Stargate, Colossus, and the New Arms Race for AI Infrastructure
The world's largest companies are pouring $700 billion into AI data centers in 2026 alone. The power grid can't keep up, the revenue math doesn't add up, and the environmental costs are mounting. Inside the biggest infrastructure bet since the transcontinental railroad.
By Henrik Larsson, Climate Tech · Mar 9, 2026
Hyperscalers are spending $700B on AI infrastructure in 2026 while facing a 6 GW power shortfall, a $600B revenue gap, and mounting environmental backlash. A data-driven analysis of the AI infrastructure arms race from Stargate to Colossus.
Frequently Asked Questions
How much are tech companies spending on AI infrastructure in 2026?
The five largest hyperscalers -- Amazon, Alphabet/Google, Microsoft, Meta, and Oracle -- are projected to spend a combined $610-715 billion on capital expenditure in 2026, with roughly 75% ($450B+) going directly to AI infrastructure including GPUs, servers, and data centers. This represents a 36% increase over 2025 spending and roughly triple the level from two years ago. Amazon leads at approximately $200 billion, followed by Alphabet at $175-185 billion, Microsoft at $145 billion, Meta at $115-135 billion, and Oracle at $50 billion.
What is the Stargate Project and how much does it cost?
Stargate is a $500 billion AI infrastructure joint venture announced in January 2025 by OpenAI, SoftBank, Oracle, and MGX (an Abu Dhabi sovereign wealth-backed fund). SoftBank and OpenAI each hold 40% ownership with $19 billion commitments each. The project plans nearly 7 gigawatts of data center capacity across multiple US sites, with its flagship facility in Abilene, Texas already operational. However, as of late 2025, reports emerged of unresolved disputes between partners and concerns that meaningful construction had stalled.
What is xAI's Colossus supercomputer and why is it controversial?
Colossus is xAI's supercomputer in Memphis, Tennessee, currently running 230,000 GPUs (150,000 H100s, 50,000 H200s, and 30,000 GB200s). It was built in just 122 days and is expanding to 2 gigawatts and 555,000 GPUs at a cost of $18 billion. The facility is controversial because xAI built and operated natural gas turbines without required Clean Air Act permits. The turbines emit 1,200-2,000 tons of nitrogen oxides per year, increasing Memphis smog by an estimated 30-60%, in a predominantly Black neighborhood with Tennessee's highest child asthma hospitalization rate.
Why is nuclear power making a comeback because of AI?
AI data centers require enormous amounts of continuous, carbon-free electricity that renewables alone cannot provide. Microsoft signed a 20-year deal to restart Three Mile Island's Unit 1 reactor (835 MW) exclusively for its AI operations. Meta announced a 6.6 GW nuclear procurement strategy for its Prometheus AI project. Google partnered with Kairos Power to deploy small modular reactors (SMRs). Amazon spent $650 million on a campus adjacent to the Susquehanna nuclear plant. These deals have made 2026 the year nuclear power is reclaiming relevance, with 15 reactors globally either under construction or restarting.
What is the $600 billion AI revenue gap that Sequoia identified?
Sequoia Capital partner David Cahn published an analysis showing that AI capital spending would require approximately $2 trillion in annual AI revenue by 2030 to justify the investment -- but current AI revenues are roughly $20 billion per year, creating a gap that requires a 100x increase. Even with optimistic projections, a $500 billion annual gap remains. Americans currently spend only $12 billion per year on AI services, and capital-intensive firms have historically underperformed conservative peers by 8.4% annually.
How does the DeepSeek breakthrough affect AI infrastructure spending?
DeepSeek's R1 model, trained for just $5.6 million using 2,000 H800 GPUs versus $80-100 million and 16,000 H100s for comparable Western models, demonstrated that frontier AI capability is achievable at a fraction of the cost. This creates a paradox: efficiency gains could reduce infrastructure spending by 30-50% in moderate scenarios, but the Jevons Paradox argument suggests that cheaper AI will drive more demand and therefore more infrastructure needs. The debate remains unresolved, but DeepSeek's success challenges the assumption that raw compute scale is an unassailable competitive moat.
Related Articles
Topics: AI Infrastructure, Energy, Data Centers, Geopolitics
Browse all articles | About Signal