KEY HIGHLIGHTS

  • Bears argue neo-cloud operators such as Nebius Group (Nasdaq: NBIS) are impaired by grid connection wait times of 3-5 years and permitting delays of up to 36 months for on-site gas generation.
  • The counterargument is structural: AI development is a US national security priority, making compute capacity a policy imperative, not merely a commercial one.
  • A scenario in which American AI companies voluntarily slow development due to infrastructure bottlenecks — ceding ground to China — is, for practical purposes, politically impossible.
  • Government intervention to accelerate data centre deployment, whether through permitting reform, grid priority or direct co-ordination, becomes the logical if underpriced backstop.
  • The near-term delays are genuine. The long-term thesis — that the infrastructure will not materialise — requires believing that the US will accept a compute deficit against China. That is a harder case to make.

A familiar critique has been making the rounds among short-sellers targeting neo-cloud operators: the data centres are not coming online fast enough, the grid is full, the permits are stuck, and the whole investment case is built on capacity that does not yet exist. For Nebius Group (Nasdaq: NBIS), which has attracted notable short interest in recent weeks, this has become the central bear argument. It is not without foundation. But it rests on an assumption about political will that, on examination, appears increasingly difficult to sustain.

The delays are real — but so is the pressure

The infrastructure challenges facing neo-cloud operators are not disputed. Grid connection timelines in key US markets have extended to between three and five years in many jurisdictions, reflecting a combination of transmission congestion, interconnection queue backlogs and utility capacity constraints that have accumulated over decades of underinvestment. Permitting for on-site natural gas generation — one of the most practical near-term solutions for data centres seeking to bypass grid constraints — can take anywhere from twelve to thirty-six months, depending on state and local regulatory environments.

For a company trying to bring compute capacity online in line with customer commitments and investor expectations, these are genuine operational headwinds. The bears are correct to flag them. The question is whether they represent a ceiling on the sector's growth or a friction point that will, in time, be resolved.

The national security floor

The bear case implicitly assumes that the bottleneck will persist long enough, and severely enough, to materially impair the economics of neo-cloud operators. That assumption requires looking at the infrastructure problem in isolation from the broader political context in which it sits.

AI development has ceased to be purely a commercial competition. It is now, explicitly and repeatedly, framed by US policymakers as a matter of national security. The competitive dynamic with China over large language models, semiconductor capability and AI-enabled military applications has elevated compute infrastructure to the status of strategic asset. This framing has consequences.

In a world where American AI companies accept a compute constraint and voluntarily decelerate their development timelines, the beneficiary is China. That is not a scenario that any serious policymaker in Washington — of either party — is prepared to countenance. The political calculus is straightforward: a compute shortage that hands Beijing a meaningful window of advantage is, by definition, a national security failure. It will not be allowed to persist.

Government as backstop

The logical implication is that if the commercial and regulatory systems cannot deliver sufficient data centre capacity fast enough, the US government will intervene to make them do so. The mechanisms are well-established: permitting reform, executive orders expediting grid connections, preferential treatment for AI infrastructure under existing national security frameworks, direct co-ordination between federal agencies and the major hyperscalers and their suppliers.

None of this is speculative. The Biden administration's executive orders on AI, the CHIPS and Science Act's implicit acknowledgment of compute as a strategic resource, and the current administration's continued emphasis on AI leadership all point in the same direction. The political infrastructure for government-backed acceleration already exists. What it requires is a sufficient trigger — and a genuine compute crisis, with China visibly closing the gap, would provide one.

What the bears are really arguing

Stripped of its technical detail, the bear case against Nebius Group (Nasdaq: NBIS) and comparable operators reduces to a timing argument: the delays will be severe enough, and last long enough, to damage the business before the cavalry arrives. That is a legitimate risk to model. Near-term earnings estimates can absolutely be impaired by permitting backlogs and grid constraints, and investors who bought on aggressive capacity ramp assumptions may face disappointment.

But the structural thesis — that neo-cloud operators are impaired in a durable, unresolvable way — requires believing that the United States will accept a sustained compute deficit against China and do nothing meaningful about it. That is the bet the bears are implicitly making. It is, to put it plainly, a bet against American industrial policy at a moment when industrial policy has rarely been more aggressively deployed.

The question of timing, not direction

The more honest framing of the risk is not whether the data centres will come online, but when. A project delayed by eighteen months is a cash flow problem. A project that never materialises is a thesis-breaker. The bear case requires the latter; the evidence — political, regulatory and commercial — points more convincingly toward the former.

For investors in Nebius Group (Nasdaq: NBIS) and its peers, the appropriate question is therefore not whether to own the sector but at what price the delays are adequately discounted. If the market is pricing these names as though the infrastructure will never arrive, that may represent an opportunity. If it is pricing them as though the permits will clear on schedule and the grid will accommodate all comers without friction, there is a case for caution.

The bears have identified a genuine operational risk. They have, however, confused a friction point with a structural impairment. In a sector where the US government has made clear that failure is not an option, those are very different things.

The data centres will come online. The grid will be expanded, the permits will clear, and the capacity will materialise — because the alternative, in the eyes of Washington, is handing China a window it would not squander. The timing is uncertain. The direction, given what is at stake, is not.