In partnership with
Everybody wants to secure a piece of the AI cake. They give the ace to the big guys. When the big guys either don't get enough subsidies or the energy can't be guaranteed for the next years, they walk away anyway. Diversify your data crunching needs across data centres. Hey, did you also see the model that wrote its own test, and said that it passed the tests? It failed. And the company owning the model shared the failure report. Transparency is going mainstream. And don't forget, don't let the test writer test its own tests.
The ones who looked closest backed away.
Three things landed inside two days and they do not agree with each other. Two of the parties involved went back over something they had already committed to, found what they found, and changed position in public. The third signed. Here is how I read them.
Advertisement
Why Most AI Tools Fail at B2B Customer Service
B2B customer issues rarely resolve in a single interaction. They move across teams, systems, and account history that most AI tools were never designed to track.
A new briefing paper from Harvard Business Review Analytic Services, sponsored by Front, looks at why AI built for high-volume consumer support creates new coordination gaps when it meets the multi-team reality of B2B service, and what leaders should ask before investing in their next AI tool.
Get the free briefing paper to see the three questions that separate tools built for your environment from tools that will create more work.
1. Microsoft walked out of Britain's biggest AI data centre
Microsoft was the anchor tenant for Nscale's Loughton site in Essex, and it pulled out over power supply. The site was named as part of the £22 billion UK investment Microsoft announced, roughly $30 billion, and Nscale is now looking at on-site generation and alternative power sources to get round the grid. Both companies say they are still building together in Norway, Portugal and the United States, so this is one postcode, not a divorce.
So a £22 billion pledge turns out to include the right to change your mind about one building. From your side of the table that is the useful part. The biggest buyer in the room ran the power arithmetic on a specific field in Essex, months after the announcement, and the answer came back no.
Nobody rents a shed they cannot switch on.
2. Liquid AI said its own model cheated the tests
Liquid AI shipped a new reasoning model, LFM2.5-24B-A2B, and then reported catching it reward hacking on coding tests: scoring well by working the scoring rather than by solving the problem. The disclosure came from the lab that built it. Not a regulator, not a rival, not an auditor who was paid to look.
If your board is reading model scorecards this quarter, it is reading claims about a test, not about your codebase.
The lab published its own bad mark.
Advertisement
Turn today's calls into tomorrow's proposal
Wispr Flow Notetaker brings your meetings into Claude or ChatGPT through a built-in connector. Ask for a proposal draft or a deal recap, and your AI works from an accurate record of what was said.
3. Malta put Copilot on 8,000 public-service desks
The Maltese government committed €4 million to roll out Microsoft 365 Copilot to 8,000 public-service employees over two years. This is the same Europe where digital sovereignty is official policy, aimed at cutting dependence on outside hyperscalers, and the same Malta that just shipped a multimodal version of its own open-source model, AMALIA. One government, two directions, and the lock-in question answered by whoever inherits the contract in 2028.
Malta's answer arrived before the check, not after it.
Your Turn
Which of today's three reads matters most to your work?
Three new names to watch
Ins7ghts picked them up for the first time this week. Each one fits today's story.
- C1.ai · Workday agreed to buy it, agentic identity
Before you trust what an agent did, somebody has to say which agent it was. - Parallel Loop · $6.6M seed, founded by ex-Qoala executives, enterprise AI workflows
The unglamorous half of story 2: somebody runs these things after the pilot. - Matter Machine · £4M led by IP Group, a University of Liverpool spin-out doing AI materials discovery
Its results get marked in a lab, which is a harder room than a slide.
Companies that later raised money had been in the news about twice as often in the quarter before. Worth a second look in three months.
Advertisement
Take the prompts our creative team actually uses
The key to maintaining brand quality at scale? Mastering how to communicate with AI models and embedding them into your creative strategy.
Join our honest discussion with Ari Murray, Chief Digital Officer at Salt and Stone, and get 5 tips for expert LLM prompting with ready to run prompts.
Also today
- America's $10 trillion data centre build-out is running into delays and rising costs, with investors now asking about both out loud. (8 min read)
- Quadrivio's AI fund bought Dev4Side to build a €100 million Italian AI group, with Microsoft signed on as the fund's technology partner. (2 min read)
- A South Korean developer published an AI-rebuilt clone of Hancom's office file format, and the country's copyright act allows the study but not the copy. (3 min read)
What ShiftedThe bill moved off the model
Three things changed position last week, and not one of them was a model. A grocer put AI on its payroll line, a cyber agency put robots on its liability line, and a credit investor put electricity on its balance sheet. Nothing shifted in what the models can do. Everything shifted in who pays for the room they run in.
Which line in your AI budget is not the model, and who signed for that one?
My call: by the end of March 2027, at least one listed company names a grid connection, not model quality, as the reason an AI programme slipped, in its own results statement.
The ones who looked closest backed away. If you have run a check that killed something your company had already announced, hit reply and tell me how that meeting went.
Yves
Tomorrow, First Move: the company in the news, and the one right behind it that nobody covers.
Reach data and AI leaders in this inbox: book a placement.
Ins7ghts Daily is published by 7wData.