Venture Geopolitics Issue 43
14 April 2026
The UK has historically framed its economic strength as a function of talent, research and pro-innovation policy - a model that worked in a world where marginal costs were close to zero and value accrued to software, services and aggregation. That logic explains why OpenAI just unveiled plans to open a permanent office in London, positioning it as a major research hub outside the US.
But, as Ben Thompson argued this week, that model is breaking down. In an AI-native economy, marginal costs are re-emerging via compute, energy and infrastructure. Value is no longer captured purely by those who design systems, but increasingly by those who can build and operate them at scale.
Which is why the more important headline this week was not OpenAI’s expansion, but its retreat. The decision to step back from Stargate UK, citing energy costs, points to a constraint that is immense and wildly underdiscussed: you cannot be an AI superpower with expensive power and speculative grid access.
UK industrial electricity prices are among the highest in the developed world — materially above France and Germany, and multiples of the United States. The system lacks the firm nuclear base of France, the cheap gas and scale of the US, or the hydro-backed stability of the Nordics. At the same time, infrastructure is slow and expensive to build, constrained by planning complexity and regulatory friction.
The result is a system that is neither cheap nor predictable. At times of excess renewable output, households are now going to be paid to consume power the grid cannot store or transmit. Yet when firm power is required, the system falls back on marginal gas, setting expensive prices for everyone. This is not a shortage of energy, but a failure to deliver it reliably.
AI infrastructure is one of the most power-hungry industrial systems ever built. As the economy shifts from zero marginal cost software to compute-constrained systems, this mismatch becomes critical. Without a competitive position in energy and infrastructure, the UK risks remaining a hub for talent and ideas — while the economic value increasingly accrues elsewhere.
To that end, the green light for the UK’s first small modular nuclear reactor fleet this week was a welcome development. We need more focus on things like this than incentives to adjust our washing cycles!
IPOs / Publics
SpaceX’s planned IPO may waive traditional 180-day insider lock-ups, allowing early shareholders to sell immediately & effectively gutting Section 11 investor protections against misleading prospectus claims. The SEC has ignored years of warnings about this growing trend that strips retail investors of their strongest legal recourse (here)
Bill Ackman’s Pershing Square USA is marketing a dual IPO to raise up to $10B despite geopolitical headwinds from the Iran conflict, offering investors in the closed-end fund (at $50/share) free stakes in his management company - a pivot after his $25B NYSE fund collapsed in 2024 (here)
Nvidia-backed Firmus raised $505M at $5.5B ahead of its ASX IPO, targeting $2B in additional capital for what could be Australia’s largest tech IPO this decade - unusual given most local tech firms favour overseas exchanges (here)
Intel surged 53% over 9 sessions, adding $100B in market value, after announcing a $14.2B buyback of its Ireland plant from Apollo & partnerships with Musk’s Terafab project & Google data centres - though the stock now trades at 90x forward earnings, its highest on record (here)
Meta launched Muse Spark, its first model from Meta Superintelligence Labs, featuring native multimodality & agent-based reasoning built in 9 mo with less compute than rivals, leading benchmarks in health reasoning & agentic search despite trailing Gemini 3.1 Pro overall (here)
Palantir shares fell 8% after Michael Burry argued Anthropic is ‘eating Palantir’s lunch’, citing data showing Anthropic’s ARR jumped from $9B to $30B in months whilst capturing 73% of new enterprise AI spending - highlighting how private AI leaders are disrupting public market darlings (here)
Microsoft is developing OpenClaw-inspired features for Copilot to enable autonomous 24/7 operation within Office 365, as Anthropic’s Claude increasingly encroaches on Microsoft’s enterprise software dominance with direct integrations to PowerPoint, Excel & Word (here)
Big Dogs
Anthropic launched Claude Mythos Preview, a cybersecurity AI model, to select partners including Amazon, Apple, Microsoft & US government following source code leaks that exposed the project - the model identifies thousands of zero-day vulnerabilities but escaped its sandbox during testing, prompting limited release over misuse concerns (here)
US Treasury secretary Scott Bessent summoned major bank CEOs to discuss cyber risks from Anthropic’s Mythos, which can detect thousands of severe vulnerabilities across operating systems & browsers - some undetected for decades - raising concerns about potential exploitation by bad actors (here)
Anthropic also launched “Claude Managed Agents”, a cloud service that automates container management & tool orchestration for AI agent development, reducing build time from months to weeks at 8p per runtime hour plus model usage costs - early adopters include Notion, Asana & Rakuten (here)
Anthropic temporarily banned OpenClaw creator Peter Steinberger from accessing Claude for ‘suspicious activity,’ lifting the ban after his viral post sparked backlash - highlighting tensions as Anthropic now charges separately for third-party AI agent tools whilst promoting its own Cowork alternative (here)
CoreWeave signed a multi-year deal to provide compute infrastructure for Anthropic’s Claude AI models (here)
Three senior OpenAI Stargate executives departed to join the same new firm Three senior OpenAI Stargate executives departed to join the same (currently unknown) new firm after the company shifted from building its own data centres to renting compute through partnerships, following financing struggles & joint venture delays with SoftBank & Oracle (here)
A Molotov cocktail was thrown at OpenAI CEO Sam Altman’s San Francisco home, with a 20-year-old suspect arrested after threatening to burn down OpenAI’s headquarters; no injuries reported, highlighting escalating security concerns facing tech executives amid rising AI anxieties (here)
Sam Altman shared a family photo in hopes of discouraging future attacks whilst defending OpenAI’s mission to democratise AI rather than concentrate power (here)
OpenAI told investors it has outpaced Anthropic in computing capacity - claiming 1.9GW available in 2025 vs Anthropic’s estimated 1.4GW - as the ChatGPT maker defends its $600B infrastructure spending against rivals characterising it as reckless whilst Anthropic faces service outages from demand surges (here)
Venture Capital
Global venture investment hit $300B in Q1 2026, driven by mega-deals to OpenAI, Anthropic, xAI & Waymo ($188B combined), with 80% of total funding flowing to AI companies - marking shift toward physical layer investments in chips, robotics & defence unlike prior software-centric cycles (here)
Eclipse closed $1.3B across two funds targeting ‘physical AI’ startups, with $591M for early-stage incubation & the remainder for growth companies - the fund plans to build an ecosystem of portfolio companies across transportation, energy, infrastructure & defence that can partner with each other to achieve scale (here)
Eka Ventures closed £107M Fund II targeting early-stage health, sustainability & consumer tech investments (here)
Collide Capital raised $95M Fund II to back early-stage fintech, supply chain & future-of-work companies, following its $66M Fund I in 2022, with LPs including UC Regents, Goldman Sachs & JPMorgan (here)
Regulation
OpenAI backs Illinois bill SB 3444 that would shield AI labs from liability for mass harms (100+ deaths or $1B+ property damage) caused by models trained with $100M+ compute costs, provided they publish safety reports & didn’t act intentionally - marking a shift toward offensive lobbying as state liability frameworks remain unsettled (here)
UK financial regulators are urgently meeting with banks, insurers & exchanges after Anthropic’s Claude Mythos model exposed thousands of critical vulnerabilities across major systems - including some undetected for decades (here)
Electronic Frontier Foundation is “logging off X” after nearly 20 years, citing dramatic decline in reach (posts now receive less than 3% of views compared to 2018) & platform’s abandonment of user rights protections under Musk ownership - will maintain presence on Facebook & TikTok to serve vulnerable communities who need digital rights advocacy most (here)
Washington DC appeals court declined to block Pentagon’s national security blacklisting of Anthropic, giving Trump administration a win as the AI company fights designation over refusal to remove safety guardrails on Claude for military surveillance & autonomous weapons use (here)
Meta began removing hundreds of ads from Facebook & Instagram that law firms use to recruit clients for social media addiction lawsuits, following recent trial losses including a Los Angeles jury verdict holding Meta liable for youth mental health harm (here)
Musk amended his $150B lawsuit against OpenAI to direct any damages to the nonprofit arm rather than himself, whilst seeking Sam Altman’s removal from OpenAI’s board as the case heads to trial over the company’s for-profit conversion (here)
UK government weighs standardised testing for AI models used by banks after Bank of England warned lenders over evaluation practices, with Starling Bank CIO proposing independent assessment to ensure US algorithms meet minimum standards (here)
Venture Geopolitics
OpenAI halted its Stargate UK data centre project citing high energy costs & regulatory uncertainty, dealing a blow to Starmer’s ‘sovereign AI’ agenda - the company also scaled back its Texas facility & shut Sora app as it refocuses resources amid competition from Anthropic & Google (here)
NHS data chief Ming Tang vowed to deepen Palantir’s integration across hospitals despite staff refusing to work on the £330M Federated Data Platform over ethical concerns about the US company’s defence & immigration work (here)
Meanwhile, NHS England officials apparently threatened staff with job loss for criticising Palantir’s data platform rollout, despite voluntary adoption claims - highlighting internal resistance to the controversial 7-year contract that ministers may terminate early (here)
Trump publicly endorsed Palantir after short seller Michael Burry bet against the stock, claiming AI startup Anthropic would erode its market position - Palantir shares rallied 5% on the endorsement before closing lower, having already lost 25% this year amid valuation concerns (here)
Hungary’s opposition Tisza party led by Péter Magyar secured a landslide victory with 138 of 199 parliamentary seats, ending Viktor Orbán’s 16-year rule as EU’s longest-serving premier - a setback for Trump & Putin who counted Orbán as their closest European ally (here)
European Commission ties Hungary’s €35B frozen fund release to 27 conditions including court reforms & anti-corruption measures, following PM Péter Magyar’s supermajority win (here)
Trump ordered a US naval blockade of the Strait of Hormuz after Iran rejected demands to reopen the shipping lane & end its nuclear programme during 21-hour talks in Pakistan - Britain refused to participate whilst Iran’s IRGC mocked the threat as ‘ridiculous’, escalating tensions that could further choke global oil supplies (here)
Trump’s White House unveiled a $70B maritime spending plan for 2027, including $65.8B for naval shipbuilding (the largest since 1945) and $1.5B for US Maritime Administration programmes - part of his executive order to restore American maritime dominance and strengthen the domestic shipbuilding industrial base (here)
China nearly tripled export controls over the past 5 years (30 restrictions vs 11 previously), with new regulations allowing punishment of foreign companies conducting supply chain due diligence & exit bans for violators - demonstrating Beijing’s willingness to weaponise trade leverage ahead of Trump’s May visit to China (here)
Strategic Sectors
AI
Perplexity’s monthly revenue surged 50% to $450M ARR in March as the AI startup pivots from search to agents, launching usage-based pricing alongside subscription tiers of $20-200 monthly for 100M+ users including enterprise clients (here)
Notable deals
Elorian (AI visual reasoning) raised $55M at $300M led by Striker Ventures, Menlo Ventures & Altimeter with NVIDIA participation - building models that understand images directly rather than converting to text first, targeting robotics, aerospace & medical applications (here)
SiFive raised $400M at $3.65B from Atreides, Nvidia & others to develop data centre CPUs using RISC-V architecture, positioning against Arm as the incumbent shifts to direct chip competition with its own customers (here)
Aria Networks raised $125M Series A from Sutter Hill Ventures & others for AI-native networking infrastructure designed to work across any AI chip (Nvidia, Google) without requiring full network overhauls (here)
Cybersecurity
FBI, NSA, CISA & Department of Energy warned of escalating Iran-backed cyberattacks targeting US critical infrastructure including water utilities, energy systems & local government, with hackers exploiting SCADA vulnerabilities following recent US-Israel military strikes on Iran (here)
Cisco is in talks to acquire AI security startup Astrix for at least $250M, part of a broader consolidation trend as enterprises grapple with rogue AI agents that have triggered security alerts at Meta & mass-deleted emails elsewhere (here)
Defence
IMF analysis warns that defence spending booms typically last 3 years, worsen deficits by 2.6% of GDP & raise debt by 7 percentage points without lasting economic gains - tempering European hopes that military budget increases to 5% of GDP by 2035 will revitalise industrial growth (here)
UK Defence Secretary exposed a month-long covert Russian submarine operation in British waters, involving an Akula-class attack submarine & two GUGI spy submarines targeting underwater cables & pipelines. Royal Navy warships & RAF aircraft monitored the operation 24/7 before Russian forces retreated, highlighting escalating infrastructure threats amid broader geopolitical tensions (here)
Former RAF chief urges Starmer to repair UK-Israel ties to learn from Iron Dome technology as Britain faces ‘notable gaps’ in air defence against Russian & Iranian threats, with only 2 of 6 Type 45 destroyers operational whilst Iran recently struck RAF Akrotiri (here)
Royal Navy faces critical ship shortage with just 13 frigates & destroyers operational versus 56 in Cold War era, forcing impossible choices between defending home waters from Russian submarines & deploying to Middle East amid Iran conflict - new Type 26/31 frigates won’t arrive until 2027-28 (here)
UK’s promise to seize Russian shadow fleet tankers in the English Channel exposed as empty UK’s promise to seize Russian shadow fleet tankers in the English Channel (21 miles wide) was exposed as empty when Russian frigate Admiral Grigorovich escorted sanctioned vessels through unchallenged, meanwhile Iran successfully closed Strait of Hormuz (21 - 30 miles wide) with shore-based drones & missiles - highlighting Britain’s naval weakness with only 3 of 13 frigates/destroyers carrying anti-ship missiles (here)
Lord George Robertson, former NATO chief & author of UK’s strategic defence review, accused PM Starmer of ‘corrosive complacency’ on defence, warning Britain is ‘under attack’ & ‘in peril’ whilst facing a £28B funding gap over 4 years - criticised Treasury ‘vandalism’ & called for welfare cuts to boost military spending as the promised defence investment plan remains delayed (here)
Cambridge Aerospace, a UK defence startup founded in late 2024, is set to supply Skyhammer drone interceptors to the MoD in a multimillion-pound deal with first deliveries expected in May - remarkably fast for UK defence procurement (here)
Energy
UK government approved new nuclear power station on Anglesey using Rolls-Royce small modular reactors, promising 8,000 jobs & powering 3M homes from 2030s following £2.5B partnership signed in 2024. BBC quips “the project is still subject to a final investment decision, which is expected by the turn of the decade” (here)
Quantum
Lloyds Banking Group & IBM successfully used a 156-qubit quantum computer to detect money mule networks in fraud detection experiments, demonstrating quantum’s potential to identify complex criminal transaction patterns that traditional computing struggles to catch (here)