Tech Digest – April 26, 2026
AI Sovereignty & Procurement
DeepSeek V4 Is the Largest Open-Weights Model — and It Runs on Chinese Chips
DeepSeek released V4 on April 24 in two variants: V4-Pro (1.6 trillion total parameters, 49 billion active) and V4-Flash (284 billion total, 13 billion active), both under an MIT open-source licence with a 1-million-token context window. Pro is now the largest open-weights model available, surpassing Kimi K2.6 and GLM-5.1. Pricing undercuts every competitor in its class — Flash at $0.14 per million input tokens, Pro at $1.74.
The bigger story is what delayed the release. Bloomberg reports that DeepSeek spent months reworking its software stack to run natively on Huawei’s Ascend AI processors — not just basic compatibility, but hardware-specific optimisation. MIT Technology Review called V4 “an early sign that China is successfully building a parallel AI infrastructure.” Germany banned DeepSeek from app stores in 2025 over illegal data transfers to China, and the data-sovereignty question will follow V4 everywhere — but the capability and price are now hard to ignore.
Note: For any European institution evaluating AI procurement, V4 creates an uncomfortable split: the cheapest, most capable open model on the market is also the one most entangled with a sanctioned chip ecosystem and an unresolved data-sovereignty question. The procurement case and the compliance case point in opposite directions.
Sources: CNBC, Bloomberg, MIT Technology Review, Simon Willison
Governance & Compliance Countdown
EU Institutions Face Back-to-Back AI Deadlines — and the Rulebooks Still Conflict
The AI Act’s high-risk system requirements become enforceable on August 2, 2026, with penalties up to €15 million or 3% of worldwide turnover. Five weeks later, the Cyber Resilience Act begins to apply on September 11, imposing mandatory reporting of actively exploited vulnerabilities for any product with digital elements. European financial regulators warned on April 25 that AI is accelerating both the speed and sophistication of cyberattacks, increasing systemic risk across markets.
Meanwhile, the IAPP published a detailed mapping of where the AI Act and the GDPR collide — processing special categories of data for AI bias detection, logging requirements that may conflict with data subject rights, and sandbox testing that still requires a GDPR legal basis. The Digital Omnibus package currently in trilogue could change both frameworks, but nothing has passed into law. The deadlines are fixed; the rules they sit on are still moving.
Note: August and September 2026 land 100 days apart. Any institution running AI in a high-risk context — hiring, benefits, public services — needs compliance architecture in place before summer. Waiting for the Omnibus to resolve the GDPR friction is a bet that trilogues move faster than enforcement dates. They rarely do.
Sources: IAPP, Help Net Security, Inside Privacy
UN Launches AI Scientific Panel as Global Market Heads Toward $4.8 Trillion
The UN’s Independent International Scientific Panel on AI held its first in-person meeting in Madrid this week, with co-chair Maria Ressa warning that AI tools are accelerating the undermining of democratic systems through “narrative warfare.” Geoffrey Hinton, speaking at the UN Digital World Conference on April 22, compared ungoverned AI to “a very fast car with no steering wheel.” UNCTAD projects the global AI market will grow from $189 billion in 2023 to $4.8 trillion by 2033 — an economy larger than Japan’s.
Note: $4.8 trillion by 2033 isn’t a forecast — it’s the scale of the governance gap. The panel that’s supposed to help steer this just had its first meeting.
Sources: UN News
Workforce Restructuring
Meta Cuts 8,000 Jobs, Snap Cuts 1,000 — Savings Go Straight to AI Compute
Meta is laying off 10% of its workforce — approximately 8,000 employees — and scrapping 6,000 open roles, redirecting savings toward a planned $135 billion in AI spending for 2026. Snap followed on April 15, cutting roughly 1,000 employees (16% of staff) and projecting $500 million in annualised savings. CEO Evan Spiegel cited “rapid advancements in artificial intelligence” that allow smaller teams to achieve the same output. AI now generates more than 65% of Snap’s new code.
Note: Snap’s 65% figure is the quiet part out loud. When two-thirds of new code is machine-written and the stock jumps 8% on the layoff announcement, the market has priced in a new headcount model. The question for every organisation writing job descriptions right now: which of these roles will exist in 18 months?
Sources: CNBC (Meta), CNBC (Snap), TechCrunch
Stanford AI Index: Productivity Up 26% in Software — but Young Developer Employment Down 20%
Stanford’s 2026 AI Index reports that AI is boosting productivity by 14% in customer service and 26% in software development, though gains remain limited in tasks requiring deeper judgment. The flip side: employment for software developers aged 22 to 25 has fallen nearly 20% since 2022. A third of organisations surveyed by McKinsey expect AI to shrink their workforce within the year. The Index also finds that people are adopting AI faster than they adopted the personal computer or the internet. As of March 2026, Anthropic leads on the Arena leaderboard, followed by xAI, Google, and OpenAI.
Note: A 26% productivity gain and a 20% employment drop in the same profession, in the same timeframe, is not a coincidence — it’s the mechanism. Entry-level knowledge work is being compressed from both sides: the people already in the role produce more; the roles that would have absorbed new graduates don’t open.
Sources: MIT Technology Review (Stanford AI Index)
Capital Concentration
Q1 2026 Venture Funding Hits $300 Billion — AI Claims 80% of Every Dollar
Global venture investment in Q1 2026 reached approximately $300 billion across 6,000 startups — more than 150% above any previous quarter and close to 70% of all venture capital deployed in the whole of 2025. AI accounted for $242 billion, or 80% of the total, up from 55% a year ago. Ten mega-rounds of $2 billion or more drove the bulk, led by OpenAI ($122 billion), Anthropic ($30.6 billion), and xAI ($20 billion).
Anthropic has now passed OpenAI in annualised revenue — reaching $30 billion ARR in April 2026 versus OpenAI’s $25 billion, a roughly 1,400% year-over-year increase driven primarily by enterprise customers. The revenue crossover, combined with the capital concentration, signals that institutional buyers are shaping the market as much as consumers.
Note: When 80% of global venture capital flows into one sector in a single quarter, it stops being an investment trend and starts being an infrastructure buildout. The parallel to fibre-optic cable in the late 1990s is instructive — the capital was concentrated, much of it was wasted, but the infrastructure it left behind reshaped everything.
Sources: Crunchbase, KPMG, TechCrunch
Three threads run through today’s digest and they pull in the same direction. China ships a frontier model on domestic chips while Europe debates where the GDPR ends and the AI Act begins. Venture capital pours $242 billion into AI in a single quarter while Meta and Snap cut 9,000 jobs to fund the next round. Stanford measures a 26% productivity gain in the same profession where entry-level employment is falling 20%. The pattern is consistent: the infrastructure is being built, the workforce is being reshaped, and the governance is still catching up. For institutions planning on a two-to-three-year horizon, the landscape they’re planning for is already behind them.
