AI data centers are surging. Here's what Anthropic and Microsoft just set in motion for real estate and construction
Two fresh signals just hit the market. Anthropic announced a $50 billion buildout for new computing infrastructure with data centers planned in Texas and New York. Microsoft confirmed a two-story Atlanta build, linked over high-speed network to its Wisconsin campus, forming a supercomputer built on hundreds of thousands of Nvidia chips.
Translation for developers, GCs, and CRE: power-hungry projects, multi-site networks, and a long runway of site work, utilities, and specialized trades. The spending continues despite talk of an AI bubble, environmental scrutiny, and political pressure over electricity costs.
The deals at a glance
- Anthropic: $50B in compute infrastructure with London-based Fluidstack. New builds planned in Texas and New York. About 800 permanent jobs and 2,400 construction jobs expected.
- Microsoft: "Fairwater 2" in Atlanta connected to the original Fairwater in Wisconsin by a high-speed network, creating a massive multi-site supercomputer. Capacity aimed at Microsoft AI, OpenAI, and other developers.
- Leasing velocity: TD Cowen reports 7.4 GW of U.S. data center capacity leased in Q3 alone-more than all of last year. Oracle led capacity grabs (largely for OpenAI), followed by Google, then Fluidstack, ahead of Meta, Amazon, CoreWeave, and Microsoft.
- Capex and obligations: Microsoft spent nearly $35B in the July-September quarter, almost half on chips. OpenAI has more than $1T in infrastructure obligations. Anthropic is scaling with Amazon and Google partnerships.
Where the demand is landing
Texas and New York remain hot, with prior disclosures pointing to sites including the Lake Ontario shore. Atlanta and the Milwaukee region are anchoring a cross-state Microsoft network. Expect more multi-campus designs, with fiber backbones connecting regions for resiliency and latency targets.
Site selection is getting filtered by three constraints: immediate or scheduled power, fiber proximity, and water policy. Parcels with substation adjacency, upgrade paths, and community support will move first.
What this means for developers and GCs
- Utilities first: Interconnection queues and substation timelines are now the critical path. Early utility engagement and phased energization plans will win schedules.
- MEP depth: Direct-to-chip liquid cooling, immersion, dense electrical rooms, and high-capacity switchgear require specialized subs. Build bench strength now.
- Permitting stack: Water withdrawal, discharge, noise (gensets), and traffic are flashpoints. Bring environmental and community teams in early to avoid delays.
- Modular delivery: Expect multi-phase campuses and repeatable blocks. Prefab skids, modular power rooms, and repeatable cooling units cut months off delivery.
- Labor planning: Anthropic projects 2,400 construction jobs across its builds-plan for peak labor availability, retention, and safety on fast-track schedules.
Power and permitting realities
Leasing 7.4 GW in a single quarter tells you everything: grid capacity is the bottleneck. Substations, transformers, and high-voltage gear have lead times that can outlast your shell schedule. Water strategy (or waterless cooling) is no longer optional near tight basins.
Local politics are heating up around electricity bills. Expect more scrutiny on energy sourcing, heat reuse, and emissions accounting, especially in towns that haven't seen facilities of this scale.
For a big-picture view on energy demand trends, see the International Energy Agency's analysis of data centers and networks here.
Supply chain watch
- Electrical gear: Large power transformers, medium-voltage switchgear, UPS, and busway remain tight. Lock in allocations early and design around realistic lead times.
- Cooling: Liquid systems, CDU skids, and heat exchangers are surging. Standardize where possible to speed approvals and commissioning.
- Fiber: Multi-terabit cross-connects between campuses (like Atlanta-Wisconsin) require early carrier coordination and route resiliency planning.
How to position for 2025-2027
- Secure power-forward parcels with documented upgrade paths and utility MOUs.
- Stand up a dedicated data center precon team: entitlements, utility, environmental, and community relations.
- Build a repeatable kit-of-parts for shells, power rooms, and cooling blocks to scale across regions.
- Develop water-neutral options: heat reuse, adiabatic alternatives, onsite treatment, and non-potable sourcing.
- Explore onsite or adjacent energy: BESS, demand response, and PPAs to de-risk curtailment and rate volatility.
- Invest in HPC-ready MEP talent pipelines and safety programs for dense, high-heat environments.
Financing signals to watch
Concern over circular deals and an AI bubble is real, but the builds keep moving. Recent chatter about public support for chips met a firm "no bailout" stance from the White House. OpenAI reiterated it doesn't want government guarantees and disclosed commitments around $1.4 trillion over the next eight years-ambitious, but a reminder that revenue must keep pace.
For you, that means contract discipline, phased scopes, and clear change-order governance. Focus on clients with secured power, credible financing, and a roadmap for multi-phase scaling.
Bottom line
Anthropic's $50B push and Microsoft's cross-state supercomputer are more than headlines. They're a build signal for sites with power, fiber, and a fast path through permits. If you can deliver repeatable, utility-coordinated campuses, the next two to three years will be busy.
If you're upskilling teams on AI use in precon, scheduling, estimating, or ops, explore curated training by job here.
Your membership also unlocks:
