SNI: WEEK 14

Welcome to all the AI news that matters this week – across tech, biopharma, medtech, complex manufacturing and insurance. The wins, the fails and the somewhere in-betweens.

tl;dr: The regulators have arrived

The first quarter of 2026 closed with regulatory bodies no longer being the audience - moving instead into leading roles. As they move off AI sidelines, examination frameworks are being applied, legislative gaps are being pressed and companies are being compelled to submit AI systems for direct scrutiny.

The most concrete actions last week were in US insurance. A 12-state pilot requires carriers to submit AI claims systems for regulatory review – the first formal examination of insurer AI behind total-loss decisions and claims payouts. Separately, Colorado's auto and health insurers will have to submit annual AI compliance reports from 1 July.

In Europe, EIOPA flagged that the Insurance Distribution Directive does not fully regulate AI advice models – a realisation that the legislative framework has fallen behind deployment. But the UK's FCA published its 2026 regulatory priorities, with AI oversight threaded across all its eight sectors.

In biopharma, this week's pressure point was model provenance. The FDA is scrutinising AI model migration after federal directives restricted the use of Claude within Elsa, the agency's internal clinical-review tool. The episode exposes a novel compliance risk: when a regulator's own AI tooling is forced to switch foundation models, what of validation continuity and decision traceability? Will the change ripple out to every submission the tool touches?

But, for all the uncertainty created globally, there is no sign that the pace of AI sector will slow. Capital continues to pour in at an extraordinary scale.

• Global venture funding hit $300bn in Q1 2026, with AI companies absorbing $242bn – 80% of the total.

• Eli Lilly signed a $2.75bn generative-chemistry alliance with Insilico Medicine, the largest AI drug discovery deal to date.

• SoftBank deployed the first $10bn tranche of its OpenAI commitment.

So, no surprise that the semiconductor supply chain, the physical foundation of every AI system, is showing visible strain.

TSMC capacity constraints are opening a window for Samsung in advanced packaging, and Intel has confirmed CPU price hikes of 10–15% for OEMs as AI data centre demand stretches lead times from weeks to months.

Here are the other headlines you might want to get across:

AI & Tech

Anthropic accidentally exposed the source code of Claude Code, revealing the extent of system-level data collection the tool performs on users' machines and placing data governance at the centre of enterprise procurement decisions

Chinese domestic chipmakers captured 41% of China's AI accelerator market, pushing Nvidia's share below 60% and splitting the global chip supply chain in ways that affect model portability and inference costs for multinationals

Oracle announced 30,000 redundancies while accelerating investment in cloud and AI data centre infrastructure, signalling that the workforce composition of a cloud-and-AI business differs materially from a legacy database enterprise

Biopharma

Pharma Mar partnered with Globant to deploy a multi-agent AI platform for cancer drug discovery, pairing its marine-organism compound library with Globant's AI engineering in a pattern increasingly common among mid-cap European pharma companies

Medicines for Malaria Venture and deepmirror launched dd4gh, a free AI drug discovery platform for researchers working on malaria, tuberculosis and neglected tropical diseases, lowering the barrier to AI-assisted compound design in areas where commercial incentives are weak

Mantis Biotech is training large language models on genomics and clinical data to create digital twins of human biology, generating synthetic patient data that could reduce both cost and timeline in early-stage drug development

MedTech

Anumana secured FDA clearance for the first AI algorithm to detect pulmonary hypertension from standard 12-lead ECGs, establishing a regulatory predicate that subsequent applicants must match or exceed

Mount Sinai deployed the OpenEvidence AI platform enterprise-wide inside its Epic EHR, extending clinical decision support beyond physicians to nurses and pharmacists and reducing adoption friction by embedding AI inside existing workflows

Canada's Office of the Information and Privacy Commissioner raised concerns about ambient AI scribes, citing privacy, accuracy and bias risks - a signal that regulators in other jurisdictions may impose disclosure or consent requirements on ambient listening tools

Jimini Health raised $17m to launch Sage, an AI mental health chatbot integrated within clinical teams rather than offered direct-to-consumer, deliberately avoiding the regulatory and liability exposure that has affected consumer mental health apps

Avo raised $10m in Series A funding to build a clinical AI platform constrained to verified medical knowledge, addressing the trust deficit that persists in clinical AI by reducing hallucination risk through curated content

Complex Manufacturing

Agile Robots completed its acquisition of thyssenkrupp Automation Engineering's European and North American assets, combining software intelligence with hardware deployment capability to offer European manufacturers a full-stack AI robotics proposition

TerraPower and SoftServe unveiled an AI-powered digital twin for the Natrium advanced nuclear reactor to reduce early-stage engineering design cycles in one of the most heavily regulated sectors

Cognichip raised $60m to build a deep learning model that assists engineers during chip design, targeting design cycles that currently stretch 18 months or longer and, if validated, shortening time-to-market for custom AI accelerators

Intel raised CPU prices by 10–15% for OEMs as AI data centre demand stretches lead times from weeks to months, with some orders now taking up to six months to fulfil - a cost increase that flows downstream to hospitals, insurers and manufacturers deploying on-premises AI

Global memory chip shortages have shifted industry focus from price competition to securing supply, pushing up component costs across consumer electronics and enterprise hardware

Insurance

EIOPA's third report on the Insurance Distribution Directive identified a regulatory gap around generative AI in digital channels, signalling that EU-level action is being prepared and giving insurers a limited window to shape the coming rules through consultation

Swiss Re flagged data centre risk accumulation as a new class of insurance challenge, with single-site construction costs exceeding $20bn and existing actuarial models unequipped to capture concentration exposure, correlated failure and business-interruption risk at this scale

Ping An now automates nearly 60% of accident and health claims, with some settled in 51 seconds - the result of a decade of investment worth billions of yuan - setting a customer-experience benchmark that competitors can expect to be measured against

BizCover became the first business insurer to offer quotes directly inside ChatGPT, allowing Australian small business owners to compare and bind coverage through a conversational AI interface and signalling potential disintermediation risk for brokers and price-comparison intermediariesBut what set podcast tongues a-wagging?

But what set podcast tongues a-wagging?

Agentic systems are arriving faster than enterprise can absorb them

Nathaniel Whittemore named Q1 2026 'AI's second moment' - the shift from capable chatbots to working agentic systems - and offered survey data on the enterprise adoption gap. Of Whittemore's power-user audience, 97% use AI daily and more than 60% are running agentic or automation workflows. His cohort is not waiting for enterprise purchasing approval either. Official procurement is running on an entirely different clock.

Amjad Masad, speaking with Marc Andreessen on the a16z podcast on 30 March, gave us data to explain why Whittemore's audience is adopting ahead of the enterprise curve. The length of time that Replit's agent can run independently has been doubling approximately every seven months: Agent 1 ran for 2 minutes, Agent 2 for 20 minutes, Agent 3 for 200 minutes, with Agent 4 expected at 2,000 minutes. Whilst these figures are to be taken with a pinch of salt, there's no question that agents are increasingly capable of substantially longer tasks. Which connects directly to the continued repricing of SaaS stocks. When agents can run independently all day, annual per-seat licencing may seem like a very odd model.

Alex Kantrowitz and Ranjan Roy on Big Technology, highlighted one reason why procurement teams have reason to pause before enterprise adoption. Framing OpenAI's Sora shutdown as deliberate resource reallocation, the podcast hosts identified the portfolio management logic: OpenAI is concentrating on coding and enterprise contracts, and products that do not contribute to that revenue story are being cut. Great for OpenAI. But those relying on the frontier lab's product only received six weeks' notice of a shutdown. Vendor continuity is still far from standard in AI.

Grid architecture is the constraint cloud pricing does not capture

The founders of Radiant Nuclear and Heron Power, speaking with a16z on 31 March, noted a key distinction: the binding infrastructure constraint in the AI buildout is transmission and distribution, ahead of generation. New power plants can be permitted, financed and built on a five-to-seven-year cycle. Grid transmission upgrades - the physical architecture that moves power from plant to data centre - operate under different constraints. Legacy infrastructure replacement, right-of-way acquisitions and interconnect queues can extend up to ten years. And are subject to regulatory processes that investor urgency cannot compress.

The implication is geographic in a way that hyperscaler pricing sheets conceal. Cloud regions near grid-constrained corridors will face capacity limits and cost escalation. The 89% pre-lease rate for capacity-under-construction proves that supply, not demand, is already the constrained variable.

Thank you for reading this week's report. Come back next week for all the AI news you need to know in your sector and beyond.