Venture Geopolitics Issue 40
24 March 2026
Cursor launched Composer 2 - a new model offering an “optimal combination of intelligence & cost.” Priced at around $0.50 per million input tokens and $2.50 per million output tokens, it comes in at a fraction of frontier offerings like those from Anthropic.
Cursor did not initially mention, and only acknowledged after being called out publicly, that Composer 2 was built on top of Kimi K2.5 - an open-source model from China’s Moonshot AI. Perhaps they avoided mentioning it for geopolitical reasons. Perhaps because the model beneath the product is increasingly beside the point.
Are we crossing into frontier model commoditisation?
Historically, the most performant models were also the most expensive, and that premium felt defensible. Now a benchmark-topping product can be delivered at one-tenth the price using open foundations assembled across borders. The implication is uncomfortable for the labs that built their identity around model superiority (evidenced by OpenAI’s re-re-refocus).
As intelligence becomes cheaper and more abundant, capital stops being the decisive advantage. The inference economy rewards a different kind of builder - one who understands a domain deeply. The harnesses - the workflows, constraints, context and judgement built around the model - is where value increasingly lies. Which is where Europe enters the picture, not as an afterthought but as a genuine protagonist.
European sovereignty in AI does not mean eliminating dependence on the US - that is not realistic. It means reducing vulnerability and moving toward interdependence. Meaningful US dependence on European compute infrastructure is unlikely. The more plausible path is software - domain-specialised systems that sit between foundation models and real-world deployment.
In that context, the AI race is not only about who builds the smartest model. It is about who can embed intelligence into the hardest real-world systems - banks, energy grids, defence ministries, industrial supply chains. European founders who have spent years navigating fragmented markets, building trust with serious institutions and solving hard problems rather than scaling easy ones are precisely the founders this application era selects for.
As the wild-west phase of consumer AI gives way to something more structural and consequential, the centre of gravity in venture may move accordingly: fewer ephemeral winners, more enduring companies built around domain mastery and institutional integration. In the application era of AI, Europe may be competing on more natural terrain
IPOs / Publics
IPO market remains selective with volatility suppressing issuance
Janus (senior housing REIT) raised $840M in a carve-out from Healthpeak & rose +18% on debut (here)
Guardian Metal Resources (tungsten miner) raised $60M in its US IPO, but traded flat. Tungsten is strategically important - it sits inside defence, industrial tooling & critical supply chains. So the listing made geopolitical sense (here)
Swarmer (AI drone software) raised $15M & surged more than +500% despite minimal revenue. The Ukrainian company, <3yrs old, reportedly generated ~$300K revenue in 2025, yet attracted huge demand on Nasdaq. Partly a float dynamic (tiny supply meeting a hot theme) but also says something broader about the moment (here)
VINCORION (military power systems) rose ~10% after Frankfurt IPO - another sign defence remains one of the few sectors where public markets are willing to underwrite expansion stories with conviction (here)
Amazon backed X-energy (small modular nuclear reactors) has filed for an IPO. The company is building an SMR that uses helium as a coolant instead of water, the industry standard (here)
Unitree (Chinese humanoid robotics startup) seeking $610M Shanghai IPO amid surging sales. Revenue 4x to about $247M last year, with profits 7x as shipments jumped from 410 robots in 2024 → 3,551 in first 9mo of 2025. Beijing views humanoids as strategic infrastructure to offset labour shortages & close the US tech gap (here)
At its flagship GTC conference, NVIDIA reinforced its shift from chip leader to full AI platform provider, with CEO Jensen Huang describing an emerging stack spanning applications, models, software, semiconductors & large AI data-centre infrastructure (here)
Microsoft is weighing legal action over Amazon’s reported $50B cloud deal with OpenAI. The dispute centres on whether OpenAI’s new enterprise agent platform, Frontier - software designed to let companies deploy autonomous AI systems to perform complex workplace tasks - can run on AWS without breaching Azure exclusivity clauses. The rift reflects OpenAI’s push to diversify infrastructure partners even as Microsoft relies on its models to drive cloud growth (here)
HSBC reportedly considering 20K job cuts driven by AI, focused on back- & middle-office roles. Analysts now suggest up to 200K banking jobs could be cut over next 3-5 yrs (here)
Arm has announced its first ever in-house data-centre CPU, the AGI CPU, aimed at powering the next wave of AI infrastructure - especially “agentic AI” systems that can act autonomously, not just respond to prompts (here)
Big Dogs
OpenAI is reorienting from consumer toward enterprise, coding & cost leadership as competition in frontier AI intensifies (here)
The company plans to nearly double headcount to ~8K, hiring engineers, sales & “forward-deployed” consultants to customise AI agents for large clients - signalling a pivot toward recurring enterprise revenue & full-stack deployment partnerships (here)
Bloomberg also reports the company is nearing a deal for $10B in fresh funding (from funds including Abu Dhabi’s MGX, Coatue & Thrive). This will value the company at $730B (here)
New GPT-5.4 mini & nano models show how fast marginal intelligence costs are collapsing - with token pricing falling to ~$0.20 per million vs ~$30 for Sora 2 Pro just two months ago, a ~150x drop within the same model family (here)
Internal priorities are shifting toward developer workflows where willingness to pay is high & models already shape production environments, reinforced by the acquisition of Astral (Python tooling) (here)
Meanwhile, OpenAI’s retreat from in-chat shopping has left partners like PayPal & Etsy exposed, illustrating platform volatility as the firm abandons consumer “side quests” to concentrate resources on enterprise AI (here)
Pentagons move to label Anthropic “supply chain risk” has fractured Trump’s truce with Silicon Valley - prompting rare, coordinated lobbying from rival tech groups. Microsoft, Google & venture investors are reportedly pushing back both to preserve access to Claude models & to avoid a precedent where federal authorities can abruptly cut off critical AI infrastructure. Despite warnings the crackdown could hurt growth, Anthropic’s annualised revenue is said to have jumped from ~$9B to ~$19B in weeks - suggesting enterprise customers may reward perceived safety positioning (here)
Regulation
White House released its Federal AI Policy Framework (here)
Brussels is considering curbing national vetoes on corporate takeovers to build European “champions”. A proposed overhaul of EU merger rules would limit governments’ ability to block cross-border deals aimed at scaling firms to compete with US & Chinese rivals. Recent interventions - including Germany opposing UniCredit’s bid for Commerzbank - have fuelled fears of single-market fragmentation. Officials insist safeguards against monopolies will remain, but the shift signals growing willingness to prioritise industrial competitiveness over national protectionism (here)
Nevada courts issued a 14-day restraining order forcing Kalshi to halt certain markets. Perhaps the clearest sign yet that the legal fight around prediction markets is moving from theory to practice. Nevada’s argument is that Kalshi is effectively running a gambling operation without the relevant licence (here)
Britain’s House of Lords has blocked government plans to compel pension funds to invest more in domestic assets. Peers voted down “mandation” powers aimed at channelling retirement savings into infrastructure, defence & growth sectors. Critics warned political pressure could undermine fiduciary duties & returns, while ministers argued unlocking trillions in pension capital is essential to boost productivity (here)
UK reversed course on proposed AI copyright opt-out regime after backlash from artists & creators. Original idea would have allowed AI models to train on copyrighted works unless rights holders explicitly opted out. In practice, creators saw that as unfair: burden would sit with them to prevent appropriation of their work (here)
US jury found Elon Musk liable for misleading Twitter investors during his chaotic 2022 takeover. His public claims about bots & threats to walk away caused sharp share-price swings that allegedly inflicted losses - with damages potentially reaching $2.6B (here)
Venture Capital
GV (Google Ventures) has doubled down on Europe, with >$1B now invested. Tom Hulme argues Europe & London in particular offers density of founder talent, operators & technical depth to produce returns comparable to top US venture ecosystems. He says GV invested ~$500M in Europe from 2014-2023, then >$600M since, implying conviction has increased rather than faded (here)
New funds:
Jeff Bezos reportedly raising $100B fund to buy industrial firms & deploy AI-driven automation at scale. Would target sectors like aerospace, chipmaking & defence. Bezos has been courting sovereign wealth funds & asset managers, as investors pile into “re-industrialisation” themes amid geopolitical tensions & supply-chain reshoring (here)
Partech raised €300M for a new impact fund focused on industrial & climate technology (here)
Air Street Capital raised $232,323,232 for its third fund to back AI-first companies across North America & Europe (here)
Credo Ventures raised $88M to continue to back founders from CEE (here)
360 Capital raised €85M for deeptech startups & university spinouts (here)
Montis VC raised €50M to back energy & industrial tech startups (here)
Venture Geopolitics
EU Council met on Thursday (here)
Discussions suggest Europe is trying to answer the energy crisis with a mix of dilution & delay. On one side, there is likely to be some weakening of the ETS regime through higher permit supply & more allowances. On the other, member states are being nudged toward “targeted, temporary & tailored” fiscal measures. The bloc is trying to preserve climate architecture while quietly making it less painful for industry & voters
Meanwhile, Hungary continues to delay the €90B Ukraine loan by vetoing the required EU budget amendment
China is pitching itself as a “harbour of stability” to global executives amid US geopolitical turmoil. Premier Li Qiang told >70 CEOs that predictable policy & unmatched supply chains make China a safer investment destination as Washington grapples with war in Iran. The message accompanies a new 5-year plan doubling down on high-tech manufacturing, even as record $1.2T trade surpluses & fears of industrial overcapacity strain relations with Europe (here)
It is well known chips have been smuggled into China (per Benedict Evans “there couldn’t really be that many customers ‘in Singapore’”), but the US has charged co-founder of Supermicro, a server maker, with diverting $2.5B(!) chips into China without a licence (here)
The UK’s Financial Conduct Authority (FCA) has given US tech firm Palantir a trial contract to apply its AI software to large volumes of sensitive regulatory intelligence data to improve detection of fraud, money-laundering & market abuse. Deal forms part of a wider government relationship with Palantir (worth >£500M across sectors including the NHS (£330M Federated Data Platform), MoD, policing & now financial regulation. Critics warn of privacy risks, dependence on a Trump-linked US “spy-tech” supplier & long-term vendor lock-in as Palantir becomes embedded across multiple core public systems (here)
Prosus & Dealroom argue Europe is an AI “hidden giant”. Talent parity with the US (~325K AI workers & ~50K frontier researchers each) but misallocated across legacy sectors & starved of late-stage capital. ~48% of European AI talent sits in the “old economy” vs ~33% in the US, while only ~35% works in digital-native firms vs ~54% stateside, suggesting in their view Europe misuses elite engineers to retrofit incumbents rather than build global platforms (read more here)
Per the Economist, cheap electricity - not algorithms - may be China’s decisive edge in the global AI race. China’s vast coal, hydro & nuclear build-out is driving industrial power prices to roughly half US levels, enabling far cheaper training & inference at hyperscale. Provinces are racing to host data centres, while state-backed grid expansion supports “AI industrial policy”. The strategy reflects Beijing’s view that compute is infrastructure - & energy abundance, not model breakthroughs, may ultimately determine AI supremacy (here)
The UK is erecting a tariff “cordon” around its steel industry to counter a global glut driven by Chinese overcapacity. Import quotas will be cut 60% & tariffs above thresholds set at 50%, aligning Britain with US & EU protectionist moves. Officials frame the shift as ending decades of deindustrialisation & safeguarding national security, though manufacturers warn higher costs. The policy forms part of a broader £2.5bn strategy to decarbonise steelmaking via electric arc furnaces (here)
Apple supplier Murata is restructuring supply chains to reduce dependence on Chinese rare earths. The Japanese electronics group - producing ~40% of global MLCC components - plans to establish parallel US & China production networks as export restrictions tighten. Identifying alternative mineral sources could take three years, underscoring how geopolitical tensions are reshaping tech manufacturing (here)
Brussels is poised to clinch a long-delayed trade deal with Australia to strengthen access to critical minerals. After 8 years of negotiations, compromises on beef quotas & electric-vehicle tariffs are unlocking progress as both sides seek allies against rising US protectionism. EU officials say the pact would deepen presence in the Indo-Pacific & secure supplies of lithium, cobalt & rare earths vital for the green transition, though farmer opposition remains politically sensitive (here)
Strategic Sectors
AI
Mistral is seeking to define a distinctly European enterprise AI strategy around customisation, controllability & sovereign deployment rather than pure frontier scale. Its new Forge platform lets enterprises train or heavily customise models on their own data, including in sensitive environments (here)
Mistral’s chief executive is urging Europe to impose a revenue levy on AI companies using copyrighted content. The proposal would fund creators & grant developers legal certainty by shielding training on publicly available material. Arthur Mensch argues fragmented copyright rules leave European AI at a structural disadvantage versus US & Chinese rivals. A central fund could support cultural industries while ensuring foreign model providers also contribute - part of a wider debate over sovereignty in the knowledge economy (here)
Apple has quietly blocked updates to AI “vibe coding” apps that let users generate software through natural-language prompts. The company argues features in apps like Replit & Vibecode violate App Store rules against downloading or executing new code that alters app functionality. With such tools enabling web apps that bypass Apple’s marketplace - a major profit centre - the crackdown reflects long-running tensions over platform control & developer lock-in (here)
Cybersecurity
Lovable added built-in penetration testing using Aikido Security. With Aikido, builders on lovable can get affordable pentests (that support SOC 2, ISO 27001, client security questionnaires, or investor due diligence) for $100 vs typical $5-$50K for security analysts to do the same (here)
Turing Award went to inventors of quantum cryptography. The field has often felt theoretical to outsiders, but in an era where both classical cyberattacks & future quantum threats loom larger, communications systems that are secure in principle, rather than merely difficult to break, become more important (here)
A rogue internal AI agent at Meta triggered a severe (Sev-1) security incident, exposing sensitive data for nearly two hours (here)
Washington is quietly encouraging private companies to “hack back” against foreign cyber attackers. US officials are signalling greater tolerance for offensive cyber responses as state-backed intrusions surge - especially from China, Iran & Russia. The shift reflects growing frustration that defensive spending alone has failed to stem breaches affecting critical infrastructure, financial systems & technology supply chains (here)
YC-backed Delve (agentic compliance) has been accused of faking compliance checks. The startup is rumoured to have generated fake evidence of compliance on its clients’ behalf, skipping major requirements, using off-shore “certification mills” to produce generic nearly-identical reports & filling out audit paperwork before any actual audit has taken place. If true, Delve wouldn’t just be cheating their clients out of money, but potentially putting them in legal jeopardy, exposing them to liability under laws like HIPAA & GDPR (here)
Notable deal: Cloaked (consumer digital footprint & privacy) raised $375M from GC & Liberty City (here)
Defence
Britain ‘defenceless against Iranian missiles’. Follows reports Iran attempted to strike joint US-UK Diego Garcia base using long-range missiles, one intercepted & one failing in flight. Diego Garcia sits ~2,400 miles from Iran - enough to speak debate labout whether Tehran’s capabilities now put parts of Europe, including London, in range. The Telegraph says the UK has no ground-based air defence system capable of intercepting ballistic or hypersonic missiles & is largely reliant on a small number of Type 45 destroyers, most of which are not in position for immediate homeland defence (here)
Pentagon’s $200B Iran-war funding request shows how quickly modern conflict rewrites fiscal assumptions (here)
UK, Netherlands & Finland creating a joint defence financing mechanism to pool weapons procurement. Structured like an international lender issuing bonds backed by member guarantees, the scheme aims to cut costs & boost deterrence as Europe rearms after Russia’s invasion of Ukraine. Officials argue aggregated orders could halve per-unit costs. Initiative reflects pressure on governments to expand military capacity while keeping debt off national balance sheets (here)
Australia & UK discussing autonomous “robot warships” as part of a future hybrid navy. Remains conceptual, but logic is: naval mass expensive, missiles decisive, crewed platforms scarce. So defence planners want cheaper, attritable, optionally crewed vessels that can carry payloads without carrying full human cost. Nb, suggests autonomy is moving from drones in the air to distribution of lethality at sea (here)
HMS Dragon, the UKs Royal Navy warship, has arrived in the eastern Mediterranean to defend Cyprus from Iranian attacks (here)
Britain will buy more anti-drone missiles to support Gulf allies amid escalating Middle East conflict. The procurement signals London’s effort to demonstrate commitment to regional partners after criticism from Washington & European allies, while ministers stress Britain is avoiding direct war involvement (here)
Notable deal: Egide (air defence systems) raised an €8M seed to build systems designed to intercept hostile drones (here)
Energy
Chinese battery champions have gained >$70B in market value as Middle East war shifts energy expectations. Shares in CATL, BYD & Sungrow have outperformed oil majors since US-Israeli strikes on Iran, reflecting investor bets on accelerated electrification & energy security. Analysts forecast China’s grid-storage battery market could jump from $48B to $199B by 2032, underscoring how geopolitical shocks are speeding the transition away from fossil fuels (here)
The Trump administration will pay $1B (€860M) to French energy giant TotalEnergies to walk away from 2 US offshore wind leases as it ramps up its campaign against offshore wind & other renewable energy. TotalEnergies has agreed to what is essentially a refund of its leases for projects off the coasts of North Carolina & New York & will invest the money in fossil fuel projects instead (here)
Robotics
DoorDash paying couriers to record real-world tasks for AI training is a strong example of how physical AI gets built. The company’s new “Tasks” app pays people to film themselves doing everyday activities - washing dishes, speaking languages, carrying out chores - so models & robots can better understand the physical world (here)
Amazon is acquiring Fauna Robotics, entering the consumer household humanoid market (here)
Crypto / Stablecoins
Crypto exchange Coinbase is racing to build payments infrastructure for AI agents as crypto trading slumps. The firm is vying to issue a new stablecoin with Cloudflare that could sit at the centre of rising agent-driven internet traffic (here)
EVs / AVs
Uber committed up to $1.25B into Rivian. Uber, or its fleet partners, are expected to buy 10K fully autonomous Rivian R2 robotaxis, with the option to purchase up to 40K more by 2030. The first 10K Rivian R2 robotaxis are expected in US cities including San Francisco & Miami by 2028 (here)