SNI: WEEK 20

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

tl;dr: The public boos, the money cheers

It's not been the best of weeks for Irish AI sentiment. Research revealed half of workers are expected to use AI without enough training. And the Irish Times argued women could be the losers in the workplace transition. This was in addition to the 720 Covalen workers taking strike action against AI, while LinkedIn cut 900 jobs - although only 78 are at risk in Ireland - as parent Microsoft restructures around AI. See more on our take here.

AI's public profile also hit a new low as a Gallup poll explained that 70% of Americans oppose new data-centres - making AI less popular than nuclear power.

So it might have been fortunate that Figure AI's livestreaming of humanoid robots running autonomous warehouse ops didn't make it into the headlines.

But, of course, the spoken-word commentators were eager to redress the balance. Nathaniel Whittemore on the AI Daily Brief spent 15 minutes explaining how AI will create more jobs than it destroys. He also hypothesised they'll be one human care-giver per 100-150 patients – scaling to millions of net new jobs globally. And Marc Andreessen was equally upbeat - claiming leading-edge programmers are now 20x more productive.

Capital couldn't have nodded more vigorously. Cerebras priced its IPO at $185, opened at $350 and closed at $311.07 – a 68% first-day pop. More than $5.5bn was raised against a $95bn closing market cap. The largest US tech listing since Uber in 2019.

Almost as significant was Long Lake's $6.3bn take-private of American Express Global Business Travel - to reinvigorate the business with AI.

Dublin also got a lift from Enterprise Ireland's €21m National Accelerator, as KPMG opened its EU AI Hub – a bet on multi-year EU AI Act compliance work.

On the pharma side, Isomorphic Labs raised $2.1bn in he largest AI drug-discovery raise on record. Purely philanthropically, Anthropic and the Gates Foundation formed a $200m AI pact for health and education in low- and middle-income countries.

But it wasn't all upside for Dario's crew. Anthropic enraged the developer community with subscription changes. And, in an unrelated decision, Microsoft cancelled its own Claude Code licences.

Which made Anthropic's explanation that the model's blackmail behaviour traces back to AI-doomer literature seem even bleaker.

So let's change the mood - with everything else worth reading this week:

AI & tech:

• MIT Technology Review frames data sovereignty as the AI question of the decade.

• Thomson Reuters publishes its 2026 AI in Professional Services Report: the survey marks the strategic pivot from generative to agentic across law, accountancy and consulting.

• Nvidia has committed $40bn to equity AI deals this year: the chipmaker is now the largest single equity-side investor in the AI capital stack and a counterparty to most labs.

• Microsoft restructures its OpenAI relationship again: the two-track strategy of consuming OpenAI capacity while building in-house models continues to reshape the most-watched relationship in enterprise software.

Biopharma:

• Owkin and AstraZeneca expand their AI-driven drug-research collaboration: federated learning now runs across AZ's data without leaving its perimeter.

• The FDA rolls out Elsa 4.0 and a new HALO Data Platform for AI-augmented reviews: the agency now runs its own AI on a platform that reviewers can query across submissions.

• Cellular Intelligence licenses Novo Nordisk's shelved Parkinson's cell-therapy programme: an AI-native TechBio picks up a clinical-stage asset Novo wound down seven months ago.

• IBM releases its MAMMAL model for open AI drug discovery: the release lands a Big Tech entry in the drug-discovery market - with no licensing fees.

• Veristat launches an AI biostatistics platform cutting data readout from 5 weeks to 5 days: the also company packages the speedup as regulator-aligned.

• The Scientist maps how AI is de-risking drug development and companion diagnostics: the feature anchors the advances inside the $2bn-per-development costs and near-decade timelines debate.

Medtech:

• Artera receives FDA clearance for AI breast-cancer prediction tool: following hard on the heels of Bayesian's sepsis monitor.

• Trump and Kennedy push to relax federal safeguards on AI healthcare tools: the proposed loosening of oversight runs counter to the BMA position on NHS data and to Iowa's Patients First Act, highlighting a growing divergence between policy trajectories.

• Corti opens its clinical-AI stack to startups: the Copenhagen company says its outscored OpenAI on HealthBench Professional and is offering startup credits with regulatory help bundled in.

• The NHS rolls out an AI skin-cancer detection tool across West Yorkshire: a live regional deployment of a clinical-AI workflow lands at exactly the moment the BMA is asking who controls the models running over NHS data.

Advanced manufacturing:

• Waymo recalls 3,800 robotaxis after one drove itself into a flood: a software-only recall, recategorising firmware faults as physical-product recalls.

• Japan opens 'world's first fully automated medicine lab', with humanoids and other robots: the facility lands the lights-out thesis in a regulated domain rather than a warehouse.

• Microsoft deploys Azure Foundry Copilot in factories: the four-customer ARUM, Cemex, Beca and Obeikan announcement installs industrial generative AI inside cement, automotive and packaging plants.

• Infineon's AI revenue pipeline hits €1.5bn: the German chipmaker pairs the milestone with strategic stakes in robotics startups, broadening its mix beyond chip manufacturing.

Insurance:

• Naked Insurance launches the first ChatGPT app to issue binding insurance quotes: distribution leaps to the assistant layer with a binding-not-indicative quote. UK broker share prices fell on the news.

• Iowa signs the Patients First Act, banning health insurers from using AI to determine prior-authorisation denials: the first US state to require human review of every AI-generated coverage decision.

• HDI Global partners with mea Platform on global AI underwriting and claims: the insurer standardises AI submission and claims processing across geographies.

• Moody's analysts warn AI liability uncertainty and data-centre risks are repricing insurer exposure: a rating-agency read on AI exposure inside commercial lines, pairing with a Tufts analysis on whether AI risks can be insured at all.

But what set podcast tongues a-wagging?

The token subsidy era is ending.

Nathaniel Whittemore on the AI Daily Brief calls Anthropic's pricing change the visible end of the AI subsidy era. The six-month window in which a $200 Claude Max plan delivered $2,000-$5,000 of API-equivalent compute is closing for usage outside Anthropic's own Claude Code harness.

Why? Because compute is now a binding constraint - and Anthropic is choosing to ration subsidy toward enterprise-routed traffic. OpenAI will likely follow.

The chip-cost bargain has broken.

Azeem Azhar on Exponential View argues that Bloomberg's report of TSMC declining to deploy ASML's chipmaking tools – on cost grounds – may be the first concrete signal that the economic side of Moore's Law has reversed.

Cost per transistor stopped falling in 2011; the narrower transistors-per-wafer-dollar measure is now turning too.

But capital is still backing the multiplier play.

Long Lake CEO Alexander Taubman makes the case to No Priors for AI-driven roll-ups of services businesses, using the announced $6.3bn take-private of Amex GBT as the hook. The Long Lake thesis takes the rising compute bill as a given and routes around it: do not run a cost-cutting playbook, give frontline employees AI superpowers, and capture the productivity uplift inside the operating company.

The operational evidence supporting the thesis appears impressive. Long Lake's HOA management business has moved from a 0-5% growth baseline to 20% annually post-AI, with employee retention rising sharply, customer retention rising, and pay positioned to rise with marginal productivity.

Taubman estimates AI penetration in the services industries on Long Lake's target list at roughly 1%, and the addressable TAM at over $20tn. Which is something worth thinking about this weekend.

Thank you for reading the latest report. Come back next week for all the AI news you need to know in your sector.