Tech Digest – June 1, 2026
Capital & Infrastructure at Scale
SoftBank Commits €75 Billion to 5 Gigawatts of French AI Data Centres
SoftBank announced plans to invest up to €75 billion (~$87 billion) in AI data centre capacity across France, targeting 5 gigawatts in total. The first phase — €45 billion for 3.1 GW across Dunkirk, Bosquel, and Bouchain — is to deliver in Hauts-de-France by 2031, with Schneider Electric as engineering partner. The deal followed direct diplomacy between President Macron and SoftBank founder Masayoshi Son, and represents SoftBank’s largest AI infrastructure investment in Europe.
Note: €75 billion is nearly ten times the entire Digital Europe Programme budget. For municipalities in northern France, this means construction demand, grid load, and land-use decisions arriving whether or not their own digital strategies are ready.
Sources: Bloomberg, TechCrunch, SoftBank
Samsung, SK Hynix, and Micron Each Pass $1 Trillion — Memory Is Now Worth More Than Oil
The three dominant memory chipmakers have each crossed $1 trillion in market capitalisation, together worth roughly 22% more than the three largest oil companies. The Wall Street Journal’s verdict: “memory is now worth more than oil.” The shift reflects relentless demand for high-bandwidth memory driven by AI training and inference workloads.
The gravitational pull is visible beyond balance sheets. Chinese tech firms now charge up to $9,000 per person for curated tours of EV plants, robotaxi fleets, and humanoid labs, as global observers bet the centre of gravity has shifted east.
Note: When the commodity class underpinning your digital infrastructure outvalues petroleum, supply-chain assumptions from three years ago are obsolete.
Sources: WSJ, Rest of World
The AI Industry Goes Institutional
Anthropic Files Confidentially for IPO — Revenue Quintuples to $47 Billion Run-Rate
Anthropic submitted a confidential S-1 to the SEC on June 1, positioning for a public listing as soon as this fall. The company’s revenue run-rate has surpassed $47 billion, up from $9 billion at the end of 2025 — a fivefold increase in roughly 18 months. Two days ago in this digest, Anthropic raised $65 billion at a $965 billion valuation. Now it’s racing to beat OpenAI to Wall Street.
Note: The S-1, when made public, will disclose customer concentration, compute costs, and margin structure. For any institution evaluating Claude as a vendor, that filing will be the most informative document published this year.
Sources: Bloomberg, CNBC, TechCrunch
Anthropic and OpenAI Fund Rival Super PACs With $124 Million — Neither Mentions AI in a Single Ad
Public First, backed by a $20 million Anthropic donation, and Leading the Future, funded with over $100 million from investors including Andreessen Horowitz and OpenAI president Greg Brockman, are flooding US midterm races — particularly Democratic primaries. Public First backs candidates favouring stronger AI oversight; Leading the Future targets those most identified with regulation. Neither PAC mentions artificial intelligence in its advertisements, running instead on immigration, healthcare, and opposition to Trump.
Note: The AI industry is now spending politically at pharmaceutical-lobby scale — and using the same playbook of running on everything except the issue it exists to influence. The policy environment shaping how EU institutions can procure and deploy AI tools is being set, in part, by ad buys in New York congressional districts.
Sources: NYT
Robots Enter the Battlefield
Humanoid Robots Deployed to Ukraine — First Confirmed Combat-Theatre Use
San Francisco-based Foundation Future Industries sent two Phantom MK-1 humanoid robots to Ukraine in February — the first confirmed deployment of humanoid robots in a combat theatre. The 1.7 m/s bipeds carried 20 kg payloads using proprietary cycloidal actuators and eight-camera navigation. Initial testing exposed critical gaps: no waterproofing, insufficient battery endurance, and limited operational suitability. The company, which holds $24 million in Pentagon contracts, plans an upgraded Phantom 2 later this year with doubled payload capacity.
Separately, OpenAI entered the physical AI space. Sam Altman announced a new robotics division hiring engineers to build robots that “first help skilled workers raise AI infrastructure and eventually serve everyone.”
Note: The Phantom MK-1 failed most battlefield tests — but it was there. The precedent matters more than the performance. Defence procurement offices across Europe now have to evaluate a category that didn’t exist twelve months ago.
Sources: CNBC, Sam Altman (X)
Energy for the Build-Out
MIT Halves Lithium Cost from Hard Rock; White House Converts Warhead Plutonium to Reactor Fuel
MIT researchers developed a low-temperature process that extracts battery-grade lithium from hard rock at roughly half the conventional cost, potentially unlocking American deposits previously considered uneconomical. Separately, the White House transferred over 50 tons of weapons-grade plutonium from dismantled Cold War warheads to five private firms selected to produce commercial reactor fuel.
Note: Both moves address the same bottleneck from opposite ends — battery storage and baseload power for the compute infrastructure above. The US is treating energy supply as an AI prerequisite, not an afterthought. EU Digital Decade targets assume energy availability that isn’t guaranteed without equivalent urgency.
The Invisible Economy
AI ‘Dark Output’ May Already Be the Invisible Majority of Economic Activity
SemiAnalysis introduced the concept of “dark output” — AI-generated economic value invisible to national statistics. The analysis identifies a growing mismatch: 37% of AI token usage flows into computing and mathematics sectors, but software investment’s contribution to GDP has not broken from its pre-AI trend. “New dark output” includes work that no firm would have paid a human to perform at prevailing prices — tasks AI has made cheap enough to do that never existed as measurable economic activity.
Note: If the fastest-growing category of output doesn’t register in GDP, the productivity statistics guiding institutional budget decisions are measuring the wrong economy.
Sources: SemiAnalysis
Public Health & Biosecurity
OpenAI Launches Rosalind Biodefense — National Labs and CEPI Get Frontier AI for Pandemic Defence
OpenAI opened its GPT-Rosalind life-sciences model to vetted partners through Rosalind Biodefense, running on two tracks: a developer programme for epidemiological modelling, early detection, and screening tools, and a government track extending access to US federal agencies and allied partners. Initial partners include Lawrence Livermore National Laboratory for medical countermeasure design, Johns Hopkins Applied Physics Laboratory for protein-engineering screening, and the Coalition for Epidemic Preparedness Innovations (CEPI) for rapid vaccine development against emerging threats.
Note: CEPI is co-headquartered in Oslo and London. The “allied partners” language suggests this won’t stay US-only — EU public health agencies should watch whether Rosalind access extends through existing transatlantic health security frameworks.
€75 billion in European data centres, trillion-dollar chip companies, a near-trillion-dollar IPO, robots in a war zone, and economic output that national statistics can’t see. Today’s throughline isn’t AI capability — it’s institutional velocity. Capital, infrastructure, political machinery, and physical deployments are scaling at rates that compress planning horizons from years to quarters. For anyone working to a 2030 roadmap, the question is no longer whether to revise it — but whether the world it describes still exists.
