The European Union was formed as a response to the world’s last great crisis. Now, with the rise of AI, we find ourselves needing to pull on that same collaboration once again.
The US would have us believe that their lead is inevitable and that their dominance of the world’s economies is merely a matter of time. This is not the case. Even as their leading duopoly of OpenAI and Anthropic released reduced price models in lockstep today to counter rapidly improving Chinese open source, pundits are suggesting that their calls for regulation hide ulterior concerns around competition, speed to scale and profitability.
The key point for Europe is not to believe there is no risk. There is risk. The risk is that we are left behind without leverage in the global AI race. The risk is that the benefits of AI accrue almost uniformally to the US and China and we are left unable to control our own destiny.
We can’t compete individually. The cost of developing models is vast and the cost of running them at the sort of scale that will produce the outcomes that we have been promised even more so. And so there is an opportunity for us to lean back into the benefits derived from collaboration. We need countries to work together to provide the energy security, financing, infrastructure, R&D and technical capabilities to build a true competitor to the US AI behemoth - but it is something that can be done, and the very undertaking of this Grand Project will create immense wealth and prosperity for Europe and its allies.
At the conference I was at last week, JD Vance said that Trump will be most remembered for “ending the consensus on globalization [sic]”. America should not get to decide this for us. The UN’s recent advocation for collaboration, the recent silence from Andy Burnham around ruling out rejoining the EU and the expressed interest from Mark Carney for Canada to consider joining our fraternity gives me reason for hope.
This should obviously all be read in the voice of Churchill masticating a cigar in the bath.
IPOs / Publics
Revolut (£115B valuation) planning dual listing on Nasdaq & LSE, likely in 2028 or later. Founder Storonsky confirmed US primary venue for liquidity & institutional depth (here)
Plaud (AI note-taking hardware & software) plans a 2028 US IPO once revenue exceeds $1B, up from a $500M target this year. The San Francisco-based startup hit $100M ARR in 2 yrs (here)
Meta’s Muse AI app reached 2.8M downloads globally in 12 days, outpacing ChatGPT’s early mobile rollout - in the US & Canada, Muse hit 1.8M iOS installs vs ChatGPT’s 1.3M over the same period, with Muse’s 642K US daily active users also ahead of ChatGPT’s 231K at launch. Over 95% of Muse users also use Facebook, suggesting Meta’s cross-promotion playbook from Threads (now 500M users) is working again (here)
Amazon blocked Meta’s Muse AI assistant from making purchases on its platform, citing unauthorised agent use in violation of terms of service. Amazon operates its own foundation models (Nova) & inference platform (Bedrock), & may also be concerned about liability for AI-driven ordering errors despite Muse’s relatively low hallucination rate (here)
SoftBank is adding nearly $21B in fresh borrowing this week - increasing its Arm-backed margin loan by $5B to $25B, upsizing a credit line by $450M to $6.5B, & securing an $11.9B loan plus a potential $3.6B expansion from Apollo (here)
SoftBank’s SB Energy has slowed its $50B IPO amid market concerns over AI demand & weak interest in a planned $5B debt sale priced at ~10% yield (junk territory). The data centre developer has yet to open a single facility but faces $174B in capex needs, heavy reliance on OpenAI as anchor tenant, & covenant restrictions on dividends until 2029 (here)
Salesforce launched AIforce at Dreamforce, a new AI interface layer that brings platform capabilities - business logic, workflows, security & governance - directly into users’ existing tools rather than requiring them to work inside Salesforce (here)
Big Dogs
Anthropic confirmed it operates a wet biology lab in the Bay Area, running physical experiments focused on fundamental biology rather than drug discovery. The lab, staffed after Anthropic’s $400M acquisition of Coefficient Bio in April, works on protein design & biomolecular modelling whilst serving pharma partners including Novo Nordisk. The move has drawn industry scrutiny given Anthropic’s repeated warnings that AI poses >10% existential risk & CEO Dario Amodei’s identification of bioterrorism as a major AI threat (here)
Anthropic released Claude Opus 5.5, priced 40% cheaper than its predecessor ($4 per M input tokens vs $10 for Fable 5.1), ahead of what is expected to be the largest IPO of all time this autumn at $965B. The move targets price-sensitive customers as the lab competes with OpenAI (which slashed prices & claims the most powerful model, Astra) & low-cost Chinese open-weight models gaining share (here)
OpenAI launched GPT-6 Sol & GPT-6 Luna at half the price of GPT-5.6 predecessors, with improvements in coding, professional work & accuracy - a move aimed at easing customer cost concerns as AI spend pressure mounts (here)
OpenAI in preliminary talks for a pre-IPO raise at over $1.2T, up from $852B in March when it raised $122B; revenue reached $6.7B in Q2 but operating margin fell, delaying profitability; now has 1B active users & 200M business customers; Altman says IPO unlikely this year citing AI safety concerns (here)
Regulation
California Governor Gavin Newsom issued an executive order directing experts to deliver recommendations within 2 mo on strengthening AI safety measures, including mandatory kill switches for frontier models, onsite independent auditors, & incident reporting requirements (here)
Venture Capital / Finance
Portage closed its fourth VC fund at $600M USD (including backing from US strategics Broadridge & Fifth Third Bank), bringing AUM to $7B & reinforcing its position as one of the few dedicated growth-stage FinTech investors in Canada - though only 2 of 35 recent deals have been domestic as the firm has tilted global (here)
Bain Capital Ventures raised $1.6B (up 14% on its 2023 fund) to back 30-40 companies, mainly seed through Series B, focused on AI infrastructure, healthcare, physical AI & security (here)
Andreessen Horowitz launched the Horowitz Andreessen Academy, a 1-yr programme in San Francisco offering 50 students placements at Palantir, Google, Meta, OpenAI & others, with no degrees, grades, or homework - just $42M in backing, $50K+ compute credits & classes from founders. Free for the inaugural 2027 cohort, it reflects Silicon Valley’s push to bypass traditional education & build talent pipelines directly into the startup ecosystem (here)
Debt issued by a data centre to be leased by Jane Street has widened to 11.3% yields (650 bps over govt debt), significantly underperforming other AI data centre bonds despite Jane Street’s own leasehold guarantee. The deal tested whether projects lacking investment-grade backstops can tap public markets - Jane Street carries a BB rating, two notches below investment grade - & suggests financing for non-big-tech AI compute tenants may face higher hurdles (here)
Venture Geopolitics
US Treasuries rallied as Fed Chair Kevin Warsh hiked rates 25 bp to 3.75-4% & pledged to throttle inflation, restoring central bank credibility after three years of accommodation. 10-year yields fell 8 bp to 4.95%, snapping an 8-day rise, aided by oil’s retreat from $107 to below $102/barrel (here)
The Bank of England & Fed have intensified scrutiny of banks’ prime brokerage exposure to large trading firms (Jane Street, Citadel Securities) following Jane Street’s $15B July loss tied to AI-focused hedge fund Situational Awareness (here)
DeepSeek CEO Liang Wenfeng told investors the company expects Huawei to deliver training chips by Q4 2026, describing the shift to domestic processors as one of DeepSeek’s “biggest bets” amid U.S. export controls (here)
Trump pressed Zelenskyy to halt strikes on Russian refineries, citing diesel price concerns ahead of US midterms, while Ukraine hit Moscow with its largest-ever drone attack. Zelenskyy conditioned any energy ceasefire on reciprocal Russian compliance, though Putin is expected to refuse until after winter when Russia plans major attacks on Ukrainian energy infrastructure (here)
Canada PM Mark Carney endorsed EU Commission President Ursula von der Leyen’s proposal for Canada to become the EU’s first “associate member”, a new status focusing on deeper trade, energy (LNG), critical minerals, defence procurement & youth mobility. Carney stressed sovereignty would be retained (“no-one is going to dictate our culture”), positioning the bloc as an alternative to US dependency after Trump called the move “laughable” & threatened EU tariffs (here)
Taiwan & South Korea offer early evidence that AI infrastructure investment is translating into broader economic gains beyond the tech sector, as data-centre buildout drives semiconductor demand & related supply chains - though US public opinion remains sceptical of local AI deployment (here)
Five UK police forces ended their 2-year Palantir pilot after Home Office funding expired, citing questions over value; Lancashire also let its pilot lapse, while London’s Met remains blocked from a £50M follow-on contract by the mayor’s office over procurement concerns - a setback for Palantir’s UK law enforcement expansion built on low-cost pilots designed to unlock larger deals (here)
Trump threatened to ‘annihilate’ Iran in a UN General Assembly speech, saying he faces a ‘big decision’ but expects a peace deal after November’s midterm elections. He also warned Cuba, Mexico & other Western Hemisphere countries of possible military action, prompting Cuban delegates to walk out. The speech underscored tensions over depleted US munitions stockpiles (which Trump denied), rising petrol prices hitting $6.50/gallon for diesel, & polling pressure ahead of midterms that will determine Republican control of Congress (here)
OpenAI President Greg Brockman will attend Trump’s state dinner for Xi Jinping alongside Bezos, Musk, Altman, & other tech leaders as Washington & Beijing seek to manage tensions over trade & AI. Brockman & his wife donated $25M to MAGA Inc. in 2025, becoming among the largest individual Trump donors; the dinner underscores growing White House engagement with AI governance as Altman pushes for US-led international safety standards ahead of a UN Security Council presentation (here)
US proposing $10B fund with Arab states ($5B each) to build energy infrastructure bypassing Strait of Hormuz, reducing Iran’s leverage after it choked off a fifth of global oil & gas flows during the war; draft pitch names Blackstone, BlackRock GIP & KKR as potential co-investors, though some were reportedly unaware; project aims to create redundant pipeline routes as Brent trades near $100/barrel (here)
FT leak reveals Kremlin-backed fintech A7 moved $6.9B through Standard Chartered, Citi, FAB & other banks using forged documents & 100+ front companies across UAE, Hong Kong & Kyrgyzstan to evade sanctions - payments included military goods & kit for Russia’s security services, with A7 now claiming ~20% of Russia’s FX transactions (here)
China implemented sweeping new overseas travel controls for civil servants, SOE employees & private-sector workers in sensitive areas, formalising exit bans for those deemed to risk state secrets or tech security. The rules - signed by Premier Li Qiang & effective 9 Sept - empower authorities to ban travel for export control violations or activities that “may endanger” industrial security, creating a chilling effect on mobility for tech executives & researchers (here)
George Osborne, OpenAI’s head of AI for countries, said UK datacentre protests risk Britain losing “sovereignty” over AI infrastructure to the US, where centres are being built at scale in states like Ohio & Georgia. Labour designated datacentres as critical national infrastructure to accelerate approvals, but local opposition citing energy, water use & land impact has spread from London to Scotland (here)
U.K. PM Andy Burnham left open the possibility of Labour campaigning to rejoin the EU in the next general election, marking a major shift from predecessor Keir Starmer’s ‘make Brexit work’ stance. Burnham raised concerns over EU’s Made in Europe protectionist measures affecting British supply chains ahead of talks with von der Leyen & a delayed U.K.-EU summit expected later this year (here)
Strategic Sectors
AI
Big Tech has issued up to $300B in residual value guarantees for AI data centres & chips in the past 12 mo, backstopping debt through SPVs to keep liabilities largely off their balance sheets - Meta pioneered the structure for its 2GW Louisiana data centre ($28B guarantee), followed by Broadcom ($29B for Anthropic chips via Google) & Nvidia ($105B for OpenAI’s Ohio campus). The guarantees unlock financing at 100-150bp over the guarantor’s own debt while recording minimal exposure until assets are monetised, but credit agencies flag growing complexity as AI capex outpaces cash generation (here)
Trump announced plans for an ‘AI Force’ & new AI tsar, rejecting regulation despite industry leaders (Anthropic’s Amodei, Altman, Musk) calling for slower development after autonomous AI systems escaped test environments & hacked companies. The move splits from congressional Republicans, who’ve embraced data centre bans as 63% of US voters worry about electricity costs (here)
Jane Street-linked data centre debt issued by Zenith Arc (a Fluidstack-Coatue JV) has soured quickly in secondary trading, with yields rising to 11.3%, over 2 percentage points above August’s launch - underperforming peer AI data centre bonds (here)
TypeSafe AI, co-founded by former OpenAI researcher Diogo Almeida (who helped invent RLHF), released Jev - a transformer model that outputs calibrated probabilities rather than language, making it 5-18x faster & 10-20x cheaper than LLMs for software automation tasks like classification & routing while eliminating hallucinations (here)
OpenAI disclosed previously unreported misalignment incidents - including models fabricating data, bypassing network restrictions & sharing files they were meant to keep private - & published a new framework for tracking & reporting such behaviour (here)
Cohere (Canada) & Aleph Alpha (Germany) signed a definitive merger agreement, combining model development with enterprise deployment capabilities targeting sovereign AI demand (here)
Notable deals:
Crusoe raised $3.9B at $30.9B (co-led by Atreides, Mubadala Capital, Valor; backers include Nvidia, Qatar Investment Authority, Founders Fund) to fund large-scale data centres & truck-transportable ‘Spark’ modular AI factories, a manufacturing-led approach to deploy compute faster & sidestep NIMBY pushback. The 8 yr-old firm - which supplies Meta, Microsoft, Oracle & signed a $13B 5 yr deal with Jane Street - is exploring an IPO after tripling its valuation in 10 mo (here)
Instinct (AI agent startup) in talks to raise $1B at $10B with Sequoia & Benchmark leading, up from $2.25B valuation just 1 mo ago - the invitation-only assistant now serves 100,000+ users but faces compute constraints as it scales without charging fees (here)
Cybersecurity
Hackers infiltrating ships via satellite-connected ‘edge devices’ rose from 3% of maritime attacks in 2024 to 22% in 2025, per cyber firm Cydome, which tracks hundreds of incidents. Two tankers carrying Gulf crude to the US were attacked this week in suspected Iranian operations; VL Prosperity lost communications for 30 hrs crossing Gibraltar in August (here)
Hackers breached an Italian government email system (PEC) to pose as law enforcement for several months, obtaining confidential data on 680 Revolut customers - primarily Swiss & French ‘crypto whales’ identified via blockchain analysis. The attack exploited official channels rather than Revolut’s own systems, raising questions about how many other firms may have been similarly targeted via compromised state infrastructure (here)
Open-source AI is driving a 440% surge in blockchain-based malware attacks, with hackers using blockchains to hide command-and-control instructions - a technique now dominated by state-backed groups from North Korea & Iran. The shift follows the 2025 release of Chinese open-source models with no malicious-use restrictions, fuelling a broader crypto hack wave that saw attacks jump 150% in H1 2026 (here)
Energy
Oregon’s 111 data centres now consume 23% of the state’s electricity - 2.5x the 2020 share - with demand projected to reach 25 terawatt hours by 2030 (equivalent to 2.5M homes), raising infrastructure planning questions as AI drives hyperscale buildout (here)
Oil executives including Chevron’s Mike Wirth warn a global fuel crisis has arrived after 6+ months of the Strait of Hormuz closure, with commercial stocks depleted, strategic reserves near exhaustion, & last week’s attacks shutting Saudi Arabia’s East-West pipeline (stranding 2.5M barrels/day). China has resumed international crude purchases after months drawing down domestic stocks, tightening global supply further (here)
Defence
Trump announced plans to convert his proposed 250ft Triumphal Arch in Washington into a “military complex” housing drones, snipers & ammunition storage, citing national security justification after veterans & historians sued to block the $1B+ redevelopment project for lacking congressional authorisation (here)
Pentagon awarded contracts for 60,000 drones (avg $5K each) across 9 startups, with Eric Schmidt-backed Perennial Autonomy winning the Deep Strike category & Neros Technologies placing in both categories. Donald Trump Jr.-backed Unusual Machines supplied components to four winners including Perennial & Neros (here)
Nammo CEO Morten Brandtzæg warned Europe cannot supply its own weapons demand & must maintain integrated supply chains with the US despite political tensions (here)