BlackRock signs union deal: AI's bottleneck is electricians

BlackRock's Global Infrastructure Partners, together with the AI Infrastructure Partnership, signed a memorandum on Aug 10 with North America's Building Trades Unions to line up skilled workers for AI data centers and energy projects. When the largest asset manager on earth starts negotiating for pipefitters, the constraint on AI capacity has moved. It isn't chips and it isn't money. It's people who can pull conduit.

What actually happened

Bloomberg reported the agreement between BlackRock's infrastructure arm, the AI Infrastructure Partnership — a roughly $30 billion joint venture with Microsoft and other large technology firms — and NABTU, which represents more than 3 million skilled trades workers.

The terms are about sequencing, not construction contracts. Both sides commit to sharing upcoming project pipeline information early, so unions can recruit and train against specific sites before ground breaks. The memorandum also points at responsible contractor programs and project labor agreements — pre-negotiated wage and condition contracts that speed hiring on large builds. BlackRock has separately committed $100 million to skilled-trades training, per Engineering News-Record.

This is a memorandum, not a construction schedule. What it tells you is what the money believes: capital is arriving faster than the workforce that turns capital into megawatts.

Why the data center labor constraint matters for your business

Every AI capacity announcement you read has a delivery date attached, and those dates are set by trades availability as much as by silicon. That's the same reason gigawatt-scale projects keep landing in 2028 and EU gigafactory compute arrives years out. A labor shortage doesn't announce itself as a shortage. It shows up as a slipped quarter.

Two practical consequences.

Your token price has no downward pressure coming. If capacity lags demand, providers have no reason to cut rates, and the cheap tiers stay rationed. Budget your automation on today's prices, and treat any decrease as a bonus rather than a plan.

Your capacity is regional and contested. Data centers compete for the same electricians as hospitals, chip fabs, and grid upgrades — and increasingly for the same permits and political goodwill. Concentrating your entire stack in one provider's one region is a bet on that region's labor market, not just on the vendor.

The build response is boring and effective: keep workloads portable, keep a second provider warm, and put the volume work on hardware or models you control so a capacity crunch throttles your costs instead of your operations. We've said the same thing about the compute bill and about power as the real ceiling. Labor is the third leg of the same table.

Sources: Bloomberg, Engineering News-Record.