Aberdeen shows where AI build-outs are heading

The National says plans for a new multi-billion-pound data centre and AI campus near Aberdeen would turn around 200 acres at Blackdog, just north of the city, into a major technology hub. The outline is big. A £10bn multi-gigawatt campus, 600MW of renewable and grid-connected power in the first phase, planning consent already granted and hopes of about 1,000 jobs.

That's the useful bit. This is what AI build-out looks like once it becomes a construction programme. It needs land, grid capacity, water, planning and a credible power story. The developers are saying the site can tap offshore wind and subsea cable connections. That is why the project looks real enough to move.

The Register's read is blunt. UK AI datacentre capacity is starting to move away from London as power shortages, planning limits and land constraints bite. For many AI workloads, low latency into the City is no longer the main issue. Access to the grid is what matters.

That matters because a lot of teams still default to London first and work backwards. The smarter question is what workload you're actually running, what latency it really needs and where the power exists. If the answer is not London, that does not weaken the project. It just changes the geography.

BAM UK&I makes the point plainly. Construction is not any industry. It works on critical infrastructure, tight margins and supply chains that can expose the whole job if controls are weak. The message is to start with specific use cases, build securely and prove the return before pushing further.

That is a sensible line. AI can help with back-office work, admin and repeatable tasks. But, if the tool can't survive procurement, governance and supply-chain scrutiny, it shouldn't be on the project in the first place.

Construction Dive argues that AI is taking over the mechanical layer of bidding. Reading drawings, extracting quantities and structuring estimates can now be automated far faster than before. That leaves the judgment work, scope, risk, vendor management and bid strategy with the people who actually know the job.

That is the right direction. Estimating has never just been about counting. It is about seeing the risk early and deciding what to bid, how to bid and what to ignore.

Bobyard says its 2.0 release automates up to 70% of quantity and material takeoff, with average takeoff times down 65% and users submitting three to five times more bids. The new bits are practical. Multi-Measure lets you draw once and get area, perimeter and volume. Review Workflow keeps a human in the loop. Legend Manager and Text Count tackle more of the dull parts.

This is not about magic. It is about shaving hours off the work that sits between drawings and a bid.

Moonshot's Kimi K2.6 landed as an open-weight 1T-parameter model with 256K context, native multimodality and broad tool support on day one. The detail that matters for construction isn't the benchmark theatre. It's the way open models are getting better at long jobs, repeated tool use and keeping state across a workflow.

If you are building internal assistants around drawings, PDFs, screenshots and messy admin, that direction matters. The model layer is getting more capable, but the real win is still in the harness and the workflow.