The race to build the physical backbone of artificial intelligence has entered a frenzied new phase, and according to The Information, the scale of capital flowing into data centers, chips, and power infrastructure now dwarfs anything the tech industry has seen before. The Information reports that the buildout is reshaping not just Silicon Valley balance sheets but entire energy grids, real estate markets, and geopolitical alliances. What stands out here is how quickly infrastructure spend has moved from a back-office concern to the single biggest variable determining who wins the AI era.
This is significant because the bottleneck has shifted. A year ago, the conversation centered on model quality. Now it’s about who can secure GPUs, megawatts, and land fast enough to keep training the next generation of frontier models.
Where the money is actually going
The spending isn’t evenly distributed. A handful of categories are absorbing the bulk of capital:
- Compute clusters: Hyperscalers are committing hundreds of billions to Nvidia-stacked data centers, with multi-gigawatt campuses now the new normal rather than the exception.
- Power generation: Nuclear deals, gas turbines, and grid upgrades have become strategic priorities. AI labs are negotiating directly with utilities and even buying power plants.
- Custom silicon: Google’s TPUs, Amazon’s Trainium, and Microsoft’s Maia chips are no longer side projects. They’re insurance policies against Nvidia dependence.
- Networking and cooling: The unglamorous middle layer (optical interconnects, liquid cooling, high-bandwidth memory) is suddenly a profit center for vendors who barely registered two years ago.
The Information’s reporting underscores how this spend is concentrated among roughly five buyers: Microsoft, Google, Amazon, Meta, and Oracle, with OpenAI and Anthropic shaping demand through massive multi-year commitments.
Why it matters now
Three dynamics make this moment different from previous tech capex cycles.
First, the time horizon. Data centers being announced today won’t come online until 2027 or 2028. Whoever locks in power and land now controls the pipeline for the rest of the decade.
Second, the cash burn. Capex is running ahead of AI revenue, and Wall Street is starting to ask harder questions. The bull case rests on the assumption that inference demand will eventually justify training spend. The bear case is that we’re building cathedrals before there’s a congregation.
Third, the energy crunch is real. Permitting new generation takes years. AI’s appetite for electricity is already running into hard physical limits in Virginia, Texas, and Ireland.
What practitioners and operators should do
For AI builders and businesses watching this unfold, a few practical moves:
- Lock in compute now if you need it in 2027. Capacity reservations are getting longer and pricier. Waiting is a position.
- Watch the second-tier neoclouds. CoreWeave, Crusoe, Lambda and others are building real differentiation. Pricing pressure from this layer matters.
- Diversify your silicon assumptions. Code written exclusively for one vendor’s stack is a liability. Portability is becoming a hedge.
- Track power, not just chips. The next supply shock won’t be GPU shortages. It’ll be substations.
The forecast worth watching
The Information has been one of the most accurate sources on AI infrastructure deals over the past two years, breaking stories on the Stargate buildout, OpenAI’s chip ambitions, and the financial gymnastics behind hyperscaler commitments. Their reporting has consistently surfaced the gap between announced numbers and what’s actually being built.
The forward question is whether the current spending pace is sustainable through a potential AI revenue plateau. If inference economics improve and enterprise adoption accelerates, today’s capex looks visionary. If model improvements slow and enterprise budgets tighten, the same spend looks reckless.
Either way, the infrastructure layer is now where the real competitive moats are being built. Models can be replicated. Gigawatts and land cannot.
For the full breakdown of deals, players, and financial maneuvering driving this cycle, the original report at The Information is worth reading in full.