Venture Geopolitics Issue 21
11 November 2025
The fight for truth and information is caught between technology, corporations and governments. In the UK, the High Court ruled that Stability AI hadn’t broken copyright law despite training its image generator on Getty Images, while in the US Meta’s detection tools were found to miss nearly all harmful content on its platforms and a new wave of ideologically charged chatbots is rapidly scaling. At the same time, the BBC — long a symbol of impartial journalism — saw its leadership resign amid controversy over editing a Trump speech, all while Anthropic pledged to preserve its AI models for posterity. All this raises a deeper question: who will take responsibility for the digital minds shaping our future?
IPOs / Public
Stocks eased from record highs: NASDAQ composite -3% over the week (here)
Still within growth-stock volatility norms. Nonetheless, headlines “SoftBank plunges 10%” & “Global stock markets fall sharply over AI bubble fears” show market remains skittish about valuations & AI bubble (here and here)
(Stocks bounced back yesterday, buoyed by US senate taking step to end longest government shutdown in history… here)
Beta Technologies (electric aviation) raised $1.02B at $7.4B – rare hardware IPO scale in a higher-rate tape (here)
BillionToOne (precision diagnostics) announced $314M upsized IPO at $4.4B after a 67% debut pop – diagnostics still price in a scarcity premium (here)
Pony AI (autonomous driving) & WeRide (autonomous driving) listed in Hong Kong, then dropped 20% & 22% (here)
Amazon sent a cease-and-desist to Perplexity (AI agents) over its Comet browser making purchases on users’ behalf. Perplexity accused Amazon of bullying (here)
Google plans orbital data centres from 2027. Cooling, land & energy constraints pushing compute off-planet! (here)
Google also in talks to deepen its Anthropic stake, potentially at >$350B (here)
Google has integrated prediction market data from Kalshi & Polymarket, a move toward mainstreaming event-based forecasting (here)
Apple close to paying Google ~$1B/yr for a 1.3T-parameter model to power Siri (here)
Microsoft has committed >$60B to neoclouds, gaining access to 200k Nvidia GB300s, on par with 2024 capex (here)
Microsoft built a simulated / fake marketplace (“Magentic Marketplace”) & found today’s agents are easy to manipulate, option-overwhelmed & collaboration-poor (here)
Epic Games & Google settled to open Android payments with 9–20% fee caps (increasing pressure on Apple’s App Store take rate) (here)
Michael Burry (big short fame) disclosed $1.1B shorts against Palantir (enterprise AI & defence) & Nvidia (AI chips). Meanwhile, Palantir posted $476M quarterly profit, ~3.5x YoY. Growth driven by suite of AI software- closed 53 deals worth >$10M in the most recent quarter vs. 16 year prior & has led to a lot of celebration of the FDR (forward deployed engineer role) Palantir pioneered (here and here)
Simultaneously, investors are selling off bonds from major US tech firms like Alphabet, Meta, Microsoft, Oracle amid growing concerns over their rising debt to fund AI infrastructure - pushing bond yield spreads to their highest level in months (FT)
Tesla shareholders approved Elon Musk’s pay package approaching $1T. Ties Musk’s reward to achieving ambitious 10-year targets: raising market value from $1.4T to $8.5T & delivering on robotaxis & humanoid robots. Vote underscores how Tesla’s governance culture has become entwined with the cult of its founder, as retail investors & a compliant board overwhelmingly supported despite mounting competition, slowing growth & governance concerns (here)
Meanwhile, Rivian (EVs) set a $4.6B plan for its CEO - incentives now look like sovereign wealth instruments than corporate rewards! (here)
Big Dogs
OpenAI flagged >$20B 2025 revenue & ~$1.4T compute commitments -morphing into sovereign-grade infra vendor - and Altman lots his rag when asked how he was going to pay the $1.4T in obligations given a mere $13B in current revenue (here and here)
On back of last weeks rumours, CFO says IPO “not imminent” while scaling datacentre spend (here)
Usage policy updates clarify people cannot use services for “provision of tailored advice that requires a license, such as legal or medical advice, without the appropriate involvement by a licensed professional” - more legal prophylaxis than pivot (here)
7 more families filed suit alleging its chatbot encouraged suicides & delusions after the company rushed its GPT-4o model to market without adequate safety testing in order to beat rivals like Google (here)
Anthropic announced it will preserve the weights of all major Claude models for the company’s lifetime & begin conducting post-deployment interviews with models before deprecation, documenting their reflections & preferences. Move acknowledges retiring AI systems can raise safety, research & ethical concerns - from avoiding misaligned “shutdown-avoidant” behaviour to respecting users’ attachment to older models & exploring possible model welfare issues. Fist public commitment by a major lab to treat model preservation and retirement as a moral issue (Zvi Mowshowitz)
Clio (legal software for law firms) raised $500M at $5B (here)
Beacon (software holdco) raised $250M to buy sub-$20M ARR products & refit them with AI (here)
Venture Capital
Roelof Botha ousted as Sequoia Capital managing partner by Alfred Lin, Pat Grady & Andrew Reed. FT reports the trio acted with the backing of wider partnership & Doug Leone, triggering what one insider described as a “revolt against an imperial style of leadership.” Another said, “Roelof’s IQ is off the charts, but he always needs to be the smartest person in the room — his emotional intelligence hasn’t kept up” (FT)
Munich Re Ventures (CVC) to wind down its $1.2B platform by Q22026 – more funds winding down but less unusual to see this from CVCs (here)
Balnord (frontier & dual-use) passed €70M target focused on Baltic security – Europe’s defence thesis keeps institutionalising (here)
Jamin Ball: software cycles toggle between 2 phases: expansionary phases (rapid, fear-of-missing-out buying with little cost discipline) & consolidation phases (scrutiny of vendors, cost-cutting, and platform bundling) (here)
Recent cycles: 2020–2021 & 2024–2025 are expansionary - first driven by low interest rates, now by AI hype. 2022–2023 was a consolidation period focused on efficiency & vendor reduction
AI boom created an environment where startups in the same niche are all growing, fueled by investor enthusiasm & buyers’ urgency to adopt AI tools - only a few will ultimately survive once spending tightens
True strength shows in consolidation phase, when weaker players fade. Investors must identify companies built on durable product quality & execution, not just those lifted by short-term expansionary tailwinds
a16z argues that despite cries of “bubble” today’s valuations are largely supported by fundamentals: S&P500 profit margins sit around 14.5% - highest ever - so elevated P/E ratios reflect stronger profitability, not exuberance. Adjusted for margins, the current market multiple (~17.75x) aligns with 2015–2018 levels, suggesting investors are paying up for sustained earnings power rather than fueling a speculative bubble (here)
Stripe data show US software startups massively outpacing Europe since 2020 – ~1400% revenue growth vs ~600% EU & ~500% UK. Patrick Collison (Stripe's co-founder and CEO) attributes to the vast, unified home market with appetite for innovation & less regulatory drag, rather than just AI. When you strip out AI, US startups still have a pretty wide lead (here)
Robinhood is opening the door for retail investors to chase the AI boom, launching a leveraged fund investing in a handful of top private AI firms. CEO Vlad Tenev says it democratizes access; critics call it reckless, noting Robinhood’s 255% stock surge, Q3 revenues up 100% & crypto trading up 300% may be fuelling dangerous exuberance (here)
Venture Geopolitics
NVIDIA CEO Jensen Huang recently held a “beer summit” (informal meeting to build rapport) in South Korea with Samsung Chairman Jay Y. Lee & Hyundai Motor Executive Chairman Chung Euisun, followed by a major announcement with President Lee Jae Myung of a sovereign AI partnership (Interconnected)
Deal includes 260k NVIDIA Blackwell GPUs worth at least $7.8B, alongside a commitment to triple South Korea’s AI investment in its 2026 national budget
Partnership highlights Huang’s strategy of turning “sovereign AI” into an offensive global sales & partnership model for NVIDIA
Deal underscores South Korea’s readiness & ambition - built on its 95%+ digital penetration, world-class semiconductor, telecom & cloud ecosystem & capacity to develop “Sovereign AI with South Korean Characteristics”
U.S. is reportedly preparing to ban sales of TP-Link routers, citing national security risks tied to the company’s Chinese origins, despite TP-Link’s claims of independence & U.S. incorporation. Move follows evidence Chinese state-backed hackers exploited TP-Link devices - though similar issues have affected Cisco & Netgear - highlighting concerns over foreign control of network infrastructure. As governments weaponize supply chains and firms are compelled to declare national loyalties, the TP-Link case reveals how the line between commerce & statecraft is eroding (here)
European Commission to pause parts of the AI Act via a simplification package on 19 Nov – under Big Tech & US pressure to boost EU competitiveness vs China (& US) (here)
In force since Aug 2024, most rules bite 2026–27; one-year grace for high-risk systems
Fines for transparency breaches deferred to Aug 2027 to give providers time to adapt
EU AI Office to centralise enforcement & streamline reporting
Meta’s internal research found that teens who felt bad about their bodies saw three times more “eating disorder-adjacent” & provocative content on Instagram than peers who didn’t (here)
Meta’s detection tools missed 98.5% of potentially harmful posts, many showing sexualized or distressing imagery not banned by platform rules
Experts warn chatbots now mirror media polarization, turning truth into another partisan choice online. Enoch, Arya& Grok claim to counter bias found in mainstream models such as ChatGPT & Gemini - unlike these established chatbots, which aim for neutrality but face liberal bias accusations, newer ones openly embrace ideological stances, especially right-wing views. Tests by The NYT found politically divergent, often inaccurate outputs (here)
Matt Levine notes First Brands allegedly used fake invoices - inflating a $179.84 sale into $9,271.25 and selling ~$2M of real receivables for ~$11M in loans (here)
Case exposes how private debt & factoring markets, built on trust & minimal verification, let billions move on spreadsheets & PDFs - efficient in good times but dangerously open to manipulation & double-pledging
WSJ reports “a string of alleged frauds by corporate borrowers is spurring a reckoning across Wall Street... sending real ripples in the credit markets,” prompting tighter due diligence & new lender task forces
FT adds, UBS chair Colm Kelleher warns US insurers are “shopping for ratings” on private-credit loans via small agencies—echoing 2007–08-style ratings arbitrage
Vulcan Elements & ReElement sealed $1.4B with Trump administration to onshore rare-earth magnets – reshoring as state venture capital (here)
US rubber supply chain deemed a national-security risk – 87% of key additives imported, 43% from China (here)
Nscale (UK data-centre infra) demonstrated cracks: “IP lawsuit”, “internal strife” & a “botched acquisition” according to Sifted, even as it leads Britain’s supercomputer project (here)
The Economist argues that despite gloomy domestic sentiment, Britain’s global reputation is holding up strongly. Surveys by the British Council & Brand Finance rank UK among the top 3 most attractive & influential nations worldwide. 4 pillars: renewed diplomatic clout post-Brexit under PM Keir Starmer, with strong ties to both Europe & US; continued business appeal, with UK ranked #2 for investment by PwC; unmatched cultural reach through media, music, & film; rising tech prowess - especially in AI (here)
All that said…while the BBC remains the world’s most trusted international news brand, top-level resignations this week (over editing a Donald Trump speech) included Director General & Head of News. Highlights: (1) lack of trust in journalism, (2) the US president’s continued global impact on media (here)
Strategic Sectors
AI
Moonshot (foundation models) released Kimi K2 Thinking open-source at just ~$4.6M training cost! (here)
Morgan Stanley sees data-centre capex rising from $245B (2024) to ~$700B by 2027 (here)
UK High Court ruled that “Stability AI was not liable for copyright infringement,” despite training its image generator on Getty Images’ content. Judge Smith called the decision “historic & extremely limited in scope.” The ruling turned on the fact that the AI model does not “store copies” under UK law (here)
Shopify reports AI-tool traffic up 7x since January, AI-driven purchases up 11x – agentic commerce gaining traction (here)
Hiring signals (Azeem Azhar)
Relative to the pre‑LLM period, workers in the highest ability quintile are hired 19% less often.
So far this year, overall job postings have dropped ~8% YoY, but AI roles on the rise. ML engineer postings +40%
Dr Seth Dobrin: the future of AI is smaller, smarter and finally here. SLMs (Small Language Models) – smaller, specialised, & efficient; 70–80% of LLM performance at a fraction of cost (here)
· The old “scale equals quality” myth is breaking down — bigger models often add huge cost for only tiny gains in accuracy
· SLMs focus on specific domains (e.g. finance, healthcare) rather than general knowledge
· SLMs cut cost 93%, energy use 60% & suit on-prem GDPR compliance – specialisation beats generalisation
· SLMs’ smaller size makes them easier to fine-tune for company-specific tasks, improving accuracy and reliability
· Big Tech pivots (Phi, Gemma, Llama) show efficiency is back in fashion – sustainable AI becomes investable
· The SLM market is expected to grow nearly 29% a year, reaching about $5.5 billion by 2032
· These developments democratise AI, reducing reliance on Big Tech’s huge, centralised systems
Cybersecurity
Google x Wiz deal (largest cyber deal in history) cleared by US DOJ for $32B (here)
1Password at ~$400M ARR & FCF positive – credentials are back in focus, potential path to IPO? (here)
Armis (asset security) raised $435M at $6.1B – OT & identity sprawl drive budgets (here)
Dataminr bought ThreatConnect for $290M – fusing external & internal threat intel (here)
Defence
Drone start-up Stark, backed by Thiel, Sequoia and NATO Innovation Fund, saw its drones fail in UK & German army trials - called a “disaster” by some, others note “the reality of testing at speed.” Despite setbacks, Stark - valued at $500M after raising major VC funding—remains in contention for €300M Bundeswehr contracts & part of Europe’s urgent drive to scale defence tech amid the war in Ukraine (Resilience Media, Harici)
Neros (low-cost military drones for the U.S. & allied forces) raised $75M led by Sequoia (here)
Crypto
Coinbase launching regulated token-sales platform, allowing U.S. investors to buy digital coins before they’re listed — first ICO-style offering since 2018 (here)
Japan now among countries providing government resources to Bitcoin mining (here)
Ripple raised $500M at $40B from Citadel & Fortress – balancing XRP with its RLUSD stablecoin (~$50M supply) (here)
Block (payments firm founded by Jack Dorsey) has expanded Bitcoin payments to over 4M merchants worldwide through Square PoS network (here)
EVs
Amazon placed UK’s largest-ever electric HGV order with Mercedes-Benz eActros 600s – zero-emission freight hits scale (here)