Venture Geopolitics Issue 36
24 Feb 2026
Impact is in the technology ecosystem’s DNA. It is easy to forget that amid the noise — the geopolitical friction, the concentration of power, the sense that the tools reshaping our world are increasingly wielded by a shrinking number of hands. But this week, I found myself noticing small but meaningful signs that the winds may be shifting.
Milton Friedman gave capitalism its operating system fifty years ago. His doctrine — that the social responsibility of a company is solely to make profits — made financial return the singular north star of corporate life. Immorality (amorality?) in the corporate sphere is rarely punished; it is often rewarded. Corrections tend to arrive only when people in positions of power decide to act.
This week had a few of those moments.
Discord severed ties with a Thiel-backed identity verification provider after researchers flagged surveillance infrastructure buried in its code. Anthropic pushed back on the Pentagon over autonomous weapons and domestic surveillance, drawing a values line even at potential commercial cost. Palantir, having quietly accumulated no-bid UK Ministry of Defence contracts worth hundreds of millions and some questionable business practices, is down 38% from its 52-week high. And while Russia has reopened a criminal case against Telegram founder Pavel Durov, his face in the headlines is a welcome reminder of sincere, values-driven entrepreneurialism holding a place on the global stage.
As power concentrates in technology platforms, the moral guardrails provided by government become less effective and the moral responsibility incumbent on the platforms grows. A company controlling critical infrastructure is not simply a business. It is, whether it chooses to be or not, a participant in governance. The Friedman doctrine was always reductive. At this scale, applied by platforms without recognition of their broader role in society it borders on negligence.
Capital still flows toward returns, and power still tends to compound in the hands of those who already hold it. But necessity has always been the mother of invention. And when institutions and markets start saying no, it is worth paying attention.
What makes this moment genuinely energising — at least from where I sit — is the calibre of founders building: clear-eyed about risk, serious about responsibility, determined to build companies that are both commercially excellent and genuinely good. The intentionality of those choosing to tackle the hardest problems is as high as I have seen. In a year like this one, that is not a small thing.
IPOs / Public
“The IPO rebound has felt ‘right around the corner’ for two years. It’ll stay that way for another month” (here)
UK cyber insurance provider CFC tapped banks for a possible IPO (here)
Finnish quantum computer startup IQM plans to go public by merging with Real Asset Acquisition Corp, a “blank check vehicle” based in the US. The deal values IQM at $1.8B & could generate > $300M in fresh funding (here)
Robinhood is launching $1B closed-end private markets fund via IPO – structure gives US retail investors exposure to private companies via a listed vehicle. Potentially opens floodgates for retail access to late-stage tech equity but also transfers valuation opacity from private markets into public wrappers (here)
Tunguz notes how anticipated mega-cap AI IPOs will break the float model. SpaceX targeting ~$1.5T, OpenAI ~$1T, Anthropic ~$380B = $2.9T combined. At a standard 15–25% float, would need to raise $432–576B in a single quarter. For context, entire US IPO market raised $469B from 2016–2025! Standard floats are arithmetically impossible. Expect 3–8% instead. But S&P500 requires 50% public float for inclusion…meaning none qualify initially. When they do, passive funds managing ~$20T must buy. Index funds cannot raise cash – they sell incumbents. At $1.6–2T, SpaceX would rival Meta for spot #6 in the S&P. The mechanics become reflexive: mega-caps sold to fund new mega-caps. The disruption is structural. These IPOs won’t just reprice tech – they will test index construction itself (here)
Klarna (BNPL) shares fell 27% after reporting rising bad loan costs, with credit losses accelerating as consumer delinquencies tick up (here)
RaspberryPi (single-board computers) added >$1B in market cap in weeks on enthusiasm that its low-cost hardware becomes a default for running lightweight AI agents like OpenClaw (here)
Reddit's stock fell 42% in 5 weeks as US logged-in daily users flatlined at 23M through 2025. While advertising revenue surged 74% to $2.1bn last year, Reddit generates just $86 per US user vs. Facebook's $303. Reddit also fined £14.47M by the UK's data watchdog for unlawfully using children's personal information (here)
Last week, Meta & Nvidia renewed a “multiyear, multigenerational strategic partnership” with Meta recommitting to using “millions” of Nvidia AI chips. This follows reports Meta was exploring Google TPUs (here).
Today, Meta announced a multi-billion dollar chip deal with AMD that could lead the firm to acquire 6GW of chip capacity. AMD also issued Meta with a performance based warrant, giving the option to acquire up to 160M AMD shares/10%. Shares in AMD surged ~14% on the news. Another demonstration of hyperscalers using capital, partnerships & supply commitments to secure strategic compute capacity in an era where AI infrastructure is becoming a pillar of national power (here)
Big Dogs
OpenAI is replacing last year’s highly publicised $100B “memorandum of understanding” with Nvidia with a simpler $30B equity investment as part of a >$100B round valuing OpenAI at ~$730B pre-money (here)
Original deal would have seen Nvidia invest $10B tranches tied to OpenAI’s compute expansion to ~10GW of capacity. Framework abandoned in favour of straight equity cheque
OpenAI told investors it plans to spend ~$600B on compute through 2030 across Nvidia, Amazon & Microsoft infrastructure
OpenAI raised revenue forecasts +27% through 2030 but now expects to burn >$111B more cash than previously projected. 2025 revenue reached $13.1B, with projections of $30B in 2026 & $62B in 2027
WAU hit 910M, short of 1B target for end-2025
Inference costs 4x in 2025, compressing GM to 33% vs 40% prior year
Training costs alone are forecast to total ~$440B through 2030
Signed long-term partnerships with McKinsey, Accenture, BCG & Capgemini to distribute its “Frontier” agent management software, explicitly targeting SaaS incumbents. Salesforce, Snowflake & ServiceNow shares fell on news
AMD agreed to backstop a $300M loan for Crusoe (AI cloud infrastructure) to buy AMD chips, effectively guaranteeing utilisation in order to win share from Nvidia (here)
Anthropic alleged DeepSeek, Moonshot AI & MiniMax created >24,000 fraudulent Claude accounts & generated hundreds of thousands – potentially millions – of interactions to distil Claude outputs into cheaper rival models. While distillation can legitimately produce smaller, more efficient systems, Anthropic argues the tactic was used to repurpose its work into competing products at lower cost (here)
Anthropic is clashing with the Pentagon over acceptable military uses; it seeks carve-outs for autonomous weapons & mass domestic surveillance, while the DoD insists on access for “all lawful use cases” & has floated labelling Anthropic a “supply chain risk” (here)
Anthropic launched Claude Code Security, triggering sharp sell-offs across cybersecurity equities as markets extended their reactive AI-disruption thesis (here)
Anthropic planning a $5-6B share sale allowing employees to sell stock at ~$350B, following funding round at $380B (here)
Saudi Arabia's state-backed AI company Humain invested $3B into Musk's xAI contributing significantly to its $20B funding round — with shares subsequently converted into SpaceX stock. The deal deepens Gulf investment in Silicon Valley AI as SpaceX eyes a potentially record-breaking $50B IPO, surpassing Saudi Aramco's 2019 flotation (here)
Starlink hit 10M subscribers in Feb, but has aggressively slashed prices - offering plans as cheap as $50/month & giving away terminals costing up to $600 to manufacture - to secure market share before Amazon's rival Leo service launches. Analysts warn margins could suffer significantly once Amazon, leveraging its vast Prime & AWS customer base, enters the market (here)
Venture Capital
Carta State of US Private Markets (here)
Seed
2025: Median post-money ~$21M (minimal AI premium); 90th percentile ~$59–60M
Series A
2025: Median ~$60M overall; AI startups valued about 25–38% higher than non-AI (~$75M vs ~$55M median); upper-end valuations reached ~$300M (≈90–95th percentile primary rounds)
Median time to series B has doubled since 2019 to 2.5 years
Series E+
2025: Median valuation $1.49B overall; strong AI divergence with AI ~$4.09B vs non-AI ~$616M median (~193% premium); upper-tail valuations reached ~$10.7B (primary) & ~$20B for bridge rounds (~top decile+), reflecting extreme concentration in breakout companies
In 2025, 58% of all cash raised on Carta at Series D went to AI startups, along with 51.7% of all cash raised at Series E+
Over the past year, median dilution on all rounds from seed through Series C has fallen from about 18% to 16%, continuing a years-long trend. 2 years ago, median dilution across those stages stood at 19%.
< 14% of all rounds raised on Carta in Q4 were down rounds, lowest quarterly figure since start of 2023
New Funds:
Thrive Capital (global) closed $10B across $1B early-stage & $9B growth
Battery Ventures (global) raised $3.25B Fund XV
Mundi Ventures (fintech, climate, insurtech) €750M first close Fund V
Dragonfly (crypto fund) $650M Fund IV
Frist Cressey Ventures (healthcare) $425M Fund IV
Quantonation Ventures (quantum) €220M second flagship, now the largest dedicated quantum fund in Europe
Tenet (AI roll-ups) targeting €80M debut fund
Redrice Ventures (consumer) £75M Fund II with up to £45M from British Business Bank
Emerald Technology Ventures (Physical AI) €52M
OSS Ventures (industry 4.0) €40M first close toward €75M
Benchmark hired Jack Altman as GP (here)
Bpifrance deployed €72B in 2025 versus ~€11B from French VCs, reinforcing how sovereign balance sheets are as if not more influential than private venture in Europe’s capital stack (here)
Private lender Blue Owl changed redemption rules so investors can no longer withdraw fixed amounts quarterly, alarming markets & reviving concerns about liquidity risks in the $Tn private-credit sector (here)
Secondary sales of private startup shares have grown from 3% of all VC exits in 2015 to 31% today (here)
Regulation
Meta is deploying $65M - its largest ever election spend - across 4 super PACs to back AI-friendly state politicians in the 2026 midterms. Two new PACs, “Forge the Future Project” (Republican) & “Making Our Tomorrow” (Democrat), reflect a deliberately bipartisan strategy. The goal is explicit: defeat state legislators advancing AI regulation before a damaging patchwork of state laws hardens into precedent (here)
Venture Geopolitics
US Supreme Court ruled Trump exceeded authority on global tariffs; a revised global 10%, then 15%, tariff announcement followed within 24 hours. Tariffs have come in at 10% today, while the White House is working on an order to raise the rate to 15%, but timeline is unclear. The court ruling has triggered refund disputes & renewed volatility, with FedEx one of the first major American companies to sue the U.S. government for a refund. Britain is potentially one of the bigger losers if tariffs climb to 15%, having secured a preferential 10% rate last May. Downing Street warned “nothing is off the table” while simultaneously insisting industry doesn’t want a trade war. Meanwhile China & Brazil would actually benefit from the new regime. The EU has hit pause on implementing their trade deal with the US, as they try to understand what Trump will do next (here)
Europe has spent recent years realising its dependencies — on Chinese rare earths, Russian gas & increasingly US military & tech provision. Analysts argue it has been too slow to recognise the leverage it holds in return. Research from the Institut Montaigne, the Geostrategic Europe Taskforce & German think-tank Dezernat Zukunft identifies significant EU chokepoints: Dutch firm ASML holds a near-monopoly on extreme ultraviolet (EUV) lithography machines essential to advanced chip manufacturing worldwide; Europe controls 80% of US uranium imports; Siemens dominates turbines used in US data centres; & the EU has 41 supply categories where China depends on European imports for over 80% of its needs, & 67 such dependencies for the US — spanning insulin, pharma intermediates, medical tech & specialist industrial machinery. The argument is that Europe has focused purely on defensive resilience (reducing its own dependencies) without leveraging these strengths offensively. Practically, this could mean lowering the threshold for the EU’s “anti-coercion instrument” — currently reactive — to allow pre-emptive deployment against threatened economic coercion. The EU & UK would also need to coordinate use of these tools (as they have on Russia sanctions). The core point: having leverage & being willing to use it are two different things & Europe has yet to bridge that gap (here)
European military officials are concerned that discussions about “tech sovereignty” will undermine security, given sheer reliance on US software for critical systems. Beyond intelligence systems & command & control systems, US tech companies are essential in communications, intel gathering & data storage. “Most European platforms rely on an American back—end”, with Lockheed Martin’s Aegis system a classic example. The European Commission will present a “tech sovereignty package” this spring, but military officials at the Munich security conference got nervous given the overly ambitious goals vs hard realities (here)
Discord has cut ties with Persona, a Thiel-backed identity-verification company, after researchers discovered code linked to government-style surveillance infrastructure & raised fears about how users’ biometric & ID data might be handled; the partnership lasted less than a month & came amid backlash over Discord’s planned age-verification system & a recent breach affecting tens of thousands of users’ data. The story highlights a growing tension across the internet: platforms are being pushed by regulators to verify ages for safety, but doing so increasingly requires sensitive biometric data handled by third-party vendors - creating new privacy & surveillance risks (here)
India hosted the Global AI Summit in Delhi using it to push leading AI firms — including OpenAI & Google — to open-source models for social purposes (healthcare, education, agriculture). PM Modi argued AI "will only benefit the world when it is shared." The effort largely failed. White House science chief Michael Kratsios told attendees the US "totally" rejected centralised AI oversight. American tech companies, feeling no regulatory pressure from Washington, showed little appetite for even baseline commitments. India secured only a voluntary pledge for companies to share usage data & multilingual model effectiveness metrics. The summit was also logistically troubled — gridlocked streets, long queues & high-profile no-shows including Nvidia's Jensen Huang & Microsoft co-founder Bill Gates (here)
India joined Pax Silica, a US-led semiconductor supply chain coalition spanning Japan, South Korea, the Netherlands, Israel & the UK; Taiwan remains absent - alliances increasingly reflect political alignment rather than pure industrial logic (here)
Sweden published a national AI strategy. The strategy prioritises adoption over research - an explicit rebuke of Canada's research-first approach, which it warns now risks falling behind. Currently ranked 44th globally on government AI strategy despite strong fundamentals, Sweden aims to reach the top ten by 2025, anchoring sovereign capability in shared infrastructure, democratic values & cross-sector collaboration before dependency on foreign platforms further hardens (here)
The New York Times published “The Looming Taiwan Chip Disaster That Silicon Valley Has Long Ignored”, highlighting Taiwan as a critical single point of failure: it produces ~90% of advanced semiconductors - including ~97% of high-end chips - underpinning ~$10T of global economic activity. The article warns a Chinese incursion or blockade could be imminent, with Western tech firms underprepared. Speaking at the WEF, US Treasury Secretary Scott Bessent called Taiwan-based chip production the world economy’s biggest vulnerability. US officials estimate a blockade could cut US GDP by 11% ($2.5T) & China’s by 16% ($2.8T), exceeding the 2008 crisis, while industry analysis suggests disruption could trigger the worst downturn since the Great Depression. Despite ~$50B in CHIPS Act subsidies & projected $200B US investment by 2030, reliance persists as US chips cost >25% more & lag technologically; most firms hold only months of inventory & even US-made AI chips still require Taiwanese packaging, meaning dependence will last for years (here)
Germany's relationship with China has soured as Chinese competition hollows out German industry, with a €90B trade deficit. Chancellor Merz is set to visit Beijing this week seeking "fair competition" (here)
China formally designated brain-computer interfaces as a strategic sector last year & now Beijing is accelerating brain-computer interface development via funding, looser regulation & national strategy goals to create world-class firms by 2030. Start-up NeuroXess has already shown early patient success controlling devices after implantation. Rapid trials & investment signal Beijing’s ambition to rival US leaders like Neuralink, though current focus remains medical use for severe disabilities rather than consumer enhancement (here)
China has restricted exports of rare earth magnets & other critical materials to dozens of leading Japanese companies. China’s commerce ministry said it would freeze the flow of “dual use” materials i.e. rare earths like dysprosium, yttrium or samarium which play tiny but vital roles in cars, planes, weapons & consumer electronics. Restrictions were aimed at curbing Japan's "remilitarisation" & nuclear ambitions, China's commerce ministry said. Takaichi has accelerated a military build-up launched in 2023 that will 2x Japan's defence spending to 2% of GDP by the end of March, making the country one of the world’s biggest military spenders despite its pacifist constitution (here)
Japan successfully retrieved rare-earth-rich mud from 6,000 metres beneath the Pacific — deeper than Mount Fuji is tall — as Beijing tightens export restrictions on minerals vital to missiles & drones. China controls over 90% of rare earth refining globally. Commercial extraction at the Minamitorishima site is unlikely before 2028 & costs remain formidable (here)
Britain’s largest public research funder UKRI’s has committed £1.6B to AI between 2026 & 2030, spanning cancer detection, clean energy & more efficient public services. Specific commitments include £137M for AI-enabled drug discovery & £36M to upgrade Cambridge’s DAWN supercomputer, alongside expanded doctoral routes co-designed with industry & career frameworks for data scientists & AI ethics specialists (here)
The UK’s largest banks are also convening to explore a domestic alternative to Visa & Mastercard (which handle ~95% of UK card payments) amid fears US political leverage could disrupt payments (here)
Strategic Sectors
AI
Another viral (“AI Doomer”) article that everyones talking about which again shook markets - this time from Citrini Research. Read it here & some commentary here, here & here
AI capex is estimated to account for 64-80% of US Q425 growth (here)
Developers are seeking credit ratings - before construction ends - to unlock billions in funding for AI data centres. Fitch rated 35+ projects in 9mo with average ~$3B deal sizes. Ratings allow banks to sell loans to institutional investors, easing financing pressure from massive build-outs. Projects rely heavily on long-term contracts with Big Tech tenants; however, specialised AI facilities carry higher risk if demand or hyperscaler support weakens (here)
a16z charts of the week shows just how good LLM retention rates are, how strong sales cycles are for publicly traded vertical SaaS despite recent headwinds & how not only are frontier models getting better & better, but open source alternatives are keeping pace (here)
US consulting growth is forecast at 7% in 2026, the fastest since post-Covid, driven by AI implementation work (here)
Meta tracks AI usage & productivity via internal tool "Clarity," tying adoption to promotions & pay reviews. Meta & Accenture do similarly (here)
Notable deals:
xAI (frontier AI lab) raised $3B Series E to expand compute infrastructure, reinforcing capital intensity at the frontier
Ineffable Intelligence (reinforcement learning startup), founded by ex-Deepmind David Silver in London, is reportedly seeking $1B at $4B to pursue “experience scaling,” shifting advantage toward those controlling simulation environments
World Labs (3D world models) raised $1B to build embodied training environments for robotics & science
Fluidstack (neo-cloud provider) is in talks to raise $100M at $7.5B, aggregating under-utilised GPU capacity as alternatives to hyperscalers gain relevance
Cybersecurity
Palo Alto Networks (cybersecurity platform) agreed to acquire Koi Security (AI agent security) for ~$400M, folding AI-agent visibility into Prisma AIRS & Cortex XDR (here)
Defence
German defence start-up Stark faces scrutiny over investor Peter Thiel’s political ties as MPs consider a €2.9B drone contract. Critics fear foreign influence amid Europe’s push for strategic autonomy, though the firm says Thiel holds <10% & no operational control. The controversy highlights tensions between rapid defence innovation, foreign capital & political trust in Europe’s rearmament efforts (here)
British XP Power-labelled components have appeared in North Korean missiles used by Russia in Ukraine despite no evidence of direct involvement, with 799 shipments worth ~$2.5M reaching Russia via intermediaries since 2022 (here)
Palantir in headlines for 2 consecutive uncompeted UK MoD contracts, raising procurement transparency concerns. In Nov 2022, Palantir was awarded a 3-year, £75.2M contract by the MoD — unadvertised, with no competitive tender, justified on grounds that only Palantir could meet the requirement. No meaningful exit plan or successor-competition provisions were built in. When the contract expired in Dec 2025, the MoD awarded a follow-on 3-year contract worth £240.6M — again directly, without competition — citing that MoD systems were now so embedded in Palantir’s architecture that switching supplier would require full rebuilds, reaccreditation at classified security levels & staff retraining. Critics call this a textbook “land-&-expand” strategy & the author (a former govt lawyer) warns the same cycle could repeat every 3 years indefinitely. The MoD has not published the contract specification, making independent scrutiny impossible. “Unless evidence to the contrary is provided to the public, it appears as if the government department responsible for defence has commercially surrendered to a single service provider” (here & here). Incidentally, shares of Palantir are under pressure, dropping more than 38% from its 52-week high of $207.52 (here)
Palantir is suing small Swiss magazine Republik after it revealed Swiss authorities repeatedly rejected the data intelligence firm on national security grounds — including an Armed Forces review warning US agencies could access sensitive military files. The lawsuit, seeking a right of reply rather than damages, has backfired spectacularly, amplifying the story internationally (here)
Energy
Notable deal: Heron Power (clean energy developer) raised $140M to accelerate grid capacity build-out (here)
Crypto/Stablecoins
Binance reportedly fired investigators who uncovered > $1B in cryptocurrency transfers linked to Iranian networks, part of $1.7B in flows during 2024–25. The exchange previously paid a $4.3B fine for sanctions breaches (here)
US-backed officials & Trump’s Board of Peace are considering a dollar-pegged stablecoin for Gaza after war damage crippled banking & cash supplies. The aim is to enable digital payments & limit cash flows potentially used by Hamas, though concerns include weak infrastructure, power cuts & economic separation from the West Bank (here)
Quantum
Notable deal: Pasqal (neutral-atom quantum computing) reportedly raising €200M at $1B+, potentially Europe’s latest quantum unicorn & one of the few frontier domains where Europe retains depth (here)
EVs & AVs
Uber launched "Uber Autonomous Solutions" offering robotaxi operators insurance, fleet financing & monitoring software, positioning itself as the commercial backbone of the AV industry. The company has signed over a dozen partnerships (including with Waymo & Baidu) & committed to deploying autonomous vehicles across 15 cities globally, including London, by end of 2026 (here)
Waymo’s entire remote guidance operation runs on just 70 human operators - a 43:1 car-to-human ratio. GM’s robotaxi service Cruise had 1.5 staff per car (here)
Space
OpenAI CEO Sam Altman argues space-based data centres are ‘ridiculous’ today despite hype tied to Musks vision: launch costs exceed Earth power costs & broken AI chips couldn’t easily be repaired in orbit. While the idea may work long-term, current tech & economics make it unrealistic (here)