Venture Geopolitics Issue 58
28 July 2026
“This week we invented the ploughshare. It will eradicate all wants and needs. The modern farmer, equipped with donkey and ploughshare, will be 100x more efficient than the farmers of yore. There will be a superabundance of food, and we will enter a post-scarcity world. All war will end and people will want for nothing beause everyone will be well fed.”
Liang Wenfeng decided to suspend DeepSeek’s $1.5bn funding round this week after his “Daoist interview” with investors was leaked. He espoused the view that AI is a step change in the human condition and his company will therefore work to be a part of the system - not driving the change, not seeking monopoly, merely playing its part and riding the wave “with profound goodwill toward the world”. It was a powerful and self-aware message from a clear-eyed and considered leader. It was also one that destabilised potential investment.
Meanwhile, in an interview with The Economist, Elon Musk offered the Dadaist equivalent. In his words, we’re just chimpanzees, eating bananas and waiting to become AI’s labradors. The heat death of the universe means “it’s really all about the journey”; money will have no purpose once AI takes over; and the Treasury should simply “issue cheques” so humans can live in abundance. There’s a glimmer of a thought embedded in there somewhere, but it’s lost within an incoherent and ill-considered argument. “The robots and AI are providing more goods and services than any human could possibly consume. What do you need money for in that case?” he says, ignoring all lessons from history.
Musk raised the largest IPO ever because markets decided his judgement and agency are exceptional. Yet he is now arguing that humanity, following his lead, will hand agency to AI within the decade.
Good VCs back founders, not just companies. They do this because great founders have a vision of what the future looks like, and a plan for how to get there. And that changes the course of society. Charisma and celebrity are useful characteristics to make that plan happen, but the vision of the future has to be one we actually want to bring into being, and the plan for how to get there has to be coherent and believable.
IPOs / Publics
CXMT (DRAM memory) raised nearly $10B & soared 466% on Shanghai debut, reflecting investor appetite for domestic chip champions amid US export controls & AI-driven memory demand (here)
Anthropic is considering requiring rank-and-file employees to sell shares via preset 10b5-1 trading plans post-IPO, an unusual step typically reserved for executives, to avoid insider trading issues while preserving its culture of internal information sharing (here)
Nvidia has announced over $750B in fresh AI deals, including a $500B partnership with SK Group & potential $250B backing for OpenAI’s Ohio data centre lease, reigniting concerns about circular financing where the chipmaker funds its own customers (here, here, here)
Alphabet, Microsoft, Amazon, Meta & Oracle hold $1.65T in off-balance-sheet debt tied to AI infrastructure - up 8x in 4 years & more than their $1.35T reported debt - using separate legal entities for chip, server & data centre financing. The exposure mirrors Enron-era structures (now legal but opaque), with the catch that facilities roll onto balance sheets once live - posing risk if AI demand undershoots the $3T+ buildout through 2028 (here)
Big Dogs
AMD will sell Anthropic tens of billions in AI servers (up to 2GW of MI450 chips from 2027) & invest up to $5B tied to deployment milestones, the latest circular deal where chipmakers back their largest customers to secure share against Nvidia’s dominance (here)
OpenAI CEO Sam Altman plans to brief the Trump administration & Congress next week on the next generation of models, particularly their impact on work, as the US develops a safety review framework expected in weeks (here)
Meta, Microsoft, Nvidia & a16z signed a letter defending open-weight AI models as the Trump administration weighs restrictions on Chinese labs following Moonshot’s Kimi K3 launch; Google, Amazon, OpenAI & Anthropic did not sign, underscoring splits over how open models affect both competition & national security (here)
DeepSeek suspended its second fundraising round (targeting ≥$1.4B at ~$68B pre-money) days after founder Liang Wenfeng’s comments on US-China AI competition went viral (here)
Meta secured a $50B Louisiana data center using secret NDAs, a rewritten tax bill passed in months, & a structure that offloaded financial risk to Entergy & Blue Owl (which sold $27B in bonds). The governor codified the secrecy playbook as ‘Louisiana Lightning Speed’ to attract more projects, even as local residents faced evictions from rising rents & regulators approved power plants without independent review (here)
Venture Capital / Finance
Goldman Sachs launched an alternative investments platform for wealthy clients & family offices seeking direct stakes in late-stage private companies (SpaceX, Stripe, Canva historically), combining existing alternatives business with new teams for direct investments & secondaries (here)
British Business Bank is developing a £100M aerospace supply chain debt fund with UK Government, Airbus, GKN Aerospace, Rolls-Royce & ADS to help UK suppliers scale, invest & compete for future aircraft programmes (here)
Khosla Ventures is raising up to $5.5B across new funds, its largest ever, with $3B earmarked for seed & early-stage AI startups - a 40% increase on its prior $4B vintage 17 mo earlier (here)
Regulation
Anthropic settled with authors whose books were used to train Claude for $1.5B, paying ~$3,000 per book to over 482,000 claimants (91% claimed). The judge ruled that training on copyrighted books is fair use but acquiring them via pirate sites was wrongful (here)
European Commission fined Google €890M ($1B) for violating the Digital Markets Act by favouring its own services in search results & restricting app developers on Google Play. Google has 60 days to comply or face penalties up to 5% of global revenue. The decision risks escalating US-EU trade tensions as Trump weighs retaliatory tariffs (here)
France passed a law restricting social media access for under-15s, the first EU country to do so, with the ban potentially live by 1 Sept (here)
Venture Geopolitics
Andy Burnham scrapped the Department for Science, Innovation & Technology (created 3.5 yr ago) in his first week as UK PM, merging it into an expanded business department whilst elevating AI to Cabinet level with a dedicated minister (Kanishka Narayan) & new AI Taskforce in the Cabinet Office (here)
OpenAI doubled federal lobbying spend to $2.22M in H1 2026, whilst Anthropic nearly tripled it to $3.53M, as AI firms push to deter federal regulation, shape data centre policy & influence export control decisions like the US block on Anthropic’s Fable model (here, here)
Elon Musk gave a combative interview to The Economist, claiming AI will eliminate the need for money by 2036 as robots deliver ‘quasi-infinite’ abundance & the Treasury issues universal cheques (here)
Washington collected over $13B from Venezuelan oil sales in 6 mo since January but has given conflicting accounts of fund allocation - Trump claimed the US was “making a lot of money”, an executive order described control as “custodial”, whilst State said “billions” were disbursed. Venezuela’s Q1 GDP growth of 2.5% was the lowest in 5 years despite sanctions relief, suggesting only partial transfer (here)
Trump administration plans to redirect $200B/yr in federal R&D spending toward individual scientists & AI applications rather than universities, positioning the shift as a competition response to China & aiming to accelerate technological discovery through decentralised funding & AI-enabled research (here)
Moonshot AI trained Kimi K3 (2.8T parameters, world’s largest open-source model) on Nvidia Blackwell chips partly in China - despite U.S. export bans - by sourcing from multiple local cloud providers, & is now seeking more Blackwell capacity for Kimi K4. Alibaba also trained its 2.4T-parameter Qwen3.8-Max on Blackwell hardware. Chinese firms still depend on Nvidia for frontier model training, even as Beijing pushes domestic chip adoption for inference, highlighting both the limits of export controls & the continued technology gap in advanced AI compute (here)
Strategic Sectors
AI
Microsoft CEO Satya Nadella warned that enterprises relying entirely on one AI provider risk outsourcing their thinking & may not survive, urging firms to retain their own data, use AI gateways, & manage multiple models rather than depend on labs’ proprietary coding tools - a pointed message despite Microsoft’s stakes in OpenAI & Anthropic (here)
DeepSeek founder Liang Wenfeng, in a rare (leaked) 4-hour investor talk, outlined an “almost Daoist” philosophy of AGI as a tide no company can own, advocating restraint over monopoly - hence the firm’s open-source stance & API pricing at 10-mo hardware payback. He argues China’s gap with the US is compute, not talent, & predicts a future where 3-4 base-model players coexist on fair-profit pricing, with DeepSeek focused narrowly on language models, reasoning & agents while ceding video generation & world models to the ecosystem (here, here)
AMD is partnering with Cerebras on disaggregated inference systems that combine AMD GPUs (for prefill) & Cerebras wafer-scale chips (for decode) in a single rack, shipping later this year (here)
Notable deal:
Nvidia committed $5B to Ilya Sutskever’s Safe Superintelligence (raised ~$3B previously at $32B), adding to the pattern of chipmaker backing for AI labs that also buy its hardware (here)
Defence
SpaceX in talks to provide the Pentagon with billions of dollars’ worth of data-centre capacity for running AI models, deepening Musk’s defence ties beyond Starlink & launch services (here)
Pentagon awarded Oracle a $7B, 10-yr contract consolidating on-premises software licences across DoD, Coast Guard & intelligence community, claiming $441M in savings (here)
Defence primes including Lockheed Martin & BAE Systems participated in a record $4.1B in venture rounds this year as Ukraine & Iran conflicts highlight demand for drones, autonomous systems & cheaper weaponry. Lockheed boosted its start-up arm to $1B, BAE committed €50M to Lakestar & Expeditions, & Airbus anchored a new €500M dual-use fund. Quantum Systems raised $1.2B at $8B (Airbus-backed), Kraken Technology raised $175M at $1B (Rheinmetall-backed), while total global defence deal value has topped $40B year-to-date (here)
Notable deal:
Cathedral, a military AI cyber startup founded by four ex-DOGE staffers (including Pentagon’s former chief data officer), raised $160M at $1.4B from a16z & Sequoia despite no revenue or contracts - valuation reflects investor confidence in founders’ government ties as they seek US cyber defence contracts targeting China. The round mirrors capital velocity in defence tech (Anduril recently raised $5B at $61B) & highlights how DOGE alumni are converting access into venture funding (here)
Energy / Climate
Valar Atomics (SMR startup) is in talks to raise $1B at $6B with Sequoia leading, though $450M was already closed at $2B in March - a multi-tranche structure now common in AI-adjacent hardware. The company recently powered an Nvidia chip with its helium-cooled reactor & announced a data centre partnership, as hyperscaler energy demand turns nuclear into a critical AI infrastructure bottleneck (here)
Space
Amazon filed with the FCC to launch 5,000+ satellites for direct-to-device mobile services by 2028, using spectrum from its $11.6B Globalstar acquisition. The constellation - estimated at $13-20B to deploy - would challenge Starlink’s partnerships with T-Mobile & Virgin O2, while Amazon’s existing broadband constellation (390 satellites) lags SpaceX’s 10,000+ (here)
EVs / AVs
Waymo is considering ending its Uber partnership in Austin & Atlanta, planning to launch its own app in January 2028 while remaining on Uber’s platform through May 2028. The move would end Waymo’s exclusivity in both cities & allow Uber to partner with competing AV providers (here)
Other
Elon Musk’s The Boring Company reportedly raising $4B at $20B, a 3.5x step-up from its 2022 round, as it expands beyond Las Vegas to Nashville, Dubai & pitches US metro networks, though Nevada regulators cited nearly 800 environmental violations last year & tunnel workers have suffered serious injuries (here)