurate but incomplete. A typical hyperscaler AI infrastructure build breaks down across five categories:
GPU Clusters (40–50% of spend)
NVIDIA H100, H200, and Blackwell GB200 NVL systems. At ~$30K–$40K per H100 and $60K–$70K per GB200, even modest cluster sizes cost billions. Microsoft alone has ordered enough GPUs to power several hundred thousand concurrent AI inference requests.
Data Center Construction (25–35%)
New builds in Virginia, Texas, Iowa, and internationally. Google is building or expanding in 12 countries. Meta broke ground on a 2GW+ facility in Louisiana. Lead times for purpose-built AI data centers now run 18–24 months.
Networking Infrastructure (10–15%)
InfiniBand and high-speed Ethernet to connect GPU clusters at scale. NVIDIA's networking division (formerly Mellanox) is a major beneficiary. Arista Networks and Juniper have also seen order books expand significantly.
Power & Cooling (10–15%)
A single modern GPU cluster consuming 100MW+ needs purpose-built power delivery and cooling. Companies are signing direct power purchase agreements with utilities and nuclear operators. Microsoft signed a deal to restart Three Mile Island.
Custom Silicon Development (5–10%)
Google's TPU v5, Amazon's Trainium2, and Microsoft's Maia chip reduce NVIDIA dependence over time. These are multi-year bets costing $2–5B per generation but pay off at the scale these companies operate.
The Second-Order Effects Nobody Talks About
When four companies simultaneously ramp capex by 50%+ in a single year, the ripple effects extend well beyond the AI industry:
Power Grid Stress
The US electric grid was not designed for 100MW+ data center clusters. Dominion Energy has 30GW of data center capacity in the queue in northern Virginia alone — more than the state's entire current generating capacity. New AI infrastructure is consuming the equivalent of 8–10 nuclear plants annually.
Real Estate Markets
Data center REIT Equinix and Digital Realty have seen demand outpace supply in every major market. Land adjacent to power substations in rural Virginia, Iowa, and Texas has appreciated 300–500% since 2023.
Construction Labor
Hyperscaler data center builds are competing with semiconductor fab construction (TSMC Arizona, Intel Ohio) and clean energy projects for the same pool of specialized electrical engineers and construction workers.
NVIDIA's Market Position
NVIDIA generated $78B in data center revenue in FY2024 — up from $15B in FY2023. The $320B supercycle is the primary demand driver. NVIDIA captures an estimated $0.40–0.50 of every dollar spent on AI compute, giving it an effective 40–50% take rate on the entire capex cycle.
Inference Cost Compression
As GPU supply scales, inference costs continue to fall. The cost to run GPT-4 class inference dropped approximately 95% between 2023 and 2025. This is deflationary for AI infrastructure startups but highly beneficial for application-layer companies building on top of cheap compute.
What This Means for AI Startups and VCs
As a VC who has tracked this market through 65+ investments, I see the AI capex supercycle creating both opportunity and compression for startups:
Tailwinds for Application-Layer Startups
- ✓ Inference costs dropping 80–95% makes AI products economically viable at scale
- ✓ Hyperscaler APIs become commoditized, removing moat advantage from model providers
- ✓ Enterprise budgets shift from "can we afford AI" to "which AI product wins"
- ✓ Distribution and workflow integration matter more than raw compute access
Headwinds for Infrastructure Startups
- ✕ Hard to compete with hyperscalers who are spending $80B+ on the same problem
- ✕ Inference margin compression hits GPU-rental businesses hardest
- ✕ Open-source model releases from Meta undermine proprietary model pricing
- ✕ Hyperscalers acquire promising infrastructure startups before IPO window opens
The smart VC money is moving up the stack. Infrastructure-layer bets made sense in 2022–2023 when the picks-and-shovels thesis was fresh. In 2026, the highest-return opportunities are in vertical AI applications — software that uses the infrastructure being built at $320B scale and turns it into defensible enterprise workflows. You can monitor how the market is pricing this shift on the AI Valuations Dashboard.
Can the Big Tech AI Capex Supercycle Sustain Itself?
The honest answer is: probably not at this exact pace, but a plateau is not the same as a crash.
The four companies driving the supercycle are doing so for rational competitive reasons. Each one fears being left behind if it slows down while the others accelerate. This creates a coordination problem where the rational individual strategy (keep spending) produces a collective outcome that may overshoot actual near-term demand.
The signals to watch for a cycle peak:
GPU lead times normalize from 12+ months to under 6 months — indicating supply has caught up to demand
Any of the four companies guides capex down year-over-year for the first time since 2022
Enterprise AI adoption metrics (daily active AI users per Fortune 500) plateau below model provider revenue projections
NVIDIA's data center revenue growth decelerates below 20% year-over-year for two consecutive quarters
Power purchase agreement prices in key markets decline, signaling data center build slowdown
None of these signals are flashing as of mid-2026. GPU lead times remain elevated, all four companies reaffirmed or raised guidance in their most recent earnings calls, and enterprise AI adoption is accelerating across financial services, healthcare, and software development. The Big Tech Earnings Dashboard tracks quarterly capex actuals vs guidance in real time.
The $300B AI capex supercycle is the largest infrastructure buildout since the interstate highway system.
The companies that own the rails will tax every train that runs on them for the next decade — and startups building on top of cheap, abundant compute will capture the rest.
Track big tech AI infrastructure spending and capex trends on the AI Spending Dashboard and Big Tech Earnings Dashboard at Value Add VC. Originally published in the Trace Cohen newsletter.