Venture capitalist and White House AI official David Sacks says the American AI data center boom could run out of road within two years, warning that permitting uncertainty and state-level moratoriums could leave the construction pipeline "closed or significantly reduced" by 2028.

Sacks made the prediction on the latest episode of the All-In podcast, released November 7, in a discussion covering the midterm elections, the ongoing AI stock selloff, and what the hosts called a "data center freeze" sweeping through statehouses. For a figure who serves as the administration's AI and crypto czar, the comments amount to a public acknowledgment that the biggest near-term threat to the AI buildout is not model quality or chip supply — it is local politics. Our readers can track the rest of this week's breaking AI news as the infrastructure story develops.

'Nobody Can Make a $50 Billion Investment in Conditions of Uncertainty'

"Nobody can make a $50 billion investment in conditions of uncertainty," Sacks said on the podcast. "I think by 2028 the pipeline will be closed or significantly reduced."

His argument rests on permitting risk rather than financing. The greatest danger to a data center project, Sacks argued, is breaking ground and then getting "enjoined" partway through construction — halted by a lawsuit, a worker strike, or a community group that flips the politics after billions have already been committed. He added that governors in the largest data center states are "basically indicating that they favor a moratorium" until certain conditions are met.

The warning lands in a political environment that has turned sharply against AI infrastructure. In this month's midterms, ballot measures restricting data center development won across at least 13 states, and community opposition has already put tens of billions of dollars of projects on hold nationwide.

Why the Buildings Matter More Than the Chips

The supply chain's exposure starts with a simple dependency: every AI chip, network switch, and cooling system needs a finished building and a grid connection first. Broadcom's chief executive made the same point on the company's September earnings call: "Before you get the chips out there, you have to get the data center shell in place and ready for production."

If construction stalls, the demand assumptions behind the sector's most aggressive growth forecasts stall with them. Consider what the suppliers are currently guiding to, according to their most recent earnings reports:

  • Nvidia reported data center revenue of $89.02 billion, up 117%, and guided its next quarter to roughly $108 billion, with $279 billion in disclosed supply obligations in its latest SEC filing.
  • Broadcom is targeting $230 billion in AI semiconductor revenue by fiscal 2028 — squarely inside Sacks's risk window.
  • Arista Networks grew revenue 58.5% and guided its fourth quarter to $3.9–4.0 billion, concentrated among a small number of large customers.
  • Vertiv, which sells the power and cooling equipment inside those buildings, reported organic orders up more than 100%, with customers typically ordering 12 to 18 months ahead.

Those numbers assume the pipeline keeps flowing. Sacks's forecast assumes it does not. Both cannot be true at once.

The Market Is Already Pricing the Split

Investors spent the past month drawing exactly the line Sacks is describing. Over the month ending November 6, Nvidia fell 21.4%, Broadcom dropped 22%, Arista slid 20.86%, and Vertiv lost 25.81% as the AI selloff deepened.

Data center landlords, by contrast, barely moved: Equinix slipped just 1.85% and Digital Realty 3.47% over the same period. The logic is straightforward — if speculative projects get canceled, the companies that already own permitted, powered capacity become more valuable, not less.

Equinix's chief executive made that case explicitly: "If there are a number of projects canceled, then it only makes what is already scarce capacity much more of a scarce resource." She noted that across the company's 3 gigawatts of planned capacity, "the land, power and permitting are largely secured."

Digital Realty has 1.5 gigawatts under construction, 63% of it pre-leased, and a record $2.5 billion backlog that gives it revenue visibility through 2027 and into 2028 — a hedge against exactly the freeze Sacks fears.

The Counterargument: Moratoriums Are Not Law Yet

Sacks's forecast is a prediction, not a settled fact. Proposed moratoriums can be narrowed, delayed, or dropped entirely, and no one yet knows how state policy will land after the midterm shakeout. Vertiv's own executives have continued to describe long-cycle demand, and hyperscalers with committed capital have historically found ways to build through local opposition.

There is also a federal dimension. The White House has pushed to accelerate permitting for large infrastructure projects, and an administration that loses patience with state moratoriums could attempt to pre-empt them — a legally fraught path that would itself become a political fight.

What It Means for the AI Buildout

The deepest significance of Sacks's comments is who is saying them. This is not a bearish analyst shorting the sector or an activist opposing data centers — it is the administration's own AI czar conceding that the political license to build may expire before the financing does.

If governors in data center-heavy states follow through on moratorium conditions, the industry's 2028 capacity targets could slip by years, forcing AI companies to ration compute, raise prices, or move capacity overseas. If instead the permitting fights get resolved project by project, the pipeline may bend without breaking — and the suppliers guiding to triple-digit growth will have dodged the scenario their own White House adviser just described.

Stay Ahead of AI

From data centers to model launches, the infrastructure story is moving week to week. Follow breaking AI news with coverage in 35 languages.

Read more AI news →