Weekly AI News: Apr 1–7, 2026

Apr 2, 2026 Research

AI data centres can warm the surrounding area by up to 9.1°C, study finds

Research published in New Scientist found that AI data centres create significant local heat islands, raising outdoor temperatures in the surrounding area by as much as 9.1°C depending on density and cooling configuration. Standard air-cooling dumps waste heat directly into the local environment, with effects measurable for several kilometres around large facilities.

Why it matters: As data-centre density increases near urban areas, local heating effects could become a planning and public-health issue alongside the well-documented water and electricity demands.

Jensen Huang calls physical AI the next wave — but US data-centre builds need Chinese electrical equipment

NVIDIA CEO Jensen Huang declared at a recent event that the "physical AI" era — AI embedded in robots, vehicles, and industrial systems — will be bigger than the software AI wave. At the same time, Bloomberg reported that US data-centre construction is running months behind schedule because critical electrical gear, including transformers and switchgear, is still predominantly manufactured in China, creating supply-chain exposure in the middle of an infrastructure buildout race.

Why it matters: The physical AI vision and the infrastructure reality are in tension — the robots and data centres powering next-generation AI depend on supply chains that geopolitics could disrupt overnight.

The US holds 46% of the world's data centre capacity

A new analysis puts the United States at 46% of global data centre capacity, followed by Europe and then the Asia-Pacific region. The concentration reflects decades of investment in US internet infrastructure, regulatory stability, and the clustering effects of cloud providers headquartered in the country.

Why it matters: As AI workloads drive unprecedented data-centre construction globally, the US lead is being challenged — particularly by China, the UAE, and the EU.

Tesla FSD closes in on 9 billion cumulative miles — 10 billion expected within 30 days

Tesla's Full Self-Driving system is approaching 9 billion cumulative miles driven, with the company's own projections suggesting the 10-billion-mile mark will be crossed within roughly 30 days. Each additional billion miles at this stage represents a larger data advantage over any competitor building a competing end-to-end system.

Why it matters: At this scale the real-world training data moat grows faster than any rival can close — the gap is now measured in years of equivalent fleet exposure.

Apr 2, 2026 Robotics

Elon Musk says Optimus Gen 3 needs more finishing before its public debut

Elon Musk posted an update that Tesla's Optimus Gen 3 humanoid robot is walking but requires additional polish before a public reveal, describing the remaining work as "finishing touches." The announcement walked back earlier timelines suggesting an imminent demo, with no firm date now given for a public showing.

Why it matters: Tesla has repeatedly set and missed Optimus milestones, so continued slippage shapes how seriously the market prices its robotics ambitions.

Tesla Supercharger network delivered a record 6.7 TWh in 2025

Tesla's global Supercharger network delivered approximately 6.7 terawatt-hours of energy across 2025, a record for the company and roughly equivalent to the annual electricity consumption of a small country. The figure underscores the scale of Tesla's charging infrastructure as the broader EV market grows around it.

Why it matters: Energy throughput at this scale positions the Supercharger network as a meaningful piece of grid-level infrastructure, not just a vehicle accessory.

NVIDIA takes a $2 billion stake in Marvell Technology in a strategic AI partnership

NVIDIA announced it will invest roughly $2 billion in Marvell Technology, a chipmaker specialising in custom silicon and networking infrastructure for data centres. The deal pairs NVIDIA's GPU ecosystem with Marvell's expertise in custom ASICs and high-speed interconnects, extending NVIDIA's reach into the full AI hardware stack.

Why it matters: NVIDIA cementing ties with custom-silicon partners signals a strategic intent to own more of the AI infrastructure layer beyond GPUs alone.

Judge dismisses lawsuit by Sam Altman's sister against the OpenAI CEO

A federal judge dismissed the lawsuit filed by Annie Altman against her brother Sam Altman, which had alleged sexual abuse. The ruling ended the legal case, though the underlying allegations attracted significant public attention during the filing period. Sam Altman denied the claims throughout.

Why it matters: The case drew unusual scrutiny to the private life of one of tech's most prominent CEOs at a moment of intense public focus on OpenAI's governance.

The OpenAI graveyard: a growing list of deals and products that never happened

Forbes catalogued a string of OpenAI initiatives that were announced, hyped, or reported — then quietly shelved or indefinitely delayed, including robotics partnerships, enterprise deals, and research lines that never reached production. The list is a reminder that OpenAI's output is more selective than its pace of announcements suggests.

Why it matters: Scrutinising what doesn't ship is as important as tracking what does — especially for enterprises building dependency on OpenAI's roadmap.

OpenAI valued at ~$852B on secondary market as Anthropic runs hot at $380B

Secondary market trading shows OpenAI demand cooling sharply — shares have dropped from their February peak — while Anthropic has seen increased buying interest, putting its implied valuation near $380 billion. The gap between the two companies has narrowed significantly compared to six months ago, reflecting shifting investor sentiment on product traction and enterprise momentum.

Why it matters: Secondary market prices are a leading indicator of which AI labs investors believe have durable advantages — and the spread is compressing.

AI memory is the clear standout in the supply chain — Micron positioned to benefit

High-bandwidth memory (HBM) has emerged as the dominant winner in the AI infrastructure supply chain, with demand consistently outpacing supply as data-centre builds accelerate. Micron is increasingly cited alongside Samsung and SK Hynix as a key beneficiary, with analysts pointing to a multi-year demand curve driven by inference-at-scale requirements.

Why it matters: Memory constraints — not just GPU availability — are now a primary bottleneck shaping which AI workloads get deployed and at what cost.

Korean developer rewrites leaked Claude Code in Python — repo becomes one of GitHub's fastest-growing ever

Developer Sigrid Jin (instructkr) published claw-code, a clean-room Python rewrite of the Claude Code source that leaked earlier this year. The project rocketed to the top of GitHub's trending charts within days, attracting thousands of stars and a wave of contributor forks from developers who wanted a more accessible entry point to the underlying agent architecture.

Why it matters: Developer communities can now study and build on the patterns inside a frontier coding agent, accelerating open tooling for autonomous coding workflows.

AI News This Week