If

On the fourteenth of October, the head of the world's biggest money manager made a warning. Larry Fink, of BlackRock: "But if we have a moratorium of some sort in this country, we will be in a recession."

The words are as spoken at about 10:33 of CNBC's full interview of 14 October 2026 (CNBC's own video); CNBC's article of the same day reports them. The warning is conditional, an if; this piece tests its size, not his honesty.

There is no national moratorium. There's a Senate bill, introduced in March. It isn't law. A county in Florida has paused new applications for a year. New York has paused state environmental permits for the biggest sites. So it's an if. How big is the if?

The Census Bureau counts what America spends building data centres. In August, it was running at eighty-five billion dollars a year. Two years earlier, thirty-seven. More than double, in two years.

The Census Bureau's Construction Spending release of 1 October 2026 (August 2026): private data-centre construction at a seasonally adjusted annual rate of $84,950 million in August 2026 (preliminary), $79,007 million in July (revised), and $36,668 million in August 2024. 84,950 against 36,668 is 2.32 times (our arithmetic). The Census count is the buildings and their fit-out; the racks and servers inside are not in it.

And yet: that's about five cents of every dollar spent on private construction. Factories still get twice as much. And against the whole economy, it's about a quarter of one percent.

In August 2026, data centres were 5.13 per cent of private construction ($84,950 million of $1,655,280 million); manufacturing plants $168,209 million, about twice as much (1.98, our arithmetic). Against nominal GDP of $32,563 billion in 2026Q2 (FRED), $84,950 million a year is 0.26 per cent: a spending rate set against the size of the economy, not a contribution to its growth.

The government's growth figures say the same. For four quarters in a row, building of this kind, offices, plants, data centres all together, took away from growth. What added to it was spending on computers and software, across the whole economy.

The BEA's contributions to the percent change in real GDP, in percentage points: nonresidential structures, all of them, not data centres alone, 0.03 in 2025Q1, then -0.17, -0.25, -0.31 and -0.29 through 2026Q1, and 0.00 in 2026Q2; information processing equipment 0.87, 0.37, 0.30, 0.67, 0.79, 0.19; intellectual property products 0.43, 0.84, 0.32, 0.24, 0.83, 0.51 (FRED, read at source; the 2026Q3 advance estimate is due 29 October).

And that's the strongest case for Fink. The bank ING says AI accounts for a third of growth this year. But it says the buildings themselves are "relatively small". The computers inside are the main cost. And, our point: no building means nowhere to put the computers.

ING, 20 August 2026: "We think AI technology and data centres account for a third of economic growth in 2026"; and on the structures, "given it is the computer equipment inside that is the major cost, rather than the physical structure, the contribution from this component will be relatively small".

One more thing, from Fink's own words in BlackRock's third-quarter earnings release. He says it's winning mandates in "hyperscaler financings". And its AI Infrastructure Partnership "made its first deployment this quarter."

So: a pause on building would hit a small slice of the economy. A pause on the computers would hit a big share of growth. Fink's warning is biggest if the first becomes the second.

The dates to watch: the BEA's advance estimate of 2026Q3 growth on 29 October 2026; the Census Bureau's September construction spending on 2 November 2026.