This article digs into the so-called AI and data-center boom. It argues that much of the capacity vendors and media tout is overstated or stuck in planning limbo.
Drawing on Wood Mackenzie data and industry reports, it shows a big chunk of announced capacity is still just speculative. “Committed” power often just means wires-only agreements, and grid, permitting, and planning headaches keep near-term growth slow.
The piece also sets optimistic vendor projections against slower real-world progress. There’s a real risk that GPU supply, data-center readiness, and grid capability just won’t line up.
Measured reality behind the AI data-center surge
Despite the headlines, the actual footprint of AI-driven data-center expansion is narrower than it seems. The numbers get tossed around a lot, but what do they really mean?
Definitions matter a lot when you’re trying to gauge real growth.
Active development vs speculative pipeline
Only a minority of announced capacity is actually moving toward construction and operation. About 33% of the roughly 241 GW U.S. pipeline is under active development.
The rest sits as speculative permits, land deals, or plans that depend on power sources nobody’s built yet.
- Active development makes up about one-third of the pipeline.
- Most of it is still in planning or speculation, not building.
Wires-only commitments and grid strain
A lot of commitments are “wires-only”, meaning data centers have to lock down their own generation. Roughly 58% of committed power falls into this category.
This can ramp up grid strain, especially in the PJM region where commitments already outpace accredited generation capacity.
- Wires-only deals mean more reliance on on-site or nearby generation.
- Grid stress climbs when demand commitments outstrip verified supply.
Definitions and data quality concerns
Industry reports toss around terms like absorption, delivered, and committed in inconsistent ways. That muddies the picture of what’s actually online and earning revenue.
This definitional fuzziness lets some players paint a rosier picture than the infrastructure supports.
Absorption, delivered, committed — why definitions matter
Look at it with conservative metrics: North American data-center absorption in 2025 was about 3 GW of IT load. That means around $90 billion in GPU and hardware buys—a lot lower than what vendor hype suggests.
- Real IT load online is much smaller than headlines make it sound.
- Hardware spending matches measured IT demand better than forecasted expansions.
Forecasts vs reality: NVIDIA and the hype
The pace of actual builds and grid readiness just doesn’t match the big projections from major vendors. The story of explosive GPU demand runs up against slower infrastructure rollout and power limits in the data-center world.
NVIDIA projections vs build pace
NVIDIA’s aggressive forecasts—like up to $1 trillion in GPU sales across 2025–2027—seem out of step with the slower pace of data-center construction and power delivery. This gap raises questions about timing and whether supply chains can really keep up.
Case study: Stargate Abilene and related announcements
Big projects often get called open or delivered in press releases, but progress is usually incremental and lags promised timelines by years. Stargate Abilene is a good example—headlines move faster than hardware delivery, permitting, or transmission readiness.
What the headlines show vs what is delivered
Announcements grab attention and investment, but the real-world pace is slower. Those repeated “open” or “delivered” tags often hide ongoing expansion work, permitting snags, and the scramble for reliable energy sources to keep big operations running.
Infrastructure bottlenecks and policy implications
Adding multi-gigawatt, reliable power isn’t a quick fix—it’s blocked by systemic bottlenecks that take years, not months, to solve. Transmission constraints, permitting delays, and under-resourced utility planning tools slow things down more than most folks realize.
Systemic constraints and the risk of misalignment
Analysts and developers often conflate phases of development. They do this to push a rapid-scale narrative that benefits vendors and financiers.
The AI industry faces a real risk here. Without coordination between policy, grid planning, and data-center design, we might see a stubborn mismatch between GPU supply, data-center readiness, and electricity availability.
Here is the source article for this story: The AI Industry Is Lying To You