Venture geopolitics hardened last week. Nvidia and AMD agreed to hand 15% of their China chip revenues to the US government in exchange for export licences – a striking example of Washington turning corporate sales into state leverage. Meanwhile, Trump’s trade policy is rewriting European venture dynamics: tariffs on US imports from the EU will surge from 1.2% to 17.5%, inflating costs for semis, batteries and sensors. The effect is to deepen the transatlantic divergence – US firms doubling down on AI and defence while European startups face higher costs, reduced capital access and potential relocation pressures. These shifts matter: the global innovation ecosystem looks increasingly fractured, with the winners concentrated in a handful of markets.
IPOs / Public
Firefly Aerospace (space and defence tech) made a strong Nasdaq debut at $45/share (target $41–43), reaching a $9.8B valuation. Shares peaked at $71.
Heartflow (medical imaging and diagnostics) jumped 51% on debut, showing Wall Street’s renewed appetite for IPOs.
Bullish (crypto exchange, backed by Thiel and parent of CoinDesk) filed for a $600M US IPO at a $4.2B valuation.
Palantir crossed >$1B quarterly revenue for first time = +48% YoY, with nearly every headline metric accelerating vs Q1. Growth trajectory: 13% (Q2 2023) → 27% (Q2 2024) → 48% (Q2 2025).
Tesla approved 96M shares for Musk as part of an “interim” retention plan after he threatened to quit.
Uber announced a $20B buyback and will trial robotaxis in London with Wayve.
Nvidia now makes up nearly 8% of the S&P 500 — the highest weighting in index history.
Bankers and investors, including Pimco, warned valuations may be tipping into bubble territory, with complacency over trade and political risk.
Big Dogs
OpenAI
Released GPT-5: a unified model with an internal router, four personalities (Cynic, Robot, Listener, Nerd), safer partial answers to sensitive prompts, and stronger coding/app-building. Widely seen as incremental, not transformational. Ben Evans critiqued how Anthropic and OpenAI are bigging up their models: “these are use cases where ‘80% correct’ is meaningless – there is only right or not right. It’s as though Apple launched the Apple 2 and claimed that it had the same uptime and throughput as a mainframe.”
Released open-weights models (gpt-oss-20B & 120B) for local use. Weights and parameters are public, though not training code/data. Altman: “Far more good than bad will come from it.”
Preparing an employee stock sale at a $500B valuation (up from $300B earlier this year).
Apple will integrate GPT-5 across iOS 26, iPadOS 26 and macOS Tahoe 26.
Anthropic released Opus 4.1 (better coding accuracy and agentic capabilities) and is eyeing a $150B raise. Claude is moving vertical into FS. Anthropic also warned the US will need +50GW more power by 2028 to fuel AI.
Fireworks (AI inference infra) in talks for a $4B valuation with Lightspeed and Index.
Decart (creative AI for real-time video style transformation) raised $100M Series B at a $3.1B valuation from Sequoia, Benchmark, Zeev and Aleph.
VAST Data (AI/HPC infra) is raising with CapitalG and Nvidia at ~$30B valuation.
Clay (personalised outreach platform) raised $100M Series C led by CapitalG at $3.1B.
n8n (workflow automation) rumoured to raise at $2.3B valuation led by Accel — 6× uplift in 4 months.
GoCardless is courting buyers, with Dutch fintech Mollie the frontrunner.
ElevenLabs launched Eleven Music — generating full songs with vocals and instruments from text prompts.
Venture capital Insights
NYT profiled the rise of “twentysomething AI CEOs”: Scott Wu (Cognition), Michael Truell (Cursor), Roy Lee (Cluely) and Alexandr Wang (Scale AI).
Should VCs back jockeys or horses? Data suggests investors consistently over-index on founder traits and underweight markets (Crunchbase)
At my own fund we had a (long) debate this week, as we do most weeks, about where value will accrue in AI software. Strong arguments in favour of value accruing to small, problem-focused, industry-specific models. “If it doesn’t fit on your computer, it’s a waste of time”.
Euclid on vertical AI moats: “speed gets you started but is it a durable moat?” History shows speed matters less than defensibility — Lotus Notes, Siebel, MySpace, Hopin, Lensa etc as poignant examples (read here).
Venture Geopolitics
Nvidia and AMD to pay 15% of China chip sales revenue to the US government in exchange for export licences.
“Trump’s trade policy is redefining European VC” — tariffs on US imports from the EU rising to 17.5% on average, hiking costs for semis, batteries, sensors and squeezing European startups’ margins and capital access. Relocation pressures building.
Abu Dhabi’s MGX is raising up to $25B in third-party capital to scale its AI strategy. “The era of government-backed deep tech is here”.
China’s exports rose +7.2% YoY in July (driven by SE Asia), but exports to the US fell >20%.
Truth Social launched AI-powered search (via Perplexity Sonar). Sources skew to Fox/Epoch Times/Washington Times, avoiding Reddit, NPR and YouTube.
Apple pledged another $100B in US investment (on top of $500B earlier this year) but faces tariff headwinds on Indian exports.
Anthropic and OpenAI approved to sell AI products to the US government - licences offered for just $1 though - to win market share (and favour).
The Economist notes how America’s tech giants are levering up like never before - MSFT finance leases tripled to $46B, Meta seeking $30B private credit, Alphabet issuing bonds for first time since 2020. Challengers like CoreWeave and xAI are pledging Nvidia chips as collateral. Structural change: lenders are taking direct technology risk.
Goldman Sachs on AI labour markets: youth tech unemployment +3pp this year; expect productivity up 15% but unemployment +0.5pp above trend during reallocation. ~7% of workers likely to change jobs. Risks front-loaded in a recession scenario. Long-term: comparative human advantage remains in physical–cognitive tasks; 50% chance of AI surpassing all human work between 2047–2081.
Strategic Sectors
Defence
· Article in NY Times explores how Giants like Google, Meta, OpenAI, Palantir are now openly building weapons, advising the Army, and dropping “AI-for-peace” pledges. Defence is becoming tech’s newest growth engine, with a16z and others piling in. How Tech Firms Like Google and Meta Are Embracing the Military - The New York Times
UK’s MoD turned to Castlepoint Systems (Australian AI cyber firm) to manage sensitive data after a major breach — highlighting AUKUS partnership depth.