While others worry AI spending has peaked, the smart money is writing $4 billion checks on the thesis that we're just getting started.
The Summary
- Bill Ackman's Pershing Square built $4 billion in Microsoft and Meta positions, betting on $700 billion in total hyperscaler AI infrastructure spending across the Big Four tech companies
- Meta raised 2026 capex guidance to $135 billion, signaling acceleration rather than retreat in AI infrastructure buildout
- Meta is building two natural gas plants in Ohio under fast-tracked laws that bypass public hearings, exposing the energy infrastructure gap that AI datacenters have created
- The physical infrastructure layer of AI is now driving energy policy, regulatory shortcuts, and multi-billion dollar conviction bets from hedge fund managers
The Signal
The narrative that AI spending might slow just hit a $4 billion reality check. Ackman isn't known for momentum plays or hype chasing. When Pershing Square goes long $4 billion on Microsoft and Meta, it's a thesis about infrastructure economics, not a bet on the next chatbot feature. His conviction is tied to a projected $700 billion in aggregate AI capex from the hyperscalers. That number matters because it represents physical infrastructure: datacenters, chips, cooling systems, and power plants.
Meta's guidance bump to $135 billion for 2026 confirms Ackman read the room correctly. This isn't maintenance capex or incremental expansion. This is the kind of spending that reshapes supply chains. Nvidia can't make enough H100s. TSMC is building new fabs. Power utilities are getting calls from Big Tech asking about gigawatt-scale commitments on timelines that used to take a decade.
"Meta raised 2026 capex guidance to $135 billion, signaling acceleration rather than retreat in AI infrastructure buildout."
The Ohio gas plants tell the rest of the story. Meta is constructing two natural gas facilities under Ohio Senate Bill 2, legislation designed to fast-track energy projects by sidestepping the usual public comment process. This isn't about renewable targets or ESG commitments. It's about physics. You cannot run inference at scale on wind power alone when the wind stops. You need baseload, and you need it now, so you build gas plants and you find a state willing to cut the permitting timeline.
The timing is everything. These moves are happening as questions swirl about AI profitability, model saturation, and whether enterprises will actually pay for AI at the scale required to justify current valuations. Ackman and Zuckerberg are answering with their balance sheets: the infrastructure layer gets built first, and whoever controls it controls the next decade of compute.
What's getting built here:
- Physical datacenter capacity that takes 18-36 months to bring online
- Energy infrastructure that bypasses normal regulatory timelines
- Positioning for the agent economy, where inference costs compound as agents call agents
This is Web4 infrastructure being laid in real time. The agent economy needs always-on compute. Tokenized assets need always-on settlement rails. Both need energy. The companies writing nine-figure checks today are the ones who will own the rails tomorrow.
The Implication
If you're building in AI, the infrastructure question just became existential. The hyperscalers are locking in power, chips, and datacenter capacity at a pace that will price out smaller players. If you're dependent on API calls to someone else's models, you're renting from landlords who are currently building moats made of concrete and natural gas turbines.
For investors, watch where the next Ohio-style energy deals get cut. States hungry for jobs and tax revenue will trade permitting speed for datacenter commitments. Those locations become the physical hubs of the agent economy. Geography matters again.