Compute & Infrastructure

A Guardian investigation exposes significant gaps between UK government announcements and actual AI infrastructure delivery. The centrepiece Nscale supercomputer project in Essex — repeatedly featured in press releases with completion dates this year — remains a scaffolding depot and scrap metal site. Planning permission was filed only after Guardian inquiries. Despite this, Nscale raised $2 billion at a $14.6 billion valuation, adding former Meta executives Sheryl Sandberg and Nick Clegg to its board, as reported by The Guardian and TechCrunch.

The investigation also found that much of the government's touted AI capacity relies on rented data centre space rather than owned infrastructure, and questions whether Nvidia-backed investments represent genuine new capacity or repackaged existing commitments. Government officials have claimed these projects will 'mainline AI into the veins of the economy,' but physical evidence suggests delivery lags announcements by months or years.

Taiwan Semiconductor Manufacturing Co. reported 30% sales growth for January-February 2026, reflecting robust AI hardware demand prior to the Middle East conflict escalation, according to Bloomberg. The growth comes even as geopolitical risks intensify around Taiwan itself and as major customers face energy and site location challenges. Apple has shifted roughly 25% of iPhone production to India — a 53% increase year-over-year — demonstrating accelerating diversification from China manufacturing, per Bloomberg.

The Middle East conflict is forcing infrastructure strategy shifts. Oracle and OpenAI cancelled expansion plans for their Texas flagship data centre, as reported by Bloomberg Tech. Oil prices above $100 per barrel are raising operational costs, while Financial Times questions why US tech firms concentrated AI infrastructure in the Gulf region given current instability. Energy constraints are emerging as the binding constraint on expansion.

Anthropic filed federal lawsuits in California district court challenging the Department of Defense's designation of the company as a supply-chain risk, with executives warning the label has caused commercial partners to pause deal negotiations and could result in billions in lost revenue, per Wired and Bloomberg. The lawsuits claim the Trump administration overstepped legal authority and violated Anthropic's First Amendment rights by escalating a contract dispute over acceptable use cases for Claude into a government-wide ban.

The dispute has drawn unusual public solidarity from competitors: nearly 40 employees from OpenAI and Google DeepMind, including Google's chief scientist Jeff Dean, filed an amicus brief supporting Anthropic's position, as reported by Wired and The Verge. A Pentagon official stated there is little chance of resuming negotiations, according to Bloomberg. The conflict centres on Anthropic's refusal to permit its models for certain military surveillance applications against Americans.

Biotech startup Cortical Labs is constructing two small-scale data centres in Singapore and Melbourne using lab-grown human brain cells integrated onto silicon substrates, experimenting with biological neural networks as computing infrastructure, according to Bloomberg. The approach represents a fundamental departure from semiconductor-based compute, potentially offering different power efficiency and processing characteristics. Separately, researchers published work reconstructing video from mouse brain activity, demonstrating progress in understanding biological neural processing, per The Guardian.

While these projects remain experimental and far from commercial scale, they signal growing interest in alternative computing substrates as conventional silicon scaling faces physical and economic limits. The initiatives are small — measured in lab installations rather than megawatts — but represent hedge bets by infrastructure investors exploring post-semiconductor pathways.