OpenAI's $750 Billion Bet: Can Growth Outrun The Losses And Public Backlash?

OpenAI is spending more money than it planned, and virtually nobody wants its datacentres in their backyards.

As losses mount and local resistance to datacentres hardens, OpenAI's escalating infrastructure spend is putting its finances, but also its social licence to build, under growing pressure.

OpenAI said yesterday that it will spend $750 billion on infrastructure through 2030 – an increase of 25% on its earlier estimate of $600 billion, The Wall Street Journal reported.

At the same time OpenAI announced a $20 billion, 3.2GW data centre campus in Georgia known as Project Camellia. OpenAI did not give a timeline for when the GPUs are expected to be operational.

The revised spending projections may cause a sharp intake of breath among some investors. Last November Sam Altman set out an intention to spend $1.4 trillion on computing capacity, which startled the horses to such an extent that OpenAI CFO Sarah Friar was forced to repent publicly and put the figure nearer to $600 billion.

That this figure has escalated once more probably won’t go down brilliantly, particularly given Friar’s reported concerns about whether OpenAI will be able to pay its compute bills if revenue growth does not match expectations and whether the company is in good enough shape for an IPO.

Growing spending commitments and huge losses

The $20 billion it plans to spend in Georgia is a rounding error in OpenAIs overall spending commitments which includes $300 billion with Oracle, $250 billion with Microsoft and $138 billion with AWS.

Quite how OpenAI plans to pay for all the compute that the hyperscalers are building for it remains unclear. Leaked internal documents suggested that OpenAI was on course to lose $14 billion in 2026, but that looks optimistic when viewed in the light of data that was leaked a few weeks ago which show OpenAI losing $38.5 billion last year. $20.92 billion of that was operating loss.

How the company plans to turn the projected $100 billion of profit by 2029 is unclear, especially given that the company is already falling short of revenue targets and is only going to face further competitive pressure from Google’s Gemini, Anthropic and the raft of cheap open source models from China.

Everyone hates datacentres

Another obstacle that OpenAI must navigate is entrenched and growing opposition to AI datacentres in the US. In one of the most politically polarised nations, hating AI datacentres is literally the only subject that almost all voters seem to agree on.

Stargate sites in Michigan, New Mexico and Texas have all been delayed, reduced in capacity or halted completely (for the time being) by public opposition to impacts on air quality, energy costs and water.

It’s part of a much wider pattern of public opposition to datacentres. To be seen as a friend of datacentre development is utterly toxic. Many counties have imposed datacentre moratoriums, a Kansas Republican Senator recently told Semafor that Americans were "on fire" against AI datacentres and the Trump-endorsed frontrunner in the Florida governor primary race Byron Donalds is becoming bogged down in the implications of being funded by a super PAC made up of OpenAI founders and tech venture capital.

In Georgia, OpenAI has clearly tried to get in front of this issue before locals have an opportunity to organise. It has committed to paying “the full cost of the infrastructure and electric-service costs” for the new data centre and has also promised to reduce its power draw by up to 1 gigawatt during periods of high demand on the grid.

OpenAI has also said the datacentre will be designed with closed loop cooling to minimise impact on water supplies.

Whether these pre-emptive concessions are enough to head off the kind of backlash that has slowed Stargate elsewhere remains to be seen. OpenAI may face tougher opponents in the form of debt markets and prospective IPO investors, who must reconcile a $750 billion spending commitment with a company already burning through tens of billions a year and falling short of its own revenue targets.

Ultimately, whether Project Camellia comes in on time and on budget may hinge less on local opposition than on whether OpenAI's business can plausibly deliver growth of the scale needed to make good on its commitments.

This article originally appeared on MES Computing's sister site Computing.