Nokia's CEO said this week that the construction of AI data centers could move twice as fast if the industry weren't held back by supply chain bottlenecks. The remark puts a number on a frustration that's been building across the tech sector for months: the physical buildout of AI infrastructure isn't keeping pace with demand, and the holdup isn't capital or ambition — it's parts and power.
The CEO's point was blunt. The constraint isn't how fast you can pour concrete or install racks. It's whether you can get the chips, the networking gear, and enough electricity to run the whole thing. Without those shortages, he suggested, the same projects that now take years could be finished in half the time. Nokia, which supplies networking equipment used in data centers, sits on the front line of that bottleneck — and has a clear incentive to see it cleared.
What's actually slowing things down
AI data centers are not like the server farms of a decade ago. They pack racks with high-end accelerators that draw far more power and throw off far more heat. That changes the bill of materials: more advanced cooling systems, beefier power distribution, denser fiber connections. Every one of those components has its own supply chain, and several of them are stretched.
Chip production is the most visible pinch point. Advanced processors and the specialized memory they need are made by a handful of manufacturers, and their capacity is booked out well in advance. Networking gear — the switches, optics, and cabling that tie thousands of processors together — is another. Nokia plays in that space, which is likely why the CEO framed the problem as an industry-wide production issue rather than something a single company can fix.
Energy is the quieter constraint. Data centers need massive amounts of electricity, and in many regions the grid simply doesn't have spare capacity sitting idle. Building a new facility is one thing. Getting a utility to guarantee the power it needs is another. In some markets, the wait for grid connections now rivals the wait for the chips themselves.
Why the two-year timeline matters
The CEO's "twice as fast" estimate isn't a forecast — it's a way of sizing the opportunity cost. If a project that would normally take four years could be done in two, that's two extra years of compute capacity, two extra years of revenue, and two extra years of progress on whatever the models are being trained to do. Multiply that across the dozens of data centers under construction worldwide, and the gap between constrained and unconstrained buildout becomes enormous.
It also reframes the problem. For much of the past year, the public debate over AI infrastructure has focused on spending — who's buying how many chips, how much capital is being poured into new sites. The Nokia CEO's comment shifts attention to the supply side of the equation, where money alone doesn't solve the problem. You can't buy a chip that hasn't been fabbed, and you can't run a data center on a grid connection that doesn't exist yet.
The innovation pressure points
Two areas get the most attention when supply constraints come up: chip production and energy. On chips, the push is toward more advanced manufacturing capacity and, increasingly, specialized designs that squeeze more performance out of the same silicon. On energy, the conversation has moved to efficiency — getting more compute per watt — and to alternative power arrangements that reduce reliance on strained grids.
Nokia's own business touches the middle of that stack. Data centers don't function as isolated chips; they function as networks of chips, and the networking layer determines how much of the raw compute you can actually use. If the interconnect is slow or unreliable, the most expensive processor in the rack sits idle. That's why a networking vendor's CEO has standing to talk about buildout speed — his products are part of what determines it.
None of this is a problem one company can solve alone. The constraints run across fabrication, packaging, power generation, grid infrastructure, and construction. But the CEO's framing — twice as fast, if not for supply — gives the industry a concrete target to argue about, and a reason to treat supply chain investment as urgent rather than routine.
No quick fix on the horizon
New fabs take years to build and ramp. Grid upgrades take even longer. Networking capacity can scale faster, but it still depends on components that are themselves in short supply. So the timeline the CEO described isn't something that flips with a single decision or a single quarter of strong earnings.
What it does do is sharpen the question for everyone building or financing AI infrastructure: how much of the current delay is unavoidable, and how much is a supply chain that hasn't caught up to a demand curve nobody predicted this fast? Nokia's CEO has put his answer at 50 percent. The next earnings cycle, and the next round of data center announcements, will start to show whether the rest of the industry agrees.
