Venture Geopolitics Issue 20
04 November 2025
Nvidia shattered another ceiling this week, becoming the world’s first $5 trillion company – now worth 9% of the entire S&P 500. One firm, larger than all but five of the world’s stock markets. Meanwhile, OpenAI finalised its shift to a for-profit model, struck a $38 billion alliance with AWS, and teased markets with a possible $1 trillion IPO in 2026. The concentration created by the AI platform shift is both breathtaking and brittle: an Azure failure just days after the AWS outage; OpenAI’s new Atlas browser exploited just days after launch; while ~95,000 viewers unknowingly tuned in to a deepfake of Nvidia’s CEO promoting a crypto scam on YouTube.
The more intelligence, infrastructure and innovation concentrate within a few technology stacks and behind a few individuals, the more essential it becomes to build networks of resilience, not fragility. The next decade of value will come from how we invest in infrastructure and design systems that scale securely.
IPOs / Public
Navan (corporate travel & expenses) fell 20% (to $4.6B) in its debut - a steep drop from private-market price of $9.2B in 2022 (here)
PonyAI (autonomous driving) launched Hong Kong IPO that could value it at $10B+ (here)
Discord (chat & community platform) filed to go public after reaching $890M in revenue (and rejecting $12B takeover offer from Microsoft) (here)
Xandu (quantum computing) to go public via SPAC, EV of $3.1B (here)
OpenAI reportedly preparing for 2026–27 IPO that could value it at $1T (here)
Nvidia became the world’s first $5T company (here)
Accounts for 9% of S&P500 by weight & larger than all but 5 of world’s stock markets (US, China, Japan, Hong Kong, India)
Forecasts $500B revenue over next 5 quarters, selling ~20M of latest GPUs (5x previous generation)
Earnings season, including Meta, Alphabet, Microsoft, Apple & Amazon (here)
Strong results. Even when one wobbles (Meta) others pick up the slack
AI-linked stocks account for ≈75% of S&P500 returns
3 companies (Apple, Microsoft, Nvidia) now exceed $4T
Common thread remains relentless AI capex: a ~$60B raise to market expectations for CY26
Microsoft Azure suffered outages after a configuration error – another reminder of big-cloud fragility and infrastructure concentration.
Coinciding with Dutch elections underscored how vulnerable public infrastructure has become to commercial infrastructure (here)
Meta created a $27B SPV for the 5GW ‘Hyperion’ data centre build out
Meta 20%, Blue Owl 80% (with $2.5B equity injection)
Project financing approach means Meta’s balance sheet & credit rating unscathed (Meta broke up leases into 4 year chunks so not even considered debt)
Matt Levine deciphered the financial wizardry – noting how big tech does not want to issue debt, in part for real risk management reasons but also to protect credit ratings (Bloomberg)
Reddit created a ‘test post’ that could only be crawled by Google’s search engine…within hours Perplexity had surfaced the content (Business Insider)
Elon Musk’s launched Grokipedia - built entirely using AI generated content and exists to challenge “woke” Wikipedia (here)
Big Dogs
OpenAI finalised shift to for-profit public benefit corporation (here)
Microsoft holds 27% stake worth ~$135B & will receive 20% of revenue until 2032 (Bloomberg)
Ambitions are colossal: ~30GW of compute commitments (≈$1.4T in infrastructure). Plans to add 1GW/week, comparable to national infrastructure projects. In fact, commitments are nearly double the value of US Interstate Highway System (Azeem Azhar)
Musk challenging the restructuring ahead of March trial; Zvi Mowshowitz titled newsletter “OpenAI Moves To Complete Potentially The Largest Theft In Human History” (here)
OpenAI signed 7-year $38B deal with AWS, taking recent commitments close to $1.5T
Will allow the company to make immediate use of AWS infra (incl Nvidia chips)
Ties OpenAI closely to Amazon (which committed to invest $8B in Anthropic)
Lessens OpenAI’s dependence on Microsoft, biggest backer (FT)
Bending Spoons (app conglomerate that buys & revamps ageing digital brands) announced acquisition of AOL for $1.4B. Also announced $710M raise at $11B (here and here)
Model echoes digital-first IAC (rolling up legacy internet assets & squeezing value from loyal, low-churn users)
HackerNews chatter: “if it goes down, check back on Monday”. Bending Spoons buys dying platforms & hike prices. If users haven’t already quit, they probably never will (here)
Lots of other Big Dogs getting Big Cash:
FireworksAI (chip access & model inference optimisation) raised $250M at $4B+ (Lightspeed, Index, Evantic, Sequoia)
Synthesia (AI-generated video) raised $200M at $4B led by GV (nearly 2x Jan valuation. Follows rejected Adobe offer mentioned last week)
Harvey (AI legal & finance tools) raised $150M at $8B+ led by a16z
HippocraticAI (safety-focused large language model for healthcare) raises $126M at $3.5B
EnduroSat (modular satellites & constellation systems) raised $104M (GV & Lux)
Substrate (particle-accelerator-based semiconductor manufacturing) raised $100M+ at $1B+ (Founders Fund, General Catalyst, Valor, Allen & Co)
Cartesia (state-space model voice AI) raised $100M (Index, Kleiner Perkins & Lightspeed)
Applied Compute (enterprise-specific AI agents) raised $80M (Benchmark, Sequoia, Lux, Elad Gil) after emerging from stealth. Early adopters include Cognition, DoorDash, Mercor.
Humans& (AI collaboration models, founded this year by ex–xAI researcher Eric Zelikman) reportedly raising up to $1B at $5B
Poolside (AI code assistant) in talks with Nvidia for $500M–$1B investment as part of $2B round at $12B
Legora (collaborative AI for legal) reportedly raising $150M at $1.7B
Venture Capital
Sequoia launched new funds ($200M seed & $750M venture growth) (here)
Meanwhile 83North announced they won’t raise another fund (~$400M remains to deploy from its 11th & final vehicle). A rare rejection of perpetual AUM treadmill (here)
With others (e.g. Lakestar) winding down, is a quiet pattern emerging? Are European firms choosing discipline over bloat?
Rory O’Driscoll (ScaleVP): “it may be apocryphal but the Red Army quote applies to A16Z at $10B: “Quantity has a quality all its own…its extra nice when your strategy means you can flood the zone & everyone else has to work twice as hard”
Jason Lemkin, SaaStr: “Who Will Buy the SaaS Companies?” (here)
We’re in an odd market: record investment capital ($476B in PE dry powder plus huge VC funds) but nearly all of it is chasing AI. So, strong (non-AI) SaaS with $20–100M ARR, 25–40% growth, solid margins & good retention face limited exit paths
Historically, could be bought by PE (5–8x revs), rolled into strategics, raise growth equity or grow to IPO. Today, paths have tightened:
PE prefers bigger or faster-growing deals
Strategics want AI or scale
Growth equity is flowing to AI
Public markets don’t reward $200M ARR growing 25–30% - judged against giants like CrowdStrike (growing 29% with 32% FCF margins & $3.6B ARR)
IPO bar now $400–500M+ ARR with 30–50%+ growth & profitability
More on bubbles, this week from Gavin Baker (Atreides Management) (here)
At peak of tech bubble, 97% of fiber laid in America was dark
Contrast with today: no dark GPUs. Biggest problem today is GPUs melting from overuse
If look at Return on Invested Capital of biggest public spenders on GPUs, they have seen around a 10 point increase in their ROICs
Venture Geopolitics
Trump met Xi Jinping in Seoul & called it “a 12 out of 10” meeting (here)
China will pause curbs on rare-earth magnets while US will reverse certain restrictions on Chinese firms & cut fentanyl-related tariffs
Beijing will resume large-scale purchases of soybeans & other US agricultural exports
Halts escalation, not rivalry – tensions over Taiwan & tech remain
Commentary split: tactical pre-midterm move vs evidence of Beijing’s ability to escalate into negotiation & extract meaningful concessions?
US allows Microsoft to ship latest Nvidia chips to UAE for first time (FT)
Middle East is crucial battleground in Washington vs Beijing
First company to receive such licence under Trump administration
Microsoft plans to increase UAE investment to >$7.9bn 2026-2029
Region will play a vital role in disseminating AI through the “global south”, running from the Middle East & southern Europe to Africa and east Asia
Pitchbook on Sovereign AI “This market…is bifurcating into two distinct, competing ecosystems led by the US & China, compelling other nations to make strategic alignment choices that will dictate capital flows & technological standards for next decade” (here)
4 critical layers for investment highlighted
Semiconductors, where US export controls are creating a protected market for domestic Chinese players & a premium for Western-aligned suppliers
Infrastructure, where sovereign wealth funds in Middle East & state-backed consortia are underwriting hyperscale datacenter buildouts
Foundation models, where capital is concentrating in a few US-based leaders who are becoming quasi-sovereign entities through deep partnerships with allied nations
Applications, nascent but high-potential area where sovereign-specific use cases in defence, healthcare, & public services will emerge
Equinix to invest £3.9B in UK data-centre campus – one of Europe’s largest, creating 2,500 jobs & doubling its British compute capacity (here)
Coppers record highs. Less from real demand or supply strain than policy distortions (The Economist)
EVs & AI data centers lift demand, but efficiency gains & China’s construction slump shrink it
Supply hiccups (Grasberg mudslide, Chile/DRC) are manageable
→ Trump’s metal tariffs, a weaker dollar, lower U.S. rates & speculative “tourist” buying driving prices
i.e. if global growth falters / tariffs escalate the rally could unwind fast
Israel introduced tax breaks to lure back tech workers after emigration rose 45% in 9mo following start of war (FT)
Measures include foreign income exemptions, tax credits & clearer equity rules, alongside broader incentives for investors & M&A oversight
Note, tech firms from Europe/UK are increasingly basing HQs in the U.S. for deeper capital markets and faster scaling, not only to escape war
Strategic Sectors
AI
Cursor launched own proprietary LLM (Composer) designed specifically for its IDE (here)
Anthropic tested whether Claude could notice changes in its own thinking. In experiments where hidden patterns were added to its reasoning, Claude sometimes detected injected neural patterns and noticed something was off - a hint that introspection may scale with capability (here)
Mustafa Suleyman (Microsoft’s AI chief) argued only biological beings can be conscious and urged the industry to stop pursuing seemingly sentient AI, underscoring a growing ethical divide as tech races toward more human-like systems (here)
NHS pilot of Microsoft Copilot (30k staff across 90 orgs) found AI saved ~43 mins per worker daily i.e. ~400k staff hours monthly, or hundreds of millions annually. Productivity rose 2.7% YoY, beating government’s 2% target (here)
The Economist argues AI erodes information asymmetry - empowering consumers in opaque markets (healthcare, mortgages, services) that account for a quarter of US spending (here)
AI adoption & labour markets
Microsofts AI diffusion report: AI use concentrated in service economies (UAE 59.4%, Singapore 58.6% of working population) (here)
Remote Labour Index (“RLI”) found top models automate only 2.5% of “economically valuable remote-work projects” (here)
Wharton study: 75% of large US firms already seeing ROI; 10% spend >$20M/yr on gen-AI (here)
Goldman Sachs survey found 37% of US corporates using AI in production, 74% expect adoption within 3 years (Goldman Sachs Global Investment Research)
Top use cases: analytics (65%), customer service (56%), sales (48%), IT (46%), document work (42%)
Only 11% cutting staff (31% in TMT), mostly via hiring freezes
Finance may see -14% headcount in 3 years
GS still forecasts a +15% US productivity uplift over the next decade from AI diffusion
Defence
Palantir sued 2 ex-engineers for allegedly stealing trade secrets & passing them to Percepta, a rival backed by General Catalyst (here)
The firm claims a “months-long charade” before the pair’s exit, only discovering the leak when Percepta’s launch went public
Interesting deal: Valthos raised $30M to develop next-gen biodefence technologies - part of a growing wave of dual-use biotech startups bridging health security & national resilience
Cybersecurity
OpenAI unveiled Aardvark, a GPT-5-powered cybersecurity assistant originally built for internal use (here)
Helps teams identify & patch vulnerabilities - part of OpenAI’s quiet push into enterprise tools
Meanwhile, new Atlas browser suffered prompt-injection exploits within days of launch. Researchers showed malicious web content could hijack agent behaviour, exposing the core security flaw of agentic AI: once models act with user credentials, the same-origin policy breaks down (here)
The takeaway: agentic browsing equals authenticated privilege escalation - requires architectural, not cosmetic, fixes
MCP servers are being rapidly adopted without proper safeguards. Vast majority lack authentication or authorisation entirely - a “crazy” oversight in 2025. SMBs reusing public MCP servers face the greatest risk, while shadow AI adoption worsens the threat (Channelholic)
Trellix’s CyberThreat Report confirms AI-powered malware is now operational (here)
APT28’s (the Russian hacking group) LameHug (their new malware) became the first publicly reported AI-driven infostealer, while Qilin (Ransomware-as-a-Service) topped global activity
Industrial targets made up 37% of all attacks, with US hardest hit
Report highlights AI–nation-state convergence: AI-enhanced tools are now in active deployment by state & criminal actors alike
Notable deals: Sublime Security (email threat detection) raised $150M (Georgian, Index, IVP & Citi); Reflectiz (third party software supply chain risk) secured $22M (Fulcrum); ConductorOne (identity security for humans & AI accounts) raised $79M (Greycroft, CrowdStrike, Accel, Felicis)
Crypto
JPMorgan tokenised a PE fund on its own blockchain, extending its Onyx platform beyond payments & repo markets into fund infrastructure. The move signals Wall Street’s deeper shift toward tokenised finance - using blockchain rails to automate settlement & investor administration (WSJ)
Around 95k viewers unknowingly tuned in to a deepfake of Nvidia CEO Jensen Huang promoting a crypto scam on YouTube. Fake livestream pushed a “giveaway” token scheme, demonstrating how generative AI & celebrity likeness are converging to industrialise fraud (here)
Tether posted $10B+ profit YTD, minting $17B new USDT in 2025 & widening its lead in stablecoins despite rising competition from Circle & PayPal. Its cash-printing model (investing reserve assets in Treasuries) continues to make it one of the most profitable entities in finance, even as transparency debates linger (here)
Core Scientific (crypto mining) rejected a $9B takeover from CoreWeave, which also started as a crypto miner. CoreWeave now serves AI workloads, and the hope is Core Scientific can transition into a CoreWeave, too. Core Scientific’s stock jumped on the news.
The story underscores miners’ surging valuations amid the AI–energy convergence – and for some is another sign of a bubble.
Meanwhile, CoreWeave agreed to acquire Marimo (open-source Python workspace) absorbing software talent to feed AI & crypto infrastructure growth (TechCrunch)
EVs
Uber plans to launch its first autonomous taxis in the Bay Area in 2026 using Lucid’s Gravity SUV equipped with Nuro’s self-driving technology (here)
Waymo is targeting 3 new cities: Detroit, Las Vegas & San Diego. Waymo executes >250k trips each week & targets 1m by end of 2026 (here)
Baidu’s Apollo Go reached 250k weekly self-driving rides (here)
Robotics
Neo, 1X’s $20k humanoid household robot, has entered limited trials. It is currently teleoperated, i.e. requires owners to sign a waiver, allowing human operators to see through the robot’s eyes and help it perform tasks around your home (here)
Elon Musk used Tesla’s earnings call to plead with investors to approve his $1T pay package, arguing he needs “enough voting control to build a robot army - unless I go insane.” As Matt Levine dryly noted, such comments are both absurd & historically unprecedented: no previous billionaire has ever asked for absolute command of humanoid robots as a compensation plan… (here)