The UK's most expensive infrastructure project is now confirmed at up to £103bn for less than half its original scope. The AI pricing your technology budget runs on shifted last month through mechanisms the rate card won't show you. Interestingly, construction robots are generating verified, repeat revenue on live job sites for the first time. The thread connecting all three: in each case, the bill arrived before anyone had built a system to read it.

THE EXECUTIVE BRIEFING
THIS WEEK’S KEY TAKEAWAYS

Full episode write-up at the bottom↓

 AI pricing is running a shell game. Per-token costs are falling; enterprise bills are climbing. Agentic workflows consume 10 to 20 times more tokens per task, and both OpenAI and Anthropic quietly raised effective prices last month through different mechanisms.

Construction robotics has crossed from pilots to production. Five structural barriers that held the industry back for 60 years are bending simultaneously. The first nine months of 2025 saw $1.36 billion in venture capital invested in the category.

HS2 will now cost between £87.7 and £102.7 billion, completing 4 stations and 140 miles. France builds comparable high-speed rail for roughly £20 million per kilometer. The UK is paying 15 to 20 times the international rate for the same technology.

7 THINGS WORTH YOUR ATTENTION
ON THE RADAR THIS WEEK

On Thursday, 28th May, Owen will be heading to Contech Connect in Paris. Following Patric Hellermann's opening keynote, we'll be joining him on the main stage for 45 minutes of grilling on their latest "State of the Project Economy 2026" research piece. 

Later, join Owen from around 11:15 am for a 2-3hr livestream diving into construction technology with a roster of top guests and interviews. [JOIN HERE]

POWERED BY:

This week produced three stories that look unrelated until you place them alongside each other. The United Kingdom confirmed its most expensive infrastructure project at up to £102.7 billion for less than half its original scope, then handed the same industry the right to bid on the next one.

The two largest AI labs quietly raised their effective prices through mechanisms most finance leads will have missed. And a new generation of construction robots has moved, for the first time, from demos into verified repeat revenue on live job sites. The thread running through all three is identical: what happens when accountability structures arrive too late?

A few months ago, this briefing carried a data point from a senior construction technology leader: the cost of intelligence had fallen 280 times in two years. The friction, he argued, had stopped being pricing. Per-token costs have continued to fall. Some of the cheapest models today run for less than thirty pence per million tokens. But two pricing moves happened in the same week last month that most finance leads will have missed entirely:

The reason this matters in construction is agentic workflows. An agentic system completes a task in a similar way a junior analyst would: reading documents, checking its own output, calling other tools, and verifying the result. That single task can consume 10 to 20 times the token volume of a simple exchange. What economists call the Jevons paradox applies: as the cost per unit falls, consumption climbs because applications that were uneconomical at the old price become viable at the new one.

Millie Bridge has spent a decade building natural language software for construction litigation. When the major AI labs told her in 2024 not to worry about compute costs, she built her company the opposite way. Every client's document set gets pre-processed down to roughly one percent of what actually matters before an expensive token is spent. Her description: "You don't need to kill a rabbit with a bazooka."

Jean-Marc Shimizu runs open innovation at Shimizu Corporation, one of Japan's five mega general contractors, after two decades at British Telecom's corporate venture arm in Silicon Valley. Shimizu has onboarded five thousand of its twenty thousand staff onto an AI platform built on Light Blue, a Tokyo University spinout, in seven months.

Ask your finance lead and your IT lead the same question separately: What did AI spending look like in April versus March, and what are we getting for it? If neither can answer with a number, that's a budget problem that compounds every quarter.

The headline from Bricks & Bytes' physical AI research this week: after 60 years of construction being the last sector to automate, the first real wave of construction robots has graduated from demos to verified production deployment.

Five structural barriers held the industry back. Each is bending:

Bedrock Robotics raised $270 million at unicorn valuation in February, retrofitting autonomous kits onto major OEM equipment. Crewline raised $7.1 million as a four-person team in April with a $26 million order book already behind it. Both are production deployments on live American job sites.

The biggest failure mode in construction robotics is the business model. Technology rarely fails first. Maria Telleria, co-founder of Canvas, earned early trust by getting her own drywall-finishing licence and working alongside the robot on site. Workers knew a human with manual skills was present if anything went wrong. Physical credibility built one deployment at a time is how the industry learns to trust new technology.

Transport Secretary Heidi Alexander confirmed to Parliament on May 19 that HS2 will cost between £87.7 and £102.7 billion. First services between London and Birmingham will not run until 2036 to 2039. The original 2012 promise was eight cities, 340 miles, open this year, at £32.7 billion. The delivered project: four stations, 140 miles. At over £400 million per kilometer, The United Kingdom pays roughly fifteen to twenty times what France or Spain pay for comparable high-speed rail.

Three separate reviews reached the same conclusions. The specification was gold-plated; trains designed to exceed any operational speed in the world made every component bespoke. Main civil contracts were let before the designs were ready, transferring risk that contractors could not price. The commercial model was target-cost on paper, cost-plus in practice. HS2 Limited lacked the commercial muscle to manage what it had bought. Board pack data, per the most recent review, was "a veil behind which less good news becomes difficult to assess."

Mark Wild, who delivered the Crossrail recovery and is now running the HS2 reset, told the parliamentary committee this week: "mostly inefficiency of work, because we started too soon."

HS2 ran for fifteen years without meaningful accountability for its cost because the commercial structure actively prevented it. Your AI budget is rising in ways most finance teams can't track, because no governance mechanism has been built to watch it. Construction robots fail commercially because no contractor wants to be first to absorb unproven risk. In all three cases, costs are diffuse, ownership is absent, and the reckoning arrives late.

The firms that get ahead of this build accountability structures before the bill arrives. HS2 proves structure determines what gets noticed and what doesn't. Energy as a project input is already repricing in real time. Your AI spend is next.

If this briefing was useful, share it with a finance or technology leader on your team. Hit reply to share your honest answers: Are you tracking AI spend? Have you seen the actual bill? Which robotics workflow would you back first? We read every email and respond. Your views help us build content that impacts our readers.