Nothing here is breaking news. That is the point.

Over the last 50 days, I have recorded six executive briefings, sat across from CTOs, founders, and chief economists, and read every report that crossed my desk. None of it qualified as a single headline. All of it, taken together, points one direction.

These patterns only show up when you zoom out. Once you see them, the comfortable story about where this industry is going gets harder to defend. Six patterns. One uncomfortable conclusion at the end.

In short: The last 50 days haven’t produced a single story big enough to dominate a news cycle. Together, they describe a structural shift the industry’s loudest voices are still mostly ignoring.
Table of Contents
  1. The platform layer is consolidating faster than anyone modeled
  2. The construction economy has split, and it isn’t coming back
  3. “Time to compute” is the new bottleneck, and it isn’t GPUs
  4. Supply chain became a boardroom problem, not a procurement one
  5. Most AI strategies are solving the wrong problem
  6. The single most important AI question executives aren’t asking

1. The platform layer is consolidating faster than anyone modeled

Three of the largest software companies in construction made the same move inside 50 days. Procore acquired DataGrid. Autodesk closed Rhumbix. Trimble signed to acquire Document Crunch on the 2nd of April. None are bolt-on chatbot plays. Each fills a specific workflow gap the incumbent could not build fast enough.

ServiceTitan reported that 59% of contractors who say AI is generating measurable impact are using AI built into software they already use. They are not shopping for AI. They are buying workflows that happen to have AI baked in.

Josh Levy, founder of Document Crunch, put it sharply: “you can’t vibe-code trust.” Foundation models let anyone build a contract review tool in a weekend. They cannot build seven years of customer feedback loops, a data pipeline running through 10,000 projects, or relationships with legal departments at top-20 contractors.

If your core platform hasn’t published an AI roadmap you can read in plain English, the gap is widening every quarter.

Go deeper: The full breakdown, including why Trimble’s Mark Schortz called construction a deterministic industry and what that means for every AI tool you are being sold, is in Executive Briefing 6.

2. The construction economy has split, and it isn’t coming back

US data center construction starts hit $25.2 billion in a single month in January 2026. Trailing twelve-month spend is $103.7 billion. Contractors with data center work are sitting on 11.2 months of backlog. Contractors without it: 7.6.

That gap is structural. The top five hyperscalers have around $969 billion in future lease commitments, $662 billion of which hasn’t broken ground. That is committed infrastructure spend that lands on real sites whether residential recovers or not.

The other side of the market is grinding. UK contract awards down 43% year on year. Project starts down 31%. The US architectural billings index is still in contraction, meaning new projects aren’t entering design while data center starts hit record highs. Office starts are forecast down 18 to 22%. Manufacturing build-to-suit down 20 to 28%.

Carl McFarland, who has run construction businesses for 31 years, was blunt: nobody rethinks their business model during their best financial years. Record profitability in AI infrastructure is masking the fact that the underlying model is about to change.

Go deeper: The full Carl McFarland conversation, including the consortium-style transformation model he is pushing for, is in Executive Briefing 6.

3. “Time to compute” is the new bottleneck, and it isn’t GPUs

At Nvidia’s GTC conference, Daniel Roberts, co-CEO of publicly traded data center operator IREN, didn’t talk about GPUs. He talked about electricians.

His phrase was “time to compute,” and it is entirely a built-world problem. You cannot deploy AI without first pouring foundations, running cable, and installing switchgear. Electrical work is 45 to 70% of total data center construction cost. Nearly one in five US electricians is over 55. The country needs 300,000 new electricians over the next decade just to meet data center demand, and 200,000 current electricians retire in the same window.

The capital is following. BlackRock just announced a $100 million Future Builders initiative to train 50,000 trades workers over five years. Google has pledged $50 million for the electrical pipeline. OpenAI is funding Trades Futures. Microsoft has publicly named electrical talent as the single biggest constraint on its data center expansion.

Patrick from Foundamental made the point that lands: none of the AI CEOs predicted they’d be short of electricians when they were selling the AI dream. The same people now write nine-figure checks to train plumbers.

Go deeper: The full breakdown, including IREN’s $9.7 billion Microsoft contract and what it tells you about labor pricing in your market, is in Executive Briefing 5.

4. Supply chain became a boardroom problem, not a procurement one

Noble Francis, economics director at the Construction Products Association, has been telling boards directly: stop optimizing for the cheapest source, start optimizing for the most secure one. There have been major disruptions every year for the last six, and there will be major disruptions every year going forward. Firms just won’t know which flavor they come in.

The numbers make the case. The average construction company is overpaying by around 5% on materials without knowing it, because they have never had structured visibility across projects, regions, and vendors. Around 40% of labor idle time on a typical project is caused by materials, not weather, not shortages.

Then Iran. US construction input prices rose at a 12.6% annualized rate through February, and ABC’s chief economist Anibal Basu notes that doesn’t yet reflect the Hormuz oil shock. UK input cost inflation just hit a 41-month high. Aaron Anderson’s analysis pegs it at roughly 130,000 fewer US permits filed for every 13% the cost index runs above trend.

If you are still treating procurement as a back-office function, you are exposed.

Go deeper: The full Noble Francis playbook, including how to diversify suppliers when the cheaper option is killing you on bid day, is in Executive Briefing 3.

5. Most AI strategies are solving the wrong problem

Matt Gough, former innovation director at Mace and three years at Neom, just published a report from interviews with 50 senior AEC leaders. Every single one was bullish on AI: budgets up, ROI being generated, pilots becoming deployments. Every single one was using AI to optimize the same operating model that has underperformed for decades.

His phrase is “don’t automate the past.” The Stanford AI Index showed inference cost for GPT-3.5 level intelligence dropped 280 times between November 2022 and the end of 2024. Intelligence is heading toward free. The friction is no longer information. It is how the industry organizes people around it, how risk gets passed down the chain, how fragmented contracts misalign incentives.

One company in his sample stood out. They run something called the rule of five: if a task takes more than five minutes, or you’re doing it more than five times, you should be using an AI tool. Every employee gets a monthly budget for whatever tooling they want, on one condition: come back and share what you learned. That was 1 out of 50.

The other 49 are making a broken operating model run 10% faster. The firms that reorganize the value chain itself, what Matt calls virtual vertical integration, will eat their margin.

Go deeper: The full Matt Gough thesis, including why construction has no moonshot the way healthcare does, is in Executive Briefing 4.

6. The single most important AI question executives aren’t asking

At Nvidia GTC, Arthur Mensch, CEO of Mistral, said the real unlock for enterprise AI isn’t smarter models. It is governance, what he calls observable deterministic processes.

The distinction every executive needs to internalize: deterministic systems give you the same answer every time. Your structural calculation software. Your payroll engine. Your spreadsheet formulas. Probabilistic systems make their best guess based on patterns. Every large language model, every ChatGPT, every Claude, is fundamentally probabilistic. 90% accurate sounds great until the 1% on a structural calc, a prevailing wage report, or a compliance submission costs millions.

Anna Berger from Trade, the construction payroll startup, put it bluntly on our panel. Trade tracks something they call payroll health: if they hit 99.7% accuracy and are off by 0.03%, they treat that week as a failure. A single cent off on a prevailing wage report can trigger fines in the tens of thousands.

Carl McFarland built on this with the cleanest framework I’ve heard. Organize your AI initiatives into two buckets: efficiency and risk mitigation. The efficiency bucket can be loose, let people experiment. The risk bucket needs governance, human review, deterministic checkpoints. The returns there, in Carl’s experience, are roughly 10x higher. Almost nobody is auditing their portfolio against this taxonomy.

Go deeper: The full two-bucket framework, including the example where Big D’s agent caught a six-figure risk a PM missed, is in Executive Briefing 6.

The uncomfortable conclusion

Six patterns, 50 days, one direction.

The platform players are absorbing the AI-native companies. The economy has split, and the side you’re on determines your labor cost, your material cost, and your margin. The bottleneck is now electricians, not GPUs. Procurement is no longer a procurement problem. Most AI strategies are aimed at the wrong target. And the firms deploying AI well are doing it with a governance taxonomy almost nobody is talking about.

Earlier in this window, a VP at a multi-billion-dollar US contractor gave me an analogy I have not been able to shake. Most of the industry is swimming. The water is warm, there is plenty of work, the data center boom is paying well. A smaller group has noticed the swell. The wave is here. They are paddling hard, exerting energy and not yet feeling the progress. When the wave breaks, the swimmers will still be in the water. The paddlers will be somewhere else entirely.

Look at the six patterns again. Then ask which one you are.

Trusted Sources