Venture Geopolitics Issue 22
18 November 2025
In a year when AI capex has become the macro engine, new jitters over rising inference costs, debt-fuelled buildouts, stretched ROI timelines and a crowded retail consensus triggered another brief wobble in markets.
Reports suggesting OpenAI’s cash burn on inference may exceed revenue and Peter Thiel’s fund unloading its entire Nvidia stake ahead of crucial earnings added to the tension.
Yet the core fundamentals - AI capex, employment, and broad S&P contribution - remain relatively steady. And some of the week’s most consequential AI developments were geopolitical: Anthropic disclosed a Chinese state-backed espionage campaign that automated 80–90% of its workflow with AI, while the US confirmed a foreign breach of the CBO and a White House memo accused Alibaba of facilitating military-linked data access for Beijing. AI power is rapidly converging with cyber power, turning the capital–compute–security nexus into a new geopolitical battleground.
IPOs/Publics
A backlogged SEC, the approaching holiday slowdown (Thanksgiving) & pressure on AI/tech stocks weigh on hopes for IPO rebound, despite US government reopening
Several tech stocks (Palantir, Tesla, Intel, Oracle, Nvidia) trading down 5-10% over past 5 days
Groww (Indian retail investing platform) raised $748M in its IPO at $9B. India’s largest fintech listing this year (here)
Pine Labs (Indian merchant payments platform) raised $440M in its India IPO, +14% on debut, a positive signal for late-stage Indian fintechs along with Groww (here)
Grayscale (crypto investment platform) submitted S-1, keeping public listing ambitions alive as regulators continue to shape the next wave of listed crypto vehicles (here)
SBI Shinsei Bank (Japanese lender) preparing Tokyo listing that could raise ~$2.6B, likely Japan’s largest IPO this year (here)
Einride (autonomous electric freight trucks) plans to go public through a SPAC merger at $1.8B (here)
Peter Thiel’s hedge fund has dumped its entire $100M Nvidia stake (here). Nb, Nvidia announces earnings this week (19th) which will be provide critical data around AI confidence
Google unveiled new AI shopping features (including a “buy it for me” tool that sets price alerts & auto-purchases when items fall below a user threshold) pushing the checkout journey deeper into Google’s own agent layer (here)
Google DeepMind unveiled a new AI weather model that delivers 2-week forecasts 50% faster & predicts hurricane paths a day earlier (here)
Meta announced it will begin grading employees on their “AI-driven impact” from 2026, making AI adoption & productivity gains an explicit part of performance reviews across the workforce (here)
Oracle shares down ~25% over past month as it leans into a late pivot from traditional business software to AI, committing to spend hundreds of bns over next few years on chips & data centres, largely tied to capacity deals with OpenAI (here)
Berkshire Hathaway has built a $4.3B stake in Alphabet, now its tenth-largest holding, while selling about $11B of Apple shares in the third quarter, modestly rebalancing from one US consumer-tech giant to another. Apple reportedly accelerating search for Tim Cook’s successor (here)
Big Dogs
Leaked documents obtained by Ed Zitron revealed detailed figures showing OpenAI’s cash inference costs may exceed revenue & clarified training compute mostly paid with Microsoft credits, while inference - the day-to-day cost of running models - is overwhelmingly cash, creating financial pressure (here)
OpenAI has begun piloting group chat in ChatGPT for users in Japan, New Zealand, South Korea & Taiwan, validating earlier reports & moving ChatGPT further into collaborative work use cases (here)
Thinking Machines Lab (AI models & tooling) reportedly in talks to raise at $50B, only 5mo after announcing its last round at $10B (here)
Anthropic (frontier AI models) plans to invest $50B in AI data centres in Texas & New York, in partnership with Fluidstack, signalling a major US infra push to secure long-term compute capacity for Claude (here)
Jeff Bezos has co-founded Project Prometheus, an AI company that has raised $6.2B to develop AI for engineering & manufacturing. Bezos will serve as CEO (here)
Anysphere (AI coding assistant Cursor) raised $2B at $29B (here)
Lovable (AI-assisted app builder) reportedly in talks to raise fresh funding at $6B, roughly 3x valuation from July, off back of “vibe coding” interface that lets users build applications with natural-language prompts (here)
Firmus Technologies (renewable-powered AI data centres) raised $327M from existing investors at $3.9B, more than 3x valuation from 2mo ago (here)
Ramp (corporate expense management) raised $300M at $32B (here)
D-Matrix (AI inference chips for data centres) raised $275M at $2B (here)
Zilch (BNPL & consumer credit services) raised $177M at a flat $2B allowing the London-based lender to keep expanding in regulated credit rather than pure-payments territory (here)
Flatpay (business payments) raised $170M at $1.7B - Europes 22nd Unicorn of the year (here)
Sakana AI (biologically inspired AI systems to make ML more efficient) raised $135M at $2.6B+ (here)
Ferroelectric Memory (low-power memory for AI compute) raised $116M (here)
Wonderful (AI agents for customer-facing enterprise work) raised $100M to build AI agents that handle voice, chat & email workflows (here)
Parallel Web Systems (AI-native web search infra), founded by former Twitter CEO Parag Agrawal, raised $100M at $740M to provide the crawling & search infrastructure used by AI agents (here)
Venture Capital
Accel Globalscape 2025 (here)
o AI infra winners breaking away: Nvidia added $1.6T in market cap, Oracle +151% - while legacy SaaS players like Adobe (–18%) & ServiceNow (–4%) declined
o Efficiency gap massive: AI-native companies run at 6–12x higher revenue / employee (Cursor $6.1M/FTE vs Salesforce ~$0.5M/FTE)
o Growth unprecedented: Lovable hit $100M ARR in 8 months; Cursor reached $500M ARR in ~2.5 years. Developer AI adoption now 90%
o Margins low but improving: AI apps sit at 7-40% gross margin today, but inference costs have dropped 97% since 2023
o AI infra boom: $4.1T in AI CapEx forecast (2026–2030), with 117GW new capacity needed & 55% already committed; major constraint is the 36GW US power shortfall
o While large part of funding has been driven by huge raises for OpenAI, Anthropic & xAI, application layer more level playing field US vs Europe & Israel. European & Israeli-founded cloud and AI application companies have attracted $30B of investment – 2/3 the funding of US peers & +36% YoY
o Agentic AI is working in production: UiPath/Fiserv at 98% automation, Decagon/Duolingo at 80% ticket deflection, Salesforce Agentforce at $440M ARR
o Public markets recovering: Cloud multiples are back above pre-COVID at 7.8x
o Funding is huge & AI-heavy: $184B in Cloud & AI (2025E), with 60% going to AI models. 80% growth over 2024
o Unicorns emerging faster than ever: 65% new unicorns are < 3 years old; time to unicorn now 18–36 months
Dave Kellogg (Balderton): we’ve entered an AI era of “haves and have-nots” where traditional SaaS metrics are losing influence. He notes investors underweight the durability of rapid AI-driven revenue: if a startup reaches $100M in 18mo, even with only 70% recurring, it’s still “70% of something” versus “100% of nothing.” As a result, many companies are de-emphasizing ARR & instead reporting revenue or product revenues. Broader caution is about timing: bubbles can outlast sceptics, making shorts perilous, but eventually a trigger will start the unwind & reveal that “this time” wasn’t different. For AI-native growth companies, his advice is Ellison-esque: win quickly, outgrow competitors, hire aggressively & accept a winner-takes-most market while capital remains plentiful (here)
Mikael Johnsson (Oxx) offers counterweight. We are still at the “Peak of Inflated Expectations”: soaring valuations, rapid company creation, modest real productivity gains. Adoption is driven by FOMO, experimental budgets & lightweight deployments, yielding fast sales cycles but shallow integration & high churn risk. Much AI revenue is circular - vendors selling to one another - while ARR quality is weak & gross margins are structurally squeezed by inference costs. In the end, fundamentals haven’t changed: durable winners will embed deeply in workflows, deliver clear ROI, & generate strong cash flows. The valuation stack may be new, but the underwriting questions aren’t (here)
→ Early AI rounds may not be about clean metrics, but about understanding the problem intensity, the value transfer, defensibility & distribution. Ultra-fast revenue growth mostly tells you a company has solved go-to-market, which is non-trivial. The hard part for investors is distinguishing between the companies where that growth sits on top of meaningful product-market fit & those where it sits on top of a temporary enthusiasm for “AI budget”
PE firm Vista has told investors it plans to shrink its workforce significantly by using AI tools to automate operational roles, some junior analyst work & parts of investor relations, or to grow without adding headcount. Staff numbers could ultimately fall by as much as 1/3 through redundancies & reduced hiring. Vista has been early in pushing AI across its portfolio (scoring companies on AI threats & opportunities) (here)
VC megafunds became so large that they shifted from discovering markets to constructing them. Tools include narrative (“It’s Time to Build”), networks (Microsoft–OpenAI, Palantir–Pentagon), regulatory lobbying (a16z in crypto) & industrial alignment (SpaceX–NASA, Tesla–subsidies). Uber & Airbnb showed that you can rewrite urban rules by scaling faster than regulators; Anduril, Palantir & OpenAI show you can fuse with national strategy itself. O’Sullivan’s summary: venture capital has evolved from funding innovation to engineering exits by constructing outcomes in partnership with the state, creating a venture–industrial complex where private capital & public power have effectively merged. Western economic development starts to look much closer to Eastern state-led models, with Big Venture as an instrument of geopolitics (here)
YC’s latest “Requests for Startups” highlighting 6 AI frontiers it wants founders to attack: rapid workforce training at scale, video generation as a core primitive, ultra-lean billion-dollar teams, infrastructure for multi-agent systems, AI-native enterprise software & LLMs replacing government consultancy work (here)
Venture Geopolitics
EU plans to delay key high-risk AI rules in the AI Act by at least a year, likely pushing enforcement and related fines to 2027. Many European startups welcome the pause as a chance to create clearer, more workable rules, though some fear it signals EU capitulation to U.S. influence. The proposal will be part of a digital simplification package expected Nov 19 (here)
Nigel Morris (Capital One) emphasizes 25% of Buy Now Pay Later (BNPL) users are using the service for things like groceries - and we should be worried given these loans aren’t reported to credit bureaus and lenders can’t see when someone has taken out different loans across different platforms (here)
Arattai (Indian messaging app by Zoho) has become a test case for digital industrial policy. Top ministers publicly promoted the app as part of a broader “made in India” push amid US tariffs that threaten India’s economy. Endorsements helped daily sign-ups jump from about 3k to 350k & delivered roughly 12M new downloads in a single month (here)
Taiwan’s chip-powered boom has created both exceptional wealth & severe economic strain. Global demand for semiconductors has pushed Taiwan’s trade surplus to 31% of GDP / current-account surplus to 16% of GDP. Normally this would strengthen the currency, but the central bank has long suppressed it, viewing weak currency & $600B in reserves as strategic protection. This model now generates risks: life insurers face a $200B currency mismatch, wages stagnate & U.S. negotiators bristle at perceived manipulation. As chip demand keeps rising, upward currency pressure intensifies, making Taiwan’s export-security strategy increasingly unsustainable (here)
Brain–computer interfaces (Neuralink, Meta & others) are moving quickly enough that mainstream outlets are now framing them in rights language. A NYT piece argues fast-moving BCI technologies are beginning to read simple thoughts, influence feelings & connect human minds directly to AI systems, raising urgent questions about mental privacy & control (here)
Europe faces critical strategic risk from growing dependence on China for key industrial metals (gallium, germanium & antimony) essential for chips, defence systems, renewables & AI. China controls most global refining capacity; Europe hasn’t built a new smelter since the 1990s & has shut a third of its base-metal plants. Rebuilding sovereignty will require €75–150bn in modernisation & policy tools such as price floors, strategic offtakes & alliances like the US–Australia minerals pact. Without this, Europe risks ceding technological competitiveness & geopolitical leverage to others (here)
Strategic Sectors
AI
“Every enterprise software firm is selling the same AI agents”. At least 7 major platforms now compete head-on with agents that forecast sales, triage support tickets, write customer emails, assist with recruiting, finance etc. Older database & streaming firms like Snowflake & Confluent are competing directly with younger AI app firms such as Sierra & Decagon. OpenAI & Anthropic are layering agent features inside ChatGPT & Claude. Its an “all or nothing” – competitive barriers have fallen, software firms can chase new budgets across functions they didn’t have a right to win previously. Execs feel valuations will go to $1T or zero (here)
Cybersecurity
Anthropic disclosed first major publicly documented AI-led cyber-espionage campaign. A Chinese state-backed group used Claude to automate 80–90% of an operation against roughly 30 global targets, including large tech firms, financial institutions & government agencies. In geopolitical terms, “AI superiority” will increasingly map straight to cyber power (here)
Google Cloud released its Cybersecurity Forecast 2026 predicting a sharp rise in prompt-injection attacks, exploitation of AI agents & cyber-physical attacks targeting European infrastructure (here)
Scope of UK Cyber Security & Resilience Bill widening to cover additional sectors including data centres & IT services. Firms could be fined up to 4% of global turnover / £17M if they fail to meet rules. Cyber has effectively joined capital & liquidity as a regulated systemic risk (here)
US Congressional Budget Office confirmed suspected foreign cyberattack. Some offices have paused communications with the agency while investigations continue. Underlines that the analytical back office of government is now firmly in the crosshairs (here)
OneTrust (privacy & compliance platform) reportedly exploring a private equity sale (here)
Defence
Secretary of War Pete Hegseth told a room of defence-tech founders & primes that US defence procurement must change & that his Pentagon intends to drive that shift. The emphasis is on faster, more flexible buying processes that can accommodate software-first, iterative products (here)
Alibaba (Chinese tech conglomerate) accused in a White House memo of supporting China’s military by providing access to data such as IP addresses, WiFi details & payment records. US agencies link this to wider campaigns including Salt Typhoon breach. Lawmakers have called for measures including delisting 25 Chinese firms. Alibaba rejects allegations as baseless but shows how commercial platforms are being treated as extensions of national cyber power vs neutral commercial actors (here)
Crypto / Stablecoins
Bitcoin broke 100k to the downside for the third time this month (here) while MarketVector Digital Assets 100 Small-Cap Index (which tracks the 50 smallest digital assets in a basket of 100), fell to its lowest level since Nov 2020 (pre Covid) (here)
Bitcoin miners seeing renewed demand as AI companies turn to them for data centre space & power. With AI workloads constrained by both GPUs & grid capacity, miners’ existing facilities & energy contracts have become valuable - driving up crypto-miner stock prices & allowing them to raise fresh billions. A cohort recently viewed as stranded asset owners repositioned for strategic national AI capacity (here)
The rapid rise of US-linked stablecoins could threaten Europe’s financial stability. If a rush to exit these tokens forced large sales of the US assets backing them, it could disrupt markets, the economy & inflation — potentially pushing the ECB to adjust interest rates (here)
>$28B in illicit funds from hackers, scammers & criminal groups flowed into major crypto exchanges such as Binance since 2023 (here)
Robotics
Neura Robotics (cognitive & humanoid robots) reportedly in talks to raise €1B led by Tether (the crypto group) valuing the startup at €8-10B (here)
Xpeng (Chinese EV maker) showcased the latest version of its humanoid robot, initially targeting roles such as office receptionists & sales assistants, blurring the line between automotive engineering & general-purpose robotics (here)
A much-trailed Russian humanoid robot reportedly staggered & collapsed during its stage debut to the theme from Rocky in front of about 50 journalists (here)
Apptronik (humanoid robots for industrial use) is raising at least $400M at ~$5B to scale Apollo, its two-legged, two-armed humanoid robot. The company, backed by Google, plans to integrate DeepMind’s AI to enable more autonomous operation in factories & warehouses (here)
EVs
Waymo (autonomous driving platform) expanding to include freeway routes across Los Angeles, San Francisco & Phoenix, extending from urban cores onto faster, higher-risk roads. A critical test of whether regulators, riders & business models are ready for autonomy at scale across entire metro networks (here)
Space
Seth Dobrin sketches a space-tech thesis where space infrastructure becomes a direct response to AI’s terrestrial energy constraints. He estimates in-orbit data centres could grow from about $1.78B in 2029 to >$39B by 2035, as part of a broader $1.8T space economy. The core idea is that as AI drives an Earth-based power crunch, continuous solar power in orbit plus the physical isolation of space-based compute become attractive for both economics & security (here)
Blue Origin (launch & space infrastructure) successfully launched its New Glenn rocket with a pair of NASA spacecraft bound for Mars, only the second flight for the heavy-lift vehicle that NASA & Jeff Bezos’s company hope will become a workhorse for ferrying people to the Moon (here)