March 30, 2026

The clean-energy narrative around AI data centres is getting stress-tested in public. A new report says Google, Microsoft, Amazon and Meta are leaning more on natural gas and other fossil generation to keep up with AI-driven load, as grid interconnection backlogs and shifting U.S. policy slow the clean buildout (AI surge forces tech firms to rely on natural gas). In the same news cycle, the Trump administration is reported to have paid TotalEnergies $1 billion to unwind offshore wind projects — a reminder that energy policy can swing fast, and corporate decarbonisation plans can get dragged with it (Weekly ESG roundup: policies, deals, and sustainability funding).

The Big Stories

The most consequential takeaway today is that the AI buildout is colliding with the realities of power availability. The report frames a widening gap between Big Tech’s 2030 net-zero/carbon-removal commitments and what’s actually powering new capacity right now: more natural gas and other fossil generation, driven by surging demand, interconnection delays, and U.S. federal policy shifts that have complicated renewable procurement. That matters because it changes the near-term “winning” generation mix for incremental data-centre load — and it puts a spotlight on the bottlenecks (grid queues, permitting, policy support) that will decide whether AI growth is paired with clean supply or just displaces emissions elsewhere.

The ESG roundup adds useful texture on why those bottlenecks are hard to shrug off. The headline geopolitical jolt is the reported $1 billion payment from the Trump administration to TotalEnergies to end U.S. offshore wind projects, alongside broader policy moves like Germany’s 2030 climate action plan and India’s cautious 2035 clean-energy goals. On the corporate side, Microsoft’s 1 million-ton carbon removal deal with Liferaft shows the “back end” of decarbonisation is still moving — but it also underlines the problem: carbon removal contracts don’t solve the immediate question of what powers next quarter’s incremental AI megawatts.

Behind the Headlines

Talent is becoming its own form of capacity, and Microsoft is leaning into that with a local, practical play in Australia. Microsoft and Victoria University open Victoria’s first Datacentre Academy launches a fee-free programme for 48 students at VU’s Footscray Nicholson Campus, offering 12-week and 16-week tracks for Datacentre Essentials and Critical Environment Technicians, backed by an industry-matched Victorian government grant. The telling detail is the cited projection that Australia will need 8,300 data-centre workers by 2030 — a signal that labour constraints are being treated less like an HR problem and more like an infrastructure constraint. If you’re tracking time-to-power and time-to-build, time-to-staff is quietly becoming part of the same critical path.

West Africa is also laying groundwork that matters to data infrastructure investors, even if it reads like bureaucracy at first glance. ECOWAS ministers meet in Freetown to advance digitalisation describes ministers endorsing regional moves to expand broadband, harmonise data protection and cybersecurity, and stand up two concrete institutions: a Regional Cybersecurity Coordination Centre and a Regional Internet Exchange Point. The practical significance is that regional policy alignment can reduce friction for cross-border networks, cloud adoption, and locally anchored traffic — all prerequisites for more viable in-region data-centre economics. The next watch item is whether ECOWAS statutory bodies move quickly from endorsement to formal adoption, because timelines (and enforcement) are what separate “digital transformation” talk from investable infrastructure reality.