SNI: WEEK 19

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: Professional doom mongering is out of fashion.

Say it quietly, because AI dooming is almost deafening in popular circles. But it might actually be tapering off among commentators, as various forces pull attitudes in the opposite direction.

An Ezra Klein essay, argued the AI job apocalypse is structurally implausible - pointing out that economists are largely sceptical of mass-joblessness predictions.

The labour market signals agreed: software engineering job postings are up 18% and US graduate unemployment is down to 5% - while new-grad hiring is up 5.6%. A blip? Or something more?

Perhaps these grads are working at startups? Atlas - Stripe's online service for incorporating U.S. companies - reached 100,000 incorporations, with first-quarter formations up 130%. And job prospects look good at Anthropic too - its annual run rate is estimated to have increased from $9bn to $44bn. In less than six months. Feel free to skip back and read that again.

All this good cheer seems to be taking the edge off of doom scenarios among policymakers as well. The EU agreed to slim down the AI Act and push high-risk obligations to 2027, the first material reversal in the Brussels timetable since enactment.

Big capital is also ratifying the vibe shift. Anthropic, Blackstone and Goldman Sachs launched a $1.5bn enterprise AI services joint venture. AI engineers will soon be sitting next to clinicians, underwriters and operations leads. And OpenAI leaked its $10bn Deployment Company with the same thesis - there's no AI transformation without organisational change - and the last mile of deployment is the knot to be unknotted. To complete the set, Sierra closed a $950m round at a valuation north of $15bn.

Microsoft's Work Trend Index explained why the sudden rush to consulting - only 19% per cent of enterprises have both high levels of individual capability and organisational readiness. And a growing segment of skilled employees are stuck in unready organisations. Venture capital seems keen to fund services arms and get the rubber in contact with the road for the sake of their earlier investments.

So, with this news absorbed, reports of Anthropic weighing a $50bn round at a near-trillion-dollar valuation, gave the doom thesis an even harder kick. As did their deal with SpaceX for roughly 220,000 GPUs and 300MW of compute. Again proving that capital is rerouting to whoever can secure power.

But some press framing kept lagging behind the data. Cypto business Coinbase laid off 14 per cent of staff - and cited AI acceleration - just as Robinhood reported its crypto trading revenue was down 47% year-on-year.

Which just goes to prove that the doom narrative has begun to dissipate. But it is certainly not dead yet.

And on that note, here's everything else worth reading this week:

AI & tech:

• Cisco acquires AI security startup Astrix for $400m: adversarial-AI gets its first Tier-1 acquisition, repricing security as enterprises defend against agent rollouts.

• Anthropic uses Code with Claude DevDay to push the agent story: memory features, multi-agent orchestration and an infinite-context roadmap all make a splash.

• Perplexity Computer is generally available on Mac: the agentic super-app lands at $500m ARR, with multi-model features as a durable moat.

• OpenAI's compute partner Cerebras is on track for a blockbuster IPO: a public-market read on whether the premium will (or won't) hold.

• Denmark pauses new grid connections as AI data centres overwhelm Europe's cleanest power supply: the first national-grid moratorium in the EU on AI compute admission, and a concrete example of why power, not silicon, may be the gating constraint.

  
  

Biopharma:

• Evotec nominates the first preclinical development candidate from its AI-and-data-integrated dermatology platform with Almirall: AI drug discovery exits proof-of-concept and produces a clinic-bound molecule on a timeline materially shorter than the industry standard.

• ByteDance's Anew Labs presents its first AI-designed therapy as TikTok's parent enters drug discovery: a generative-AI-designed IL-17 small molecule debuts at the major Boston immunology meeting, signalling a structural Chinese-tech footprint in pharma R&D.

• Sanofi commits $294m to expand its AI Center of Excellence in Toronto: big-pharma capex specifically for AI deployment, with Toronto becoming the integration point for Sanofi's AI-DD pipeline.

• Taoiseach Micheál Martin opens AstraZeneca's new Dublin facility, calling it 'a vote of confidence' in Irish life sciences: the geographic and capital signals are real for a pharma-manufacturing modernisation cycle now visibly running through Ireland.

• AI detects pre-tumour pancreatic cancer signals: the kind of concrete clinical milestone that may help dampen popular AI doomerism?

Medtech:

• Abbott receives concurrent FDA clearance and for AI-enabled imaging device: it seems that regulator-friendly geometries on both sides of the Atlantic are now table stakes for the largest device makers.

• Nature Digital Medicine publishes 'the absence' – a paper on the lifecycle risk-management gap for AI-based medical devices in radiology: the next regulatory pressure point lands as more devices clear initial review without commensurate post-market obligations.

• Enzo Health raises $20m to expand its AI home-health platform: the category keeps attracting capital as adoption friction is low and reimbursement maths friendly.

• A startup says it has invented a beanie that reads your mind: the consumer-BCI hype cycle returns with a non-invasive form factor.

• InformationWeek details the silent erosion of enterprise AI by data poisoning: a story that hits clinical data integrity harder than the headline tech narrative - CISOs should read it twice.

Advanced manufacturing:

• AbbVie chooses North Carolina for a $1.4bn manufacturing campus: pharma capex continues to favour US sites with predictable AI-process integration roadmaps, repricing the regulatory and energy story regionally.

• Intel and FPT partner on AI-driven autonomous factories: tier-2 manufacturers are the new test bed for autonomous production, with tier-1s skipping the pilot stage and going straight to deployment.

• Panthalassa raises $140m to power AI at sea: maritime power and ocean-cooled compute join orbital data centres as the next frontier infrastructure bets, with the same investor logic – secure non-grid generation early.

• Applied Materials acquires NEXX to broaden its advanced-packaging portfolio: the factory step that bonds memory to GPUs is the bottleneck behind every AI server build; larger panels mean more chips per shift, and packaging now gates the GPUs you build the way power gates the GPUs you run.

• Siemens releases its Eigen Engineering Agent for industrial workflows: the agent-engineering product ships into general availability.

Insurance:

• Federato pitches an AI-native underwriting workbench at carriers: underwriting workbenches converge on the agent-orchestration patterns first proven in software engineering.

• Governance infrastructure is now the precondition for agentic AI: carriers are building scaffolding ahead of agent rollout in ways their tech-sector counterparts are not.

• Sidley Austin flags FCA and EIOPA drift on AI-assisted pricing models: UK and EU divergence on insurance AI becomes operational, tying directly to the EU AI Act delay signal in the frame.

• A Canadian insurtech launches an insurance shopping app inside ChatGPT: distribution moves to the assistant layer before underwriting catches up.

• Insurance Thought Leadership argues underwriting fundamentals must come before AI: a practitioner reminder of analytics first, AI second lands as a useful balance to the agent hype.

But what set podcast tongues a-wagging?

China's not what you think it is.

Azeem Azhar on the Exponential View reports from a week in China where he discovered that Zhipu alone is serving 5.5 trillion tokens per day and that developers are joining the platform at the rate of 10 per minute.

And this is despite compute constraints being universal across the labs he visited, largely due the shortage of Nvidia chips. However, Azeem contends, research and product velocity continue regardless. Perhaps even more surprisingly, Anthropic's Claude is the preferred technical-team choice in Chinese labs, despite the geopolitical decoupling story. Policymaker intentions do not seem to be making major inroads on practitioner behaviours.

New roles are arriving.

Aaron Levy at Box is hiring agent engineering roles, reports Nathaniel Whittemore on the AI Daily Brief. These are internal FTEs whose only job is to wire agents to run on top of Box, Salesforce and Workday. The point? Agents expose an infinite backlog of work that was always there, hidden behind execution capacity. And it means that the work no longer drains you through typing. It drains you through judgement. The bottleneck for the next 24 months will be coordination, evaluation, context curation and the human ability to keep up.

The Western AI stack is a foreign-policy lever.

Despite Azeem Azhar's findings on China, Sarah Rogers, Under Secretary of State for Public Diplomacy, on the a16z Show frames AI free-speech as a foreign-policy priority on par with internet freedom in the early 2010s. Rogers names American structural advantages 'that a lot of people don't appreciate' – and presents EU regulation as the petri dish for similar US legislation.

Users cuddle up to deterministic AI before they use agents.

Descript CEO Laura Burkhauser on the Cognitive Revolution maps the hierarchy of AI tooling inside a video editor base. Purpose-built deterministic features – studio sound, green screen, voice overdub – are universally loved. Agentic editing copilots like Descript's Underlord are wanted but flagged as not-yet-good-enough.

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