SNI: WEEK 16

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 weight of scale

This week, AI got very physical.

Sam Altman's home was targeted twice - a Molotov cocktail then gunfire  - with prosecutors treating the arson as potential domestic terrorism. Altman's ‘I don’t want the ring of power’ response did not necessarily make everyone feel safer. And the violence wasn’t isolated: an Indianapolis city councillor had 13 rounds fired at his door because of his support for data centres.

The attacks arrived at an already-charged moment. OpenAI accused Elon Musk of 'legal ambush' after he changed his objectives weeks before the $100B+ damages trial. And while that goes on, OpenAI's backers are questioning its $852B valuation.

The physical infrastructure commentary this week was also decidedly aggressive. Nvidia was accused of ‘spraying’ capital around like a sprinkler for control over the industry. And Anthropic's multi-gigawatt deal with Google and Broadcom for custom TPUs was portrayed as a deeper land grab into the AI supply chain.

And, we were asked, is the sector’s ring of power poisoned anyway? Gartner expects more than 2,000 wrongful death claims linked to AI by end of 2026. News which dropped as insurers began writing AI exclusions into commercial polices.

Regulatory uncertainty also added weight. A proposed amendment to the EU's Digital Omnibus would push the AI Act's high-risk compliance deadline by 12 to 24 months. For compliance teams already invested in EU AI Act readiness, the delay is a disruption rather than a relief. The obligations do not disappear.

Instead, as we noted in one of our two keynotes at Manufacturing the Future, it shifts the burden of work - and complicates the patchwork of timelines moving at different speeds.

But not everything was uncomfortable. Here's our list of other happenings that are worth reading this week:

AI & tech:

OpenAI acquired Hiro Finance in a push into vertical AI for personal finance: the first frontier model company to move into a regulated financial-services vertical, with compliance and data-governance implications that differ from consumer AI.

Vibe coding won a $6.6bn industry valuation, 14 months after Andrej Karpathy's throwaway tweet: Cursor, Replit and GitHub Copilot account for the bulk of the market; enterprise adoption is now the primary commercial driver.

Biopharma:

Agentic AI moves into post-approval commercial pharma workflows: AI agents are restructuring where the $500m of typical launch spend goes.

Bristol Myers Squibb partnered with Oxford BioTherapeutics on AI-guided T-cell therapies: the multi-year collaboration gives BMS access to OBT's discovery platform.

The constraint in biopharma has shifted from compute to data architecture: Is building biology-native data infrastructure the primary differentiator between drug discovery programmes with compounding returns and those without?

AI health tech is booming. The cures are not: TNW analysis finds the gap between AI-generated hits and approved medicines remains across every major disease area.

Medtech:

IMDRF drafts framework to align regulators before individual jurisdictions drift: the draft covers validation, transparency and lifecycle management.

Approved but Not Proven: Is there a Validation Gap in FDA-Approved AI Medical Devices?

AI scribes reduce charting burden when used frequently: but the benefit is dose-dependent.

ProSomnus wins FDA clearance for its sleep apnea device: The AI-driven oral product has been cleared via the 510(k) pathway.

Advanced manufacturing:

Hyundai Motor commits $26bn to US investment through 2030: including AI and robotics manufacturing: the largest known foreign commitment in US manufacturing this year so far.

TSMC on course for fourth successive record-profit quarter: But the revenue concentration from Nvidia is now a governance concern for downstream manufacturers.

Kuka outlines 'Automation 2.0', combining AI and industrial robotics: the German automation specialist is integrating AI into its current lines rather than building new.

Manufacturers test physical AI at pilot scale: Tata analysis of active industrial pilots finds full process automation needs longer.

Kaleris announces AI-powered yard gate automation: logistics yards emerge as an early production deployment site for autonomous AI systems.

Insurance:

NAIC advances AI governance model law and cyber risk standards: the March meeting produced draft guidance on AI in claims handling, underwriting decisions and consumer-facing applications.

AI-enabled attacks require new cyber risk underwriting models: existing pricing was calibrated on human-speed attack timelines; AI-enabled intrusions are faster and cheaper.

Mutual insurers' governance structure provides an advantage in AI oversight: long-term alignment between policyholder and shareholder interests allows mutuals to prioritise accuracy and fairness over short-term efficiency.

But what set podcast tongues a-wagging?

Three conversations this week focused on structural AI risk. But from different angles.

Jensen admits failure.

Jensen Huang's most candid moment on the Dwarkesh Podcast was not about Nvidia's supply chain - it was about what it failed to do.  Nvidia was not in a position to make the multi-billion dollar capital commitment into Anthropic that Google and AWS made in the early rounds. 'That was my miss.'

The structural consequence: Anthropic's training compute dependency on Google and Broadcom is a direct result of that missed relationship. Jensen explicitly frames Anthropic as 'a unique instance, not a trend'.

SaaS churn could become quarterly.

Ben Horowitz on the a16z podcast made the case against two structural axioms that have governed enterprise software strategy for a generation. First: the idea that you cannot buy engineering velocity dissolves now it is clearly purchasable with compute and data. Second: lock-ins are weakening because its increasingly AI agents using the systems.

Our take? Horowitz is papering over contractual and regulatory constraints that still bind complex industries - technical portability and legal portability are not the same thing. The more defensible version of his argument is: the rate at which enterprises re-evaluate software is accelerating.

Is the Pentagon's 'original sin' a procurement template?

Emil Michael, Under Secretary of War for Research and Engineering, on Big Technology named the structural error that produced the Anthropic supply chain designation: the DOW entered the AI era with a single vetted vendor on classified networks, creating a monopoly dynamic where Anthropic could set contract terms.

His proposed remedy - competitive outcome-based procurement replacing cost-plus contracts, with Google as the model - is now being implemented. The implication for regulated enterprises across healthcare, insurance and financial services is, perhaps, what happens to procurement leverage when we depend on a single AI provider and that provider decides to renegotiate.

The terrible public mismanagement aside, the DOW's experience could be read as a leading indicator of where procurement terms, value alignment requirements and supply chain risk designations are heading across regulated industries.

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