Venture Geopolitics Issue 23
25 November 2025
November’s flurry of frontier–model releases (GPT-5.1, Opus 4.5, Gemini 3) has marked a turn in the AI power landscape. The contest is not just about topping leaderboards; it’s about controlling the rails of distribution. Google’s Gemini 3 makes this shift unmistakable – Google isn’t just advancing capability – it’s fortifying an ecosystem. The result: a product wave that has impressed markets and quietly rattled OpenAI.
As the market celebrates Google’s stack, the threats of concentration remain unsettling - another Cloudflare outage highlights the fragility of the internet’s backbone, while a look inside CoreWeave reveals an infra darling burdened by $14B in debt and dependence on hyperscalers (who are simultaneously racing to build their own sovereign compute stacks).
The boom is real – but so are the fault lines.
Meanwhile, the line between Big Tech and politics has spread to Europe. Washington’s intention to tie Europe’s treatment of Google, Microsoft and Amazon to trade concessions on steel and aluminium represents a direct fusion of antitrust and diplomacy.
In this new order, those who control the data, the compute and the defaults don’t just win markets – they set the tempo of global power.
IPOs/Public
More public market deterioration last week despite Nvidia’s strong results. Sentiment improved later in the week/this week thanks to rising expectations of a Fed rate cut (here)
No major IPOs scheduled for Thanksgiving week
SaaStr summary “the software IPO freeze”: 2021: 46 IPOs, 2022: 0, 2023: 1, 2024: 4…8 so far in 2025 i.e. this year is running at ~17% of 2021’s pace & only just back to pre-2019 “normal” of 9–10/year (despite 1,000+ unicorns in the wings…) (here)
Kraken (crypto exchange) filed for US IPO as digital-asset firms rush to list ahead of potential regulatory whiplash around 2026 midterms. On back of raising $200M at >$20B (here)
Cloudflare suffered widespread outage that took down X, ChatGPT, Spotify & others. Latent bug in a bot-mitigation service was triggered by a routine config change i.e. not a cyberattack. Yet another reminder of systemic exposure (here)
Nvidia reported above expectations: $57B Q3 revenue, $32B net income & guided to $65B next quarter. The stock’s initial pop faded in a broader tech selloff, underlining how fragile AI sentiment has become (here)
Meta scored major win as federal judge threw out FTC’s attempt to unwind Instagram & WhatsApp (here)
Google’s Gemini 3.0 launch impressed the market & crystallised next phase of AI race
Google spending ~$93B/year in capex, running its core Gemini stack on in-house Ironwood TPUs that match Blackwell-class performance & pushing distribution through Search, Android & Workspace, making “good enough inside the product you already use” arguably a more important battleground than benchmark leaderboards (here)
Oriol Vinyals (Google DeepMind) says the performance delta between Gemini 2.5 & 3.0 is one of the largest they have ever seen purely from better pre-training & post-training (here)
Benioff: “Holy shit. I’ve used ChatGPT every day for 3 years. Just spent 2 hours on Gemini 3. I’m not going back. The leap is insane — reasoning, speed, images, video… everything is sharper & faster. It feels like the world just changed, again” (here)
Amin Vahdat (Google’s head of AI infra) told employees the company needs to 2x serving capacity every 6 months to keep up with demand (here)
CoreWeave (AI cloud) profiled by The Verge as the AI industry’s “ticking time bomb”: ~$1.4B Q3 revenue sitting on ~$14B of debt & $34B of lease commitments, paying >$300M in quarterly interest while relying on a handful of hyperscaler customers who are all building their own chips & data centres (here)
Big Dogs
OpenAI’s GPT-5.1 launched as a same-price upgrade with better instruction-following, more grounded reasoning & new personality presets
Now chasing Google, Altman forecasts “rough vibes” in a leaked memo. He warned employees that OpenAI’s revenue growth could slow to single digits by 2026, according to The Information (here)
OpenAI & Intuit (personal finance software) struck a vertical finance deal to pipe users’ financial data into ChatGPT, extending OpenAI’s enterprise beachhead into consumer money management (here)
ChatGPT Group Chats rolled out globally (here)
Anthropic announced Claude Opus 4.5, its third major AI model launch in 2 months. Anthropic says ideal users include professional software developers & knowledge workers like financial analysts, consultants & accountants (vs ChatGPT’s heavily consumer focussed upgrades) (here)
Microsoft & Nvidia backing Anthropic: reportedly committing up to $15B of new investment, pushing valuation to ~$350B (here)
xAI in talks to raise $15B at ~$230B, largely to fund “Colossus” supercomputer cluster in Memphis; yet another data-point that AI capital allocation now looks more like sovereign-scale industrial policy (here)
Databricks reportedly raising at valuation >$130B; bet here is that whoever owns the data plane for enterprises will be structurally indispensable, regardless of which frontier model is fashionable (here)
Lambda raised $1.5B+ taking total funding to $2.3B, to expand its GPU rental business; the neocloud trade remains simple: borrow money, buy GPUs, rent them to hyperscalers & model labs, hope the cycle outlasts your debt maturities (here)
Revolut completed a secondary share sale valuing it at $75B - a 66% jump from last year & underlining the rapid growth of Europe’s most valuable financial technology company (here)
Kalshi (regulated prediction markets) raised $1B at $11B less than 2 months after a $300M round at $5B; in a world where volatility is ambient, markets for “what happens” are attracting substantial capital (here)
Luma AI (multimodal world models) raised $900M at $4B+ to keep building photorealistic 3D world models on top of everyday video. Backing from Saudi Arabia’s sovereign wealth fund (Humain), underscoring the regions intention to become a heavyweight in AI (here)
Genspark (AI workplace agents) raised $275M at $1.25B to automate day-to-day office workflows with agents that span scheduling, email & internal tools (here)
Suno (AI music platform) raised $250M at $2.45B & has already hit $200M ARR, but faces lawsuits from major record labels over alleged unauthorised use of copyrighted music to train its models (here)
RapidSOS (real-time emergency data) raised $100M, taking total funding >$450M, to keep wiring emergency services into real-time data feeds – a textbook case where VC is quietly upgrading critical national public infra (here)
Venture Capital
Atomico published annual State of European Tech (here)
Europe’s tech ecosystem had a leap forward, valued at less than $1T a decade ago to ~$4T now
In 2025, almost 60% more people in Europe started companies than in 2023
Highest number of founders starting new ventures in Europe than any other year (>27k). Company creation +35% YoY
Deep tech & sustainability two of the continent’s fastest-growing sectors: 36% & 18% funding respectively
But by Series C+, >30% of repeat founders relocate HQ outside Europe
US accounts for 2/3 of global private tech investment
Global exit activity reached $608B in 2025, up sharply from $364B in 2024 → US remained 50% of global exit value
Path to greater firepower for Europe lies in unlocking access to its deepest pools of capital: US pension funds allocate ~0.03% of AUM to venture, European counterparts invest ~0.01% (with 0.023% in France & Benelux and just 0.001% in the UK & Ireland)
Rory O’Driscoll, Jason Lemkin & Tomasz Tunguz on 20VC note how (1) Oracle’s CDS spreads are reportedly 3x those of Amazon & Microsoft, (2) subprime auto delinquencies are at record levels, (3) private-credit vehicles like Blue Owl’s non-traded BDC are freezing redemptions…the kind of micro stress points that tend to show up before the narrative shifts (here)
Venture Geopolitics
Commerce Secretary Howard Lutnick warns EU to relax tech regs & resolve cases against Google, Microsoft & Amazon if it wants lower tariffs on steel and aluminium. The move challenges EU regulatory sovereignty & shows how trade leverage, tech governance & geopolitical power now converge (here)
Amazon will invest $50B from 2026 to expand AI & supercomputing for US government customers, adding 1.3GW of capacity across its Top Secret, Secret & GovCloud regions (here)
EU plans to tighten foreign direct investment rules amid rising concerns over strategic dependence, security risks & Beijing’s leverage. Policymakers say the framework will curb regulatory “race to the bottom”. Chinese FDI jumped 80% to €9.4B in 2024 (here)
As anticipated, Brussels proposed easing GDPR data-sharing limits for model training & delaying strict AI Act rules until 2027 (here)
New research shows that 87 of top 100 Chinese AI researchers who were working in the US in 2019 are still at American labs, underlining that US progress in AI remains heavily reliant on Chinese-born talent even as political rhetoric hardens (here)
Oracle co-founder Larry Ellison is scaling back plans for a £10B science hub in Oxford after the departure of Sir John Bell; philanthropic mega-labs are proving as exposed to personnel & political risk as any startup, which matters when they are meant to anchor national research ecosystems (here)
EU cookie-banner fatigue may finally be ending as Brussels reviews the rules that created the pop-up spam in the first place. One proposal would be to stop people being bombarded by constant requests to accept or reject “cookies” when browsing the internet & instead consent or reject internet cookies for a 6-month period (here)
US officials are debating whether to let Nvidia sell its H200 chips to China, a potential reversal that would reveal just how hard it is to sustain strict tech controls once domestic lobbying, market pressure & allied complaints start to bite (here)
$100k robot dogs from Boston Dynamics are quietly becoming standard kit for some police departments, raising fresh ethical questions about surveillance & where to draw the line between assistive tech & “remote presence” in state power (here)
The Trump administration has officially disbanded the Department of Government Efficiency (DOGE). The initiative was expected to run for ~2 years (here)
France & Germany Join Forces with Mistral AI & SAP SE to Launch a Sovereign AI for Public Administration. This collaboration aims to deploy AI native solutions to enhance the efficiency, transparency & responsiveness of public services (here)
UK announced a major AI growth package combining billions in public–private investment. Measures include £100M for AI hardware startups via a new government “first-customer” scheme, £500M for new Sovereign AI Unit, expanded compute access for researchers & a new AI Growth Zone in South Wales backed by £10bn in data-centre investment (here)
Strategic Sectors
AI
Yann LeCun (world-model pioneer) is leaving Meta to launch a startup built around his JEPA world-model vision, after repeatedly calling LLMs a “dead end” for human-level intelligence; his exit is both a real loss for Meta’s long-term research & a signal that alternative architectures may finally have room to be productised outside Big Tech (here)
TikTok’s AI content dial will let users choose how much AI-generated content shows up in their feed; if widely adopted, this is effectively a user-level content-moderation preference & could become a de facto standard that other platforms are forced to copy (here)
OLMo 3 (Allen Institute for AI) launched as a frontier-scale open model with a fully open-sourced training pipeline. The strategic point is that Western firms & researchers get a genuinely auditable alternative to closed or foreign models, which matters for sovereignty (here)
AIG, Great American & WR Berkley are among insurance companies seeking permission from US regulators to exclude AI liabilities from policies (here)
Cybersecurity
Palo Alto Networks (security + observability) agreed to buy Chronosphere (observability) for $3.35B, fusing Chronosphere’s huge telemetry pipeline with Palo Alto’s AI-driven Cortex platform. Security, observability & AI are converging (here)
CrowdStrike says it fired a “suspicious insider” last month for sharing screenshots of internal systems, following claims by the hacking collective Scattered Lapsus$ Hunters that they had breached the cybersecurity firm (here)
Amazon has built an internal Autonomous Threat Analysis system that pits specialised AI agents against each other in realistic test environments to discover software vulnerabilities & propose defences at machine speed, freeing human security teams to focus on higher stakes attacks (here)
Defence
Helsing & Stark (European defence tech) both opened new UK facilities – Helsing a maritime manufacturing site, Stark its first drone factory – small but symbolic steps in Europe’s attempt to onshore more of its defence industrial base rather than relying on imported kit (here and here)
US company Twenty (offensive cyber tooling) raised $38M to help military hackers automate the process of finding & exploiting target networks. “AI for cyber-offence” is now a funded category (here)
NestAI (physical AI for unmanned systems & defence) raised $115M to build AI products for unmanned vehicles, autonomous operations & defence C2 systems, backed by investors like Tesi & Nokia (here)
Energy
UK’s nuclear task force says Britain is the world’s most expensive place to build nuclear plants due to excessive regulation, risk-aversion and sprawling bureaucracy (here)
Fuse Energy (renewable power), a UK start-up founded by former Revolut executives, is in advanced talks to raise new funding at a valuation of about $5B, which would mark a 5x increase from when it raised money 4 months ago (here)
X-energy (advanced modular nuclear-reactors), backed by Amazon, has closed a $700M funding round led by Jane Street, as it races to deliver three big contracts with Amazon, Dow Inc and Centrica to build a fleet of ~150 small modular reactors in the US & UK (here)
Robotics
Flexion (robotics autonomy software) raised $50M. As more capital floods into humanoids & logistics robots, the leverage may sit in the software layer that lets hardware generalise rather than in the humanoid chassis itself (here)
Space
SpaceX (launch & heavy lift) suffered another explosion as its first upgraded Starship V3 booster blew up during early pressure testing in Texas; beyond the memes, each failure pushes questions about whether Starship can meet 2026 orbital demo timelines & how tight the race with Blue Origin is for key NASA missions (here)
Crypto
More than $1T has been wiped off crypto markets over the past 6 weeks as concerns over AI valuations, higher-for-longer US rates & general risk-off sentiment hit speculative assets hardest (here)
A sharp rise in so-called “wrench attacks” targeting wealthy crypto holders - a homeowner was held at gunpoint by someone posing as a delivery driver in San Fran & was forced to handover credentials for $11M in crypto (here)
EVs
Waymo has gained California DMV approval to expand autonomous vehicle operations across much larger areas of the Bay Area & S.California. Some areas still require approval for carrying paying passengers, but Waymo plans to launch service in San Diego by mid-2026. The expansion follows rapid nationwide growth, including new markets, freeway-capable rides & preparations for commercial launches in multiple U.S. cities (here)