Major Data Center CEO on Labour Demand, Power Scarcity, and the Real Work Behind AI Buildouts

The AI data centre boom has a labour problem, and it looks a lot like every workforce crisis the construction industry has been dealing with for the last decade. The difference is scale, urgency, and the fact that the world’s largest asset manager just put $100 million on the table to try and fix it.

In a recent All-In interview from Nvidia’s GTC conference, IREN co-CEO Daniel Roberts gave one of the most operationally honest accounts of what it actually takes to build AI infrastructure at scale. Not the GPU specs. Not the model benchmarks. The dirt, the crews, the grid connections, and the brutal reality that you cannot deploy artificial intelligence without first deploying electricians.

For anyone in construction, this is not a tech story happening somewhere else. It is your story, wearing a different hard hat.

IREN’s playbook: bootstrap with Bitcoin, pivot into AI compute

IREN started in Sydney about eight years ago. The thesis was simple and, in hindsight, unusually disciplined: the digital world was expanding faster than the physical infrastructure could support it. Use Bitcoin mining as the first customer to bootstrap data centre assets, generate cash flow, and layer in higher-value compute workloads as they emerged.

Roberts described a false dawn in 2020 when they signed an MOU with Dell to start bringing in AI customers, but the timing was too early. They retreated to Bitcoin, kept building the platform, and waited. About two years ago, demand for AI compute started hitting in earnest and has escalated month by month since. Today, IREN is swapping out Bitcoin mining rigs for AI chips across its entire portfolio. The company signed a $9.7 billion contract with Microsoft late last year, and Roberts noted that deal represents only about 5 percent of IREN’s total capacity. Even a massive contract is now just one line item in a bigger capacity race.

The bottleneck is not chips. It is time to compute

One of the most useful things Roberts did in the interview was reframe the constraint. For much of the industry, people talk about GPU availability or power access. For IREN, because they started assembling land and power years before anyone else cared, the bottleneck has shifted to what Roberts called time to compute.

That phrase deserves unpacking for a construction audience because it is entirely a built-world problem. Time to compute means getting tradespeople to remote sites. It means pouring foundations, installing cooling systems, solving supply chain gaps for transformers and switchgear, and bringing enormous electrical systems online in locations that were never designed for this intensity of industrial activity. It means workforce housing. It means local community engagement. It means the kind of sequencing and logistics work that construction project managers do every day, except the client is a hyperscaler who needed it operational yesterday.

The labour crisis: 439,000 workers short and counting

Here is where the IREN story connects directly to the challenge facing every contractor, subcontractor, and workforce planner reading this.

The U.S. construction industry is currently short approximately 439,000 workers, with electricians, pipe layers, and HVAC technicians among the hardest positions to fill. The Associated Builders and Contractors estimates the industry will need to attract 349,000 net new workers in 2026 alone just to keep pace with demand. Add the projected need for 456,000 more in 2027, and the picture gets worse before it gets better.

For data centres specifically, the numbers are even more stark. Electrical work accounts for 45 to 70 percent of total data centre construction costs, according to the International Brotherhood of Electrical Workers. Nearly one in five electricians currently working in the United States is over 55. The country needs an estimated 300,000 new electricians over the next decade just to meet AI-driven data centre demand, and another 200,000 current electricians are expected to retire during that same period.

Roberts described IREN’s approach to this problem with a level of detail that most tech executives skip entirely. The company hires locally first, then expands the radius by 20 miles at a time when local talent is exhausted. They deliberately locate near areas where heavy electrical infrastructure already exists but previous manufacturing has closed down, allowing them to leverage sunk capital expenditure and retrain local workforces for a new industry.

That is a smart playbook. It also mirrors what the best contractors have always done: go where the infrastructure is, invest in the community, and build a workforce rather than just poaching one.

BlackRock’s $100 million bet that capital alone is not enough

The scale of the labour problem has now attracted attention from the very top of the financial food chain. In March 2026, BlackRock announced its Future Builders initiative, a $100 million philanthropic programme aimed at training 50,000 skilled trades workers over the next five years.

The context matters. BlackRock is not doing this out of abstract goodwill. The firm is one of the largest infrastructure investors in the world and is directly backing projects like Meta’s estimated $27 billion Hyperion data centre in Louisiana. When Larry Fink says that America needs an estimated $10 trillion in infrastructure investment by 2033 and that capital alone is not enough, he is talking about his own portfolio being constrained by a shortage of electricians.

Microsoft’s president Brad Smith has publicly identified electrical talent shortages as the single biggest challenge for data centre expansion in the United States. In some cases, Microsoft has had to employ electricians commuting 75 miles to site or temporarily relocating them. Oracle reportedly shifted data centre completion dates from 2027 to 2028 partly due to labour shortages, though the company has disputed that characterisation.

The signal from all of this is consistent. The firms writing the largest cheques in artificial intelligence are now spending significant time and money trying to solve a construction workforce problem. That should tell every contractor, training provider, and workforce planner in the industry something about where the centre of gravity is moving.

The energy siting strategy construction teams should understand

Roberts made another point that deserves attention from anyone involved in site selection or infrastructure development. IREN has used 100 percent renewable energy since inception, relying on hydropower in British Columbia and wind and solar in West Texas. But the strategic insight is not the sustainability angle. It is the siting logic.

In West Texas, there is roughly 45 to 50 GW of wind and solar generation capacity but only about 12 GW of transmission infrastructure to export that power to urban load centres in Dallas and Houston. If transmission is the choke point, then moving compute to the source of stranded generation becomes a genuine arbitrage rather than a marketing exercise.

For contractors and developers, this inverts the traditional logic of where you build and why. Data centres are not following population density. They are following excess power capacity, grid connections, and available land. The firms that understood this early, IREN among them, now own optionality that late entrants cannot easily replicate.

The data centre boom is a construction workforce story

Peak crew sizes on data centre campuses have moved from 750 workers to 4,000 or 5,000 on sites like DataBank’s Red Oak campus. That is the population of a small town arriving at a construction site. Workers are seeing pay increases of 25 to 30 percent compared to previous roles, with some electricians and project managers earning well above $200,000 annually. Construction wages in the sector averaged $40.55 per hour in January 2026, with weekly earnings running about 25 percent higher than the private sector average.

But the boom comes with a structural tension. Once a data centre is built, the operational workforce is a fraction of the construction workforce. The work is project-based, not permanent. That feast-or-famine cycle is familiar to every contractor, and it means the industry needs to think about workforce development not as a one-time recruitment exercise but as a continuous pipeline.

The companies that win this infrastructure cycle will not be the ones with the most ambitious AI strategy decks. They will be the ones that can get power, permits, crews, and capacity before everyone else works out that those were the scarce inputs all along. For construction, that is not a warning. It is a generational opportunity.