UK sovereign AI infrastructure
Thirty Days of Building
8/9/2026 · AI-generated, human-reviewed
There is a particular kind of frustration that comes with building infrastructure. The work that matters most is the work nobody can see. Planning applications. Connection offers. Gas nominations. Switchgear ratings. Thermal efficiency curves. None of it photographs well and none of it makes a headline.
So every so often it is worth stopping and writing down what actually got done.
Here is our last thirty days.
We secured a firm grid connection offer
We now hold a formal connection offer for 25 MVA at 33 kV into our Dundee campus, from a named grid supply point, with an engineered route.
That sentence takes two seconds to read and it took months to earn. Grid capacity in the UK is the single hardest constraint on AI infrastructure right now. Queues are long, offers are conditional, and plenty of announced projects quietly never get one. We have ours in writing.
Our gas is live, and we tripled the roadmap
Our sovereign gas service is live at the site boundary at 25 MWth. This month we locked the staged plan that takes it to 50 MWth at month six and 75 MWth at month twelve.
Gas primary generation is a deliberate choice, not a compromise. It means we are not waiting in a queue to serve our first customers, and it means our capacity ramp is something we control rather than something we are granted.
We finished the cooling engineering
We completed the free cooling design and a full waste heat recovery study, landing on a design PUE of 1.08.
We also studied a heat driven cooling system and decided against it, because the numbers did not justify it against the free cooling scheme we already have. Publishing the things you decided not to build is a reasonable test of whether anyone is actually doing the engineering.
Our model is public, and the next one is training
MOTHER CORE V3 is published and openly available on Hugging Face. Anyone can download the weights and test them.
We say this plainly because it matters: MOTHER CORE is trained from scratch. It is not a fine tune of somebody else's model with a British name on the front. That is a slower, harder and more expensive road, and it is the only road that produces something genuinely sovereign.
V4 is in training now on a substantially expanded corpus.
We released MOTHER EXO v5
MOTHER EXO is our embodied model. One shared vision, reasoning and memory core feeding specialised output heads for aerial systems, ground vehicles and humanoid units.
The scenario we designed it around is disaster response. Drones searching from the air, autonomous ground vehicles carrying supplies and crew, and humanoid units doing the physical work that people cannot safely do. One coherent intelligence coordinating across all of it, with a human in the loop at every decision that matters.
v5 went live this month. You can read the release at motherai.uk/exo.
We added partners
We executed a new strategic partnership and signed an engagement covering a two thousand GPU deployment for the campus. We also brought in written statements from partners across flooring, generation, hardware, white space fit out and professional services, so that anyone assessing this project can hear it from them rather than from us.
That last point is deliberate. The most useful thing we can do for anyone evaluating Media Stream AI is put them in front of the people we work with.
What comes next
The next thirty days have one objective: first tenants in the hall.
Everything above is input. The output is a customer taking delivery of powered, cooled, secured white space in Dundee and putting compute into it. That is the only measure that counts, and it is what we are pointed at.
We will keep writing these. Some months the list will be shorter. We will publish it anyway.
Built in the UK. Owned in the UK. Run in the UK.