The scramble to bring AI data centers online is quietly flipping the balance of power between the companies building them and the companies feeding them, according to a new report from The Information. As detailed in The Information, the sheer desperation to get compute running is handing new bargaining power to whoever controls the scarce pieces: electricity, land, transformers, cooling, and skilled crews. This is significant because it marks a shift from a buyer’s market to a seller’s market for the unglamorous parts of the AI stack.
What stands out here is where the leverage is moving. For two years the story was about GPUs and who could get allocation from Nvidia. Now the choke point has moved down the chain to power and physical infrastructure, and the owners of those assets know it.
What’s actually changing
When you’re racing to stand up a gigawatt of capacity before a competitor does, you stop negotiating hard on price. Speed becomes the only thing that matters. That single change reshapes every contract in the pipeline.
Here’s who gains the upper hand:
- Power providers and utilities. Grid interconnection is the new bottleneck. Utilities can dictate timelines, pricing, and terms because a half-built data center with no electricity is worthless.
- Equipment suppliers. Transformers, backup generators, and switchgear now carry lead times measured in years. Vendors are booking orders far in advance and holding firm on price.
- Landowners near power. Sites with existing substation access or a path to it command a premium that would have looked absurd in 2022.
- Specialized contractors. The crews who can actually build and cool these facilities are scarce, and scarcity sets the rate.
The companies losing leverage are the ones you’d least expect to see squeezed: well-funded hyperscalers and AI labs with more capital than options. Money isn’t the constraint anymore. Time is.
Why it matters now
This is happening at the exact moment capital spending on AI infrastructure is peaking. Microsoft, Google, Amazon, Meta, and OpenAI-linked buildouts have pushed combined data center capex into the hundreds of billions. When everyone wants the same transformer in the same quarter, the person holding the transformer wins.
The knock-on effects are already visible. Power deals are getting creative, from nuclear restarts to behind-the-meter gas. Construction timelines are being quoted as competitive advantages. And smaller AI companies without deep pockets or long-term supply relationships risk getting locked out entirely, because they can’t offer the volume or the certainty that suppliers now demand.
What to watch over the next 1-3 years
The leverage shift won’t reverse quickly. Building new power generation and manufacturing more heavy electrical gear takes years, so the supply crunch is structural, not a blip.
Expect three things:
- Vertical integration. The biggest players will keep buying their way upstream, locking in power plants, land banks, and equipment orders to remove the bottleneck from someone else’s control.
- Longer, tighter contracts. Suppliers will push for multi-year commitments with less flexibility. If you want the capacity, you sign for the duration.
- A widening gap. Companies that secured power and hardware early will pull ahead. Latecomers will pay more and wait longer, or partner up to survive.
The practical takeaway
If you’re building or buying AI infrastructure, stop treating power and hardware as commodities you can source on demand. Lock supply early, even if it feels premature. Sign the long contract. Treat grid access as a strategic asset, not a line item.
And if you’re a business simply renting compute rather than building it, understand that the pricing you see downstream reflects this scramble. The cost of your AI workloads is tied to a transformer shortage you’ll never see.
The AI race was supposed to be won on models and talent. It’s increasingly being won on who can plug in first. Readers can find the full breakdown at The Information.