01
China's Humanoid Robots Run a Half-Marathon Faster Than Any Human Has Ever Done It
At the Beijing E-Town Half Marathon on 19 April, more than 300 robots raced alongside 12,000 human runners in the world's largest human-versus-machine endurance event. The winner — a bipedal humanoid named "Lightning," built by Chinese smartphone maker Honor — completed the 21-kilometre course in 50 minutes and 26 seconds, beating the human world record of 57 minutes set by Uganda's Jacob Kiplimo in Lisbon last month. At least four robots finished in under an hour. The previous year's winning robot took two hours and 40 minutes. Lightning crashed into a railing near the finish and required handlers to set it upright before completing the race.
Why it matters: China's three leading humanoid manufacturers already hold the top global shipment rankings, and the Beijing race is state-backed industrial theatre with commercial intent. For manufacturers and logistics operators evaluating automation partners, the pace of Chinese hardware development — and its alignment with government procurement policy — is now a material supply-chain and competitive-positioning variable, not a future consideration.
02
AI Hardware Is Now the Largest Single Driver of Global Trade — Accounting for a Third of All Growth in 2025
The McKinsey Global Institute's 2026 trade report, covering more than 90% of global goods trade, finds that shipments of AI-related hardware — semiconductors, servers, and networking equipment — grew nearly 40% in 2025 and accounted for approximately one-third of all global trade growth that year. The United States added roughly half of the world's new data centre capacity, with US trade in AI-related goods rising by an estimated $220 billion, or 66%. China, constrained by export controls on advanced chips, saw AI-related trade grow by only 16%, or approximately $85 billion. Asian supply hubs — Taiwan, South Korea, and parts of South-East Asia — were the primary producers. Despite tariff rates reaching their highest level since the Second World War, global trade grew faster than the global economy in 2025. The report concludes that AI infrastructure has become a structural driver of trade flows rather than a cyclical one.
Why it matters: AI hardware has become a geopolitical commodity. Supply chains for the chips, servers, and networking gear that power AI run overwhelmingly through Taiwan and South Korea — economies exposed to significant geopolitical risk. Companies with data centre or AI infrastructure ambitions should be stress-testing the concentration risk in their hardware supply chains and reviewing whether their procurement is with geopolitically aligned partners.
03
Anthropic's Own Usage Data Maps Which White-Collar Jobs AI Is Already Doing — and How Far It Could Go
Economists Maxim Massenkoff and Peter McCrory at Anthropic published a study introducing "observed exposure" — a metric measuring not what AI could theoretically do in a given role, but what it is actually doing, based on real enterprise usage of Claude. Theoretical AI coverage exceeds 94% of tasks in business and finance and computer and mathematical occupations. Actual adoption remains a fraction of that. Early labour market data shows a 6% to 16% fall in employment among workers aged 22 to 25 in the most AI-exposed roles — driven by slower hiring rather than layoffs. The researchers name a "Great Recession for white-collar workers" as a plausible scenario if adoption accelerates faster than the economy can absorb.
Why it matters: This is the most granular map yet of AI's actual — not theoretical — penetration into knowledge work. For boards, CEO’s and Human Resource Officers, the finding that entry-level hiring is already slowing in high-exposure roles has immediate implications for graduate pipelines, workforce planning, and the internal talent structures that feed senior management. The gap between observed and theoretical exposure is the timeline you have left to act.
04
UAE Sets Two-Year Deadline to Run Half Its Government on Autonomous AI Agents
On 23 April, Sheikh Mohammed bin Rashid Al Maktoum, Vice President and Prime Minister of the UAE, announced that 50% of UAE government sectors, services, and operations will run on agentic AI within two years — systems capable of executing multi-step tasks and making decisions without human intervention. The initiative was announced under the directives of UAE President Sheikh Mohamed bin Zayed Al Nahyan. Implementation will be overseen by a dedicated Cabinet taskforce led by the Minister of Cabinet Affairs. Ministers and directors-general will be assessed on their pace of adoption. The UAE frames the move as positioning itself as the first government globally to operate public services at this scale through autonomous systems, building on two decades of digital infrastructure development including the UAE Pass identity platform.
Why it matters: No government has committed to agentic AI deployment at this scale with this timeline. For multinationals operating in the Gulf, this signals a procurement and compliance environment that will shift materially within 24 months. For enterprise AI vendors, it is the largest single government deployment signal of 2026. Competitors in Singapore, the UK, and the EU should treat this as a strategic benchmark.
05
AI Data Centre Electricity Demand Rose 50% in 2025 and Will Triple by 2030 as Tech Capex Hits $400bn
The International Energy Agency's report Key Questions on Energy and AI, published 16 April, finds that electricity consumption from AI-focused data centres surged 50% in 2025, far outpacing overall data centre demand growth of 17% and global electricity demand growth of 3%. The capital expenditure of five major technology companies exceeded $400 billion in 2025 and is set to rise a further 75% in 2026 — exceeding global investment in oil and gas production. AI-specific data centre capacity more than tripled in the past 18 months. The SMR nuclear pipeline tied to data centre offtake agreements has grown from 25 gigawatts at end-2024 to 45 gigawatts today. The IEA warns that supply-chain bottlenecks — transformers, grid connections, high-bandwidth memory — are now the primary constraint on deployment speed.
Why it matters: Energy infrastructure is now the binding constraint on AI competitive advantage. Companies that have secured power — through PPAs, nuclear offtakes, or co-location agreements — hold a structural moat. For industrial and manufacturing CFOs, the IEA's finding that AI applications could reduce energy costs by 3 to 10 percentage points sets a clear ROI benchmark for adoption decisions this year.
Editor's Note
The defining shift of this week's stories is not what AI can do — it is who is moving fastest to deploy it at scale, and who controls the infrastructure on which it runs.
China leads on humanoid hardware. The UAE has set a government-wide deadline for autonomous AI. The IEA confirms that tech capex now exceeds global oil and gas investment. Anthropic's labour data shows the disruption of knowledge work is already under way — quietly, at the hiring stage. McKinsey's trade figures make clear that the compute and chips powering all of it flow through a handful of geopolitically concentrated supply chains. Access to energy, hardware, and talent is no longer a technology question. It is a strategic one.
— The AI Thesis Editorial
🌿 Good News · AI Making a Difference
AI Reveals Ocean Currents That Have Never Been Observable Before — Using Satellites Already in Orbit
Researchers at UC San Diego's Scripps Institution of Oceanography and UCLA published a study in Nature Geoscience on 20 April describing GOFLOW — Geostationary Ocean Flow — a deep learning system that converts routine thermal imagery from existing weather satellites into hourly, high-resolution maps of ocean surface currents. Previous satellite methods revisited the same ocean location only once every ten days, too slow to capture currents that form and disappear within hours. GOFLOW produces maps hourly and reveals small-scale features — eddies, boundary layers, vertical mixing zones — that were previously visible only in simulations, not real-world observations. Validated against shipboard measurements in the Gulf Stream, it matched existing techniques while revealing far finer detail. Because it requires no new hardware, it could be integrated directly into weather forecasts and climate models. The project was funded by the Office of Naval Research, NASA, and the European Research Council.
This Week's AI Tip
Use AI to argue against you — before anyone else does.
Most professionals use AI to help build a case. The more powerful move is to use it to break one. Once you have drafted a strategy, proposal, or decision memo, paste it into an AI and ask it to steelman the opposition. You will surface blind spots faster than any internal review cycle.
The key is to be specific about what kind of pushback you want. A generic "challenge this" prompt produces generic criticism. Targeted prompts produce targeted intelligence.
Before: You send a board paper or investment proposal for internal review and wait several days for feedback — often receiving comments that are polite rather than rigorous.
After — try this prompt: "Here is my proposal for [X]. Please argue the strongest case against it across three dimensions: (1) financial risk and assumptions I may have got wrong, (2) execution and operational risk, and (3) strategic risk — including competitive responses or market shifts I may have underweighted. Be direct and specific. Do not soften the critique."
The output will not always be right. But it will raise questions worth answering before the room does.