SNI: WEEK 09

Welcome to all the AI news that matters this week - across biopharma, medtech, manufacturing and insurance.

---

tl;dr: the price of position

This week, the cost of having – or lacking – a clear position in the AI value chain became visible. In stock prices. In capital flows. And in and competitive strategy. In one case, the position was geopolitical.

But despite common misconceptions, capital is not retreating from AI. It is moving directionally – toward specific chokepoints in the value chain where advantages compound and competitors struggle to follow.

In AI & tech

• DeepSeek withheld its V4 model from US chipmakers and gave early access to Huawei, reframing AI competition along supply-chain lines

• The 'SaaSpocalypse' sell-off forced a reckoning between software incumbents and AI model providers over who owns the enterprise stack – Salesforce posted record $11.2bn quarterly revenue while its sector lost 10%

• HSBC backed the incumbents, issuing buy ratings at what it called historical valuation lows

• Nvidia reported Q4 revenue of $68.1bn, up 73% year on year – CEO Jensen Huang told investors they 'got it wrong' on the SaaSpocalypse

• Anthropic pushed into enterprise, unveiling agents for HR, finance and investment banking

• AI chip startups absorbed more than $1.1bn in venture capital in a single week

• Meta agreed a deal with AMD potentially worth more than $100bn for its MI450 processors

In Biopharma

• IQVIA signed an agreement to acquire drug discovery assets from Charles River Laboratories – including an AI platform and more than 20 years of scientific data

• Bruker expanded its collaboration with Noetik to scale spatial biology foundation models toward one billion spatially resolved human cells

  
  

In Medtech

• DeepHealth secured CE marking for its remote imaging platform, connecting more than 400 scanners across physical boundaries

• Oura launched its first proprietary LLM for women's health, built on biometric data from its wearable ring

  
  

In Manufacturing

• RLWRLD raised $26m for robotics foundation models trained directly inside live industrial operations

• Axelera AI raised more than $250m to scale production of its edge AI processor

• Apple announced it will bring Mac mini production to Houston as part of a $600bn commitment to domestic manufacturing

In Insurance

• Legal moats are emerging as key competitive weapons

• Clear Group CEO Mike Edgeley argued that AI will widen the gap between adopting and non-adopting brokers rather than eliminate the intermediary

• Concirrus and Sixfold launched or expanded AI-native underwriting systems

And if you're still hungry for more, here's the detail on each:

AI & Tech

The week's dominant question shifted from who has the best technology to who is allowed to buy it – and what that answer means for hundreds of billions of dollars in market capitalisation.

DeepSeek withholds access from US chipmakers: DeepSeek withheld early access to its forthcoming V4 model from Nvidia and AMD, granting it instead to Huawei – a break from standard industry practice that signals deepening technological decoupling between the US and China. The move reframes AI competition along geopolitical lines. Position in the value chain now depends on which supply chains a company is permitted to access, and which governments it is permitted to serve. For CEOs tracking the 'price of position', this is the sharpest version of the theme: a position you cannot reach at any price.

The SaaSpocalypse and the stack war: The B2B software sector suffered its worst month in years in January, with the S&P North American Software Index falling more than 10% as investors concluded that AI agents could make seat-based licensing models obsolete. ServiceNow shares fell roughly 11% after reporting decelerating remaining performance obligations. SAP dropped 15% after missing cloud backlog targets. Microsoft slid 10% as Azure's cloud growth decelerated and capital expenditure projections exceeded $100bn for the fiscal year. Major institutional funds including Vanguard and BlackRock began rotating out of application software and into hardware and energy.

Salesforce: used its Q4 earnings on 26 February to mount a full defence. Total revenue reached a record $11.2bn for the quarter, up 12% year on year. The company announced a $50bn share buyback, raised its dividend by 6% and introduced a new metric – agentic work units – designed to measure whether AI agents completed tasks rather than merely processed tokens. Salesforce did not disclose per-unit pricing for the new metric. CEO Marc Benioff mentioned 'SaaSpocalypse' six times on the call and presented an architectural diagram that placed SaaS platforms at the top of the enterprise technology stack, with AI models sitting below as interchangeable engines. This directly contested the vision OpenAI presented when it launched its Frontier enterprise agent earlier in February, which placed AI on top and SaaS platforms beneath as data sources.

HSBC weighed in: with a contrarian note, arguing that consumer AI platform developers have 'little to no experience creating enterprise-class software' and would be 'architecting from scratch in unfamiliar, highly complex areas'. The bank rated Oracle, ServiceNow, Salesforce, HP and CrowdStrike as buys, calling sector valuations 'at historical lows'. Anthropic pushed into the same contested territory on 24 February, unveiling enterprise agents for HR, finance and investment banking and positioning Claude as what William Blair described as a 'platform-level intelligence layer across enterprise workflows'. The tension is real: Salesforce posted record quarterly revenue in the same quarter that its sector lost 10%. The sell-off may reflect a repricing of long-term architecture rather than near-term weakness – but that distinction matters less to shareholders watching the drawdown in real time.

Nvidia Q4 reinforces the hardware layer's dominance: the world's most valuable company reported Q4 fiscal year 2026 revenue of $68.1bn on 26 February, up 73% year on year, with data centre revenue reaching $62.3bn. Guidance for Q1 was $78bn – well above the $72.6bn consensus estimate. CEO Jensen Huang told CNBC that investors 'got it wrong' on the SaaSpocalypse, arguing that AI agents would increase demand for enterprise software by expanding its addressable market rather than cannibalising it. The results exposed a contradiction at the heart of the sell-off: the companies building AI's compute layer are growing faster than at any point in their history, while the companies expected to deploy it are being repriced downward.

Meta agreed a deal with AMD: for its MI450 processors in an arrangement potentially worth more than $100bn across a 6-gigawatt deployment. AMD issued Meta a performance-based warrant for up to 160 million shares at $0.01 each – giving Meta a path to a 10% equity stake in the chipmaker as milestones are met. The deal came days after Meta announced a separate long-term partnership with Nvidia for millions of chips. OpenAI has forecast a need for $665bn in compute costs over five years, with Amazon reportedly considering a $50bn investment – conditional on either an IPO or the achievement of artificial general intelligence.

A fictional memo rattles $300bn in market value: The market's fragility was exposed on 22 February when Citrini Research published a Substack essay titled 'The 2028 Global Intelligence Crisis' – a fictional macro memo set in June 2028 describing a scenario in which agentic AI sharply reduces white-collar employment, pushes unemployment above 10% and triggers a 38% drawdown in the S&P 500. The essay contributed to an estimated $300bn sell-off the following day – though the figure reflects a broad market reaction rather than a documented institutional repositioning. Citrini's founder said he did not anticipate the scale of the response: 'If I thought that stocks were gonna move on this, I wouldn't have made it free.' Shares of Visa, Mastercard, Uber and DoorDash fell as the memo argued that AI agents could compress fees, erode friction-based business models and reduce demand for human-operated software.

Jim Reid, a strategist at Deutsche Bank, said the report's 'vibes-to-substance ratio is undeniably high' but added that 'that doesn't mean it will ultimately be wrong'. Joshua Brown of Ritholtz Wealth Management argued the memo assumed 'every single negative piling on top of each other without an offset in sight' – a pattern that 'is very rarely how these things end up'. The episode illustrated how sensitive capital markets have become to AI's second-order effects, even when the catalyst is a thought experiment rather than a change in institutional positioning.

AI chip capital surges toward specific positions: While software stocks fell, capital poured into the hardware layer at extraordinary scale. In a single week, three AI chip startups raised a combined $1.1bn. MatX secured $500m for an SRAM-based LLM accelerator; Axelera AI raised $250m for its low-power edge processor; and SambaNova closed $350m alongside a partnership with Intel. Each is pursuing a distinct architectural bet on how AI inference will work at scale.

Biopharma

The week's two largest moves both centred on consolidating the infrastructure – data, platforms and scientific assets – that makes AI-driven drug development viable over time.

IQVIA acquires Charles River discovery assets: IQVIA signed an agreement to buy selected discovery service assets from Charles River Laboratories, including five sites specialising in in vitro drug discovery and a small molecule AI platform supported by more than two decades of scientific and operational data. These assets have enabled more than 100 molecules to progress into clinical trials, with several now commercially approved.

Bruker and Noetik scale spatial biology AI: Bruker Spatial Biology expanded its collaboration with Noetik, building on a prior study of more than 3,500 patient samples with the CosMx Spatial Molecular Imager. Noetik uses CosMx data to train foundation models that simulate human biology at tissue level, imaging thousands of patients across what it says is the largest subcellular spatial transcriptomic dataset in oncology.

BreezeBio raises $60m for genetic medicine delivery: BreezeBio – formerly GenEdit – raised $60m in a Series B round to advance its NanoGalaxy nanoparticle delivery platform and develop its own therapeutic pipeline. The company's polymer nanoparticles can target specific organs and be dosed multiple times, addressing a persistent limitation of existing gene therapy delivery methods.

Its lead programme, BRZ-101, delivers genetic instructions into immune cells to promote tolerance in Type 1 diabetes – coercing the body into suppressing the autoimmune response that destroys insulin-producing beta cells. No curative treatment currently exists for the condition. BreezeBio is advancing BRZ-101 into final preclinical studies this year. The round was co-led by Yuanta Investment and DSC Investment.

MedTech and digital health

Two developments this week illustrated how AI is becoming structurally embedded in medical devices and health platforms.

DeepHealth secures CE mark for remote imaging platform: DeepHealth, the health informatics subsidiary of RadNet, secured CE marking for TechLive – its FDA-cleared remote imaging and radiology management solution – enabling European commercialisation. TechLive extends expert oversight across MRI, CT, PET/CT and ultrasound from a single consolidated interface, allowing certified technologists to operate scanners across physical boundaries. Within RadNet, TechLive's largest customer, the platform has connected more than 400 scanners and delivered a 42% reduction in MR room closure hours and a 27% increase in access to complex procedures. European workforce shortages in radiology have reached what DeepHealth described as critical levels in multiple markets, making remote scanning capacity a structural priority rather than a convenience.

Oura launches proprietary women's health AI: Oura launched its first proprietary large language model, designed to deliver personalised women's health guidance across menstrual, pregnancy and menopause-related topics. The model is built on biometric data collected through the Oura Ring – sleep patterns, activity levels, stress indicators and cycle tracking data – combined with clinician-reviewed medical research.

Clinical lead Tanvi Jayaraman said the model interprets questions 'through the lens of what's happening in that person's body, not just a generic symptom search'. The model runs on Oura's own infrastructure, with no data shared with third-party systems. Oura raised $900m in October 2025 at a valuation of approximately $11bn, led by Fidelity Management & Research Company. The women's health model is available for testing through Oura Labs.

Manufacturing

The companies building AI's physical infrastructure are consolidating their positions through structural reorganisation and large-scale capital deployment.

RLWRLD raises $26m for industrial robotics AI: Seoul-based RLWRLD raised $26m in a Seed 2 round on 25 February, bringing total funding to $41m for robotics foundation models trained directly inside live industrial operations. Unlike companies developing models in lab environments, RLWRLD trains its AI on real production floors through partnerships with CJ Logistics, Lotte and other major industrial operators across South Korea and Japan.

The approach creates a proprietary real-world data advantage: each deployment generates training data that improves the model's performance across subsequent sites. Headline Asia founding partner Akio Tanaka said the company's ability to 'accumulate real-world data' through industrial partnerships is 'a critical foundation for building long-term competitive advantage'. RLWRLD plans to launch its robotics foundation model in the first half of 2026.

Axelera AI raises $250m+ for edge AI chips: Dutch startup Axelera AI raised more than $250m in a round led by Innovation Industries, with participation from BlackRock and Samsung Catalyst Fund. The company's Metis chip delivers 214 trillion operations per second while consuming around 10 watts – making it suitable for battery-powered devices such as warehouse robots and connected industrial equipment.

Metis uses an architecture Axelera calls digital in-memory computing, which stores and processes data in the same SRAM modules rather than shuffling it between separate circuits. The company is developing a successor chip called Europa, capable of 629 trillion operations per second with up to three times the performance per watt of competing products. The funding will scale chip manufacturing and expand customer support.

Apple brings Mac mini production to the US: Apple announced on 25 February that it will bring Mac mini production to Houston later this year – the first time the compact desktop has been manufactured in the United States. The move expands Apple's Houston footprint, where it began mass-producing advanced AI servers in 2025 at a 250,000-square-foot facility. The company has sourced more than 20 billion US-made chips from TSMC, Broadcom and Texas Instruments since August, and is investing in partner facilities in Texas, Arizona and Kentucky that produce wafers, packaging and cover glass.

The expansion is part of Apple's $600bn commitment to domestic manufacturing and AI infrastructure over four years. CEO Tim Cook said the company shipped AI servers from Houston ahead of schedule. The reshoring signals that the 'price of position' extends beyond chip design into the physical supply chain: Apple is building manufacturing capacity where it controls both the hardware and the workforce pipeline, including a new Advanced Manufacturing Centre for hands-on training in AI, automation and smart manufacturing.

Insurance

Here is the clearest evidence that AI competition is moving from operational efficiency into strategic positioning – with patents, platforms and market structure all in play.

AI patents emerge as competitive weapon: An analysis published by Insurance Thought Leadership posits that AI patents are fast becoming insurance's most powerful competitive weapon, yet most carriers have no strategy to compete. And some are firmly ahead - lest we forget that State Farm, USAA and Allstate have filed 77% of all AI patents

Broker stocks fall as AI displacement fear returns: US insurance broker stocks suffered their worst single-day decline since 2008 on 9 February, after Insurify announced a ChatGPT-powered insurance shopping tool. The sell-off hit large brokers despite many having predominantly commercial, relationship-driven books.

A briefing published by Clear Group CEO Mike Edgeley argued that the market overreacted. For commercial insurance – where risk understanding, insurer negotiation and claims advocacy are central – the structural barriers to disintermediation remain high. Edgeley made the case that AI will widen the gap between brokers who adopt it and those who don't, rather than eliminate the intermediary. The firms that invest early will deepen their competitive position. Those that hesitate risk being outperformed by better-equipped peers.

Concirrus launches Inspire underwriting platform: Concirrus formally launched Inspire, an AI-native underwriting platform for specialty insurers and MGAs covering the full underwriting lifecycle from submission to bind. The platform combines line-specific underwriting applications with a line-agnostic policy management foundation, designed to scale volume without proportionate increases in headcount.

General Magic raises $7.2m for SMS-based insurance AI General Magic raised an oversubscribed $7.2m seed round led by Radical Ventures, with participation from a16z Speedrun. The Toronto-based company builds AI agents that handle pre-quote, post-quote and claims coordination over SMS. In early deployments with a large general insurer, the agents reduced time-to-quote from roughly 30 minutes to under three minutes.

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