Tech Digest – April 7, 2026
The Compute Economy
Samsung Posts Record $38 Billion Quarter as Anthropic Revenue Triples and OpenAI Plans $600 Billion in Five-Year Spending
Samsung Electronics reported a preliminary Q1 operating profit of 57.2 trillion won (~$38 billion), an eightfold year-over-year increase driven by AI memory chip demand so explosive it triggered shortages across the market. Revenue rose 68% to $88.7 billion, comfortably clearing analyst expectations. The profit is roughly three times Samsung’s previous quarterly record, set just last quarter.
On the buyer side, Anthropic disclosed that its run-rate revenue has leapt from roughly $9 billion at end of 2025 to over $30 billion, while signing a 3.5-gigawatt TPU capacity deal with Google and Broadcom — coming online from 2027 on top of 1 GW already in use. The company now has over 1,000 enterprise customers each spending more than $1 million annually, a figure that has more than doubled since February.
OpenAI is scaling even more aggressively, planning to spend $121 billion on compute in 2028 alone while burning $85 billion that year. CEO Sam Altman has committed to $600 billion in five-year spending and is eyeing a Q4 IPO, though CFO Sarah Friar has signalled internally that the company may not be ready. OpenAI recently closed a $122 billion funding round at an $852 billion valuation.
Note: Three numbers frame the scale: $38 billion in quarterly chip profits, $30 billion in annual AI lab revenue, $600 billion in forward spending commitments. Each would have been unthinkable two years ago. Any institution budgeting for digital infrastructure is operating in a market whose price signals are being set by these capital flows — whether or not they buy AI directly.
Sources: Reuters, CNBC, Anthropic, Bloomberg, Wall Street Journal
Autonomous Intelligence
AI Runs 50 Experiments in 72 Hours and Invents a Memory System That Beats Every Human Baseline
UNC researchers gave an AI system 72 hours of autonomous operation to conduct research on long-context memory. The system ran 50 experiments without human intervention and produced a novel memory architecture that outperforms every human-designed baseline on the task. The preprint demonstrates that the full research loop — hypothesis generation, experimental design, execution, and evaluation — is becoming automatable as a continuous, unsupervised process.
Note: The frontier just moved past “AI assists researchers” to “AI is the researcher.” Seventy-two hours and 50 experiments is a pace no human lab sustains. For institutions funding R&D or setting research programme timelines, the productivity denominator has changed.
Sources: arXiv (2604.01007)
Security Economics Rewritten
AI Labs Share Intelligence Against Chinese Model Theft as Bug Bounty Programmes Buckle Under Discovery Speed
OpenAI, Anthropic, and Google have begun sharing attack data through the Frontier Model Forum to detect and counter Chinese distillation attempts — where competitors systematically query frontier models and use the outputs to train cheaper copies. Anthropic documented 16 million exchanges from DeepSeek, Moonshot AI, and MiniMax. US officials estimate the practice costs Silicon Valley labs billions annually. The rare industry collaboration reflects how acute the intellectual property threat has become.
At the other end of the security stack, the Internet Bug Bounty programme — running since 2012 and backed by Meta, GitHub, Shopify, and TikTok — paused new submissions on March 31. The reason: AI-assisted vulnerability discovery has expanded so rapidly that the balance between findings and remediation capacity has “substantively shifted.” Node.js has already stopped paying bounties, and Google halted AI-generated submissions to its open-source vulnerability programme last month.
Note: On one front, AI makes it too easy to copy models. On another, too easy to find bugs. The security economy assumed human-speed discovery on both sides of the ledger. When AI accelerates offence and defence simultaneously, the pricing models — for IP protection and for vulnerability remediation alike — break.
Sources: Bloomberg, HackerOne, InfoWorld
Infrastructure Meets Neighbourhood
Indianapolis Councilor’s Home Shot Up 13 Times Over Proposed Data Centre
An Indianapolis city councilor reported that his home was shot at 13 times over a proposed neighbourhood data centre, with a note left reading “NO DATA CENTERS.” The incident is the most extreme example yet of physical backlash to data centre siting as AI infrastructure pushes build-outs into residential areas. Roughly half of US data centres planned for 2026 already face delays from component shortages — the projects that do advance are increasingly meeting community resistance that goes well beyond planning objections.
Note: The cloud has a postal address. Data centres need land, power, water, and community consent — and AI-driven demand is forcing them into new geographies faster than local planning processes can absorb. European municipalities evaluating data centre proposals are watching this dynamic arrive.
Sources: CBS News
National Strategies in Motion
South Korea Deploys Thousands of AI Companion Dolls for Its Elderly as Japan Targets 30% of the Global Physical AI Market
South Korea is distributing thousands of ChatGPT-enabled companion dolls to elderly citizens living alone — now roughly 20% of the population. The Hyodol doll, distributed through government welfare programmes, monitors activity via infrared sensors, reminds users to eat and take medication, and provides conversational companionship. Over 12,000 units are deployed, with studies showing measurable reductions in depression and dementia symptoms. The company is adapting the product for international markets in 2026.
Japan’s METI, meanwhile, has set a target of capturing 30% of the global physical AI market by 2040 — an industrial policy signal that positions robotics and embodied AI as a national strategic priority alongside semiconductors and energy.
Note: Both countries face populations aging faster than workforces can replace them. South Korea deploys conversational AI as care infrastructure; Japan targets manufacturing dominance in physical AI. Two complementary bets on how automation addresses demographic decline — and a question for EU member states on similar demographic curves with no comparable deployment programme.
Sources: Financial Times, TechCrunch
China Flies World’s First Megawatt-Class Hydrogen Turboprop
A 7.5-tonne unmanned cargo aircraft powered by China’s AEP100 megawatt-class hydrogen turboprop engine completed a 16-minute maiden flight in Zhuzhou, covering 36 km at 220 km/h and 300 metres altitude. The test establishes that China now commands the complete technology chain for hydrogen-fuelled aviation engines, from core components to full integration. Initial applications target unmanned cargo and island logistics, with regional passenger aviation as the next horizon.
Note: The EU’s ReFuelEU Aviation regulation mandates sustainable aviation fuel adoption from 2025 onward. China is flight-testing the propulsion hardware. Sixteen minutes is a proof of concept — but the distance between “proven physics” and “commercial deployment” is where institutional procurement timelines should start.
Sources: CGTN, Fuel Cells Works
Policy for the Intelligence Age
OpenAI Proposes Robot Taxes, a Public Wealth Fund, and Four-Day Workweek Pilots
OpenAI published an “Industrial Policy for the Intelligence Age” proposing automated-labour taxes, a public wealth fund that gives every citizen a direct stake in AI-driven growth, and subsidised 32-hour workweek pilots with no loss in pay. Sam Altman framed it as a new social contract “on the scale of the New Deal,” calling for a rebalancing of the tax base from labour income to capital as corporate profits expand and reliance on human wages shrinks.
The automation OpenAI proposes to govern is already visible in cultural production: AI-assisted stories drive nearly 20% of Fortune’s web traffic, and an AI-generated singer holds 11 simultaneous slots in the iTunes top 100. Yet the picture is more complex than displacement alone — tech job openings have doubled since mid-2023 to a three-year high, and one startup is marketing an agent layer that operates fleets of microbusinesses for a single human owner, suggesting AI may create new economic structures as fast as it disrupts old ones.
Note: The company building the automation is now proposing the safety net. The policy specifics — capital gains taxes, a public wealth fund, 32-hour workweek pilots — are concrete enough to enter real legislative conversations. EU policymakers revising employment provisions under the AI Act will find a ready-made comparison framework, whether they agree with it or not.
Sources: OpenAI, Axios, TechCrunch, Wall Street Journal, Business Insider
Today’s digest traces one arc: the AI economy has reached a scale where its infrastructure, its security model, and its social contract are all under stress simultaneously. Samsung’s record quarter and OpenAI’s $600 billion spending plan show the capital pouring in. Bug bounty programmes pausing and a councilor’s home under fire show the systems that weren’t built for this pace. East Asian governments are already deploying — companion dolls in Korean homes, hydrogen engines over Chinese airfields, industrial AI targets in Japanese policy. And OpenAI itself, the company planning to spend more on compute than most nations spend on defence, is now drafting the tax and labour policy to manage what comes next. The institutions that adapt will be the ones that recognised this wasn’t a technology shift but a structural one.
