The pattern running through three recent Big Tech stories is not really about AI's promise or its perils. It is about who pays for both, and who decides. OpenAI is projecting a $278 billion cash burn through 2030, Nvidia's chief executive is declaring a 0% chance that AI destroys the world by that same year, and the UK government is weighing whether to tear up a £330 million NHS contract with Palantir. In each case, the industry is asking someone else to absorb the cost and the risk while insisting that the matter is settled.
The Math Nobody Wants to Underwrite
As Tom's Hardware reported, OpenAI reportedly expects to spend $278 billion more than it generates between 2026 and 2030 on compute and adjacent infrastructure, with a total compute tab of $856 billion against a revenue surge that is tenfold but far smaller than the outlay. That is not a rounding error. It is a number that exceeds the national budgets of Indonesia and Norway. A private company is proposing to spend, in the space of five years, more than most governments collect in a year, and it is doing so on the premise that the returns will arrive later.
For US technology companies, this is the central fact of the moment. The money has to come from somewhere. It can come from investors who are willing to wait, from customers who are willing to pay more, or from public balance sheets that are asked to treat AI capacity as strategic infrastructure. The story is logged as a Big Tech story because it is one, but its resolution will be decided well outside the technology press. When a single company's deficit is larger than the revenue of entire economies, capital markets, utilities, power grids and, ultimately, taxpayers are in the frame.
The second-order effects for the US market are already visible in the logic of the spending. Compute capacity is not a software product that scales at marginal cost. It is land, power, water, chips and construction. Those are physical constraints with physical bills. The company carrying the largest projected deficit is also the one setting expectations for the rest of the sector. If the projections hold, the industry's capital discipline is being set by an entity that has publicly accepted a deficit larger than most national budgets. If they do not hold, the correction will not be confined to one firm.
The 0% Doctrine
Into this comes Jensen Huang, chief executive of Nvidia, telling the world, as Tom's Hardware reported, that there is a 0% chance AI destroys the world by 2030, that the industry should go as fast as it can irrespective of anyone else, and that new regulations should be rejected. He dismisses Anthropic's warnings as unsubstantiated and places his confidence in safety mechanisms.
The statement is worth taking seriously not because of the probability claim, which is unfalsifiable in the relevant window, but because of the posture. It is a declaration that the pace of deployment is not a matter for public negotiation. That is a claim about governance, not about technology. And it sits awkwardly beside the first story. If the buildout requires public infrastructure, public permitting, public energy and, in some scenarios, public money, then the public has a legitimate claim on the terms. A doctrine of maximum speed with no external constraint is a doctrine that says the bill can be sent, but the questions cannot.
For US consumers, the practical consequences are not abstract. Energy prices, grid reliability, water use and local land decisions are the visible edges of data center expansion. When an industry leader says speed should not be conditioned on anyone else's concerns, that is a statement about how those edges will be managed. It is a direct answer to the people who live near the infrastructure and pay for the grid.
