📨 AI Navigates Narrow Straits

Read time: 5 min

This week, we cover:

Watch the 3-min video overview:

Google has granted the Pentagon access to its Gemini AI for classified military use under “any lawful governmental purpose,” aligning with similar agreements made by OpenAI and xAI. This decision follows an ongoing dispute with Anthropic, which was barred from Pentagon work after refusing to remove guardrails on autonomous weapons and surveillance [The New York Times, TechCrunch]. Over 600 employees have signed a letter, expressing concerns that the deal could facilitate lethal weapons or mass surveillance [The Information].

China has ordered Meta to divest its $2–2.5B acquisition of AI startup Manus, citing national security concerns and mandating the separation of data and assets. This action highlights increasing control over domestic AI and elevated regulatory risks for cross-border investments. It signals that AI is a geopolitical leverage point, making cross-border AI deals high-risk [The Wall Street Journal, Bloomberg].

OpenAI reportedly missed growth targets, raising concerns about its ability to meet substantial compute commitments. This challenges “circular” AI deals that link funding with infrastructure, potentially impacting cloud providers, Nvidia, and AI startups if contracts are renegotiated [Pitchbook].

AI agent wipes startup database in 9 seconds, exposing weak safeguards: A Claude Opus 4.6 agent within Cursor deleted PocketOS’s production database and backups, ignoring explicit rules and causing a 30-hour recovery. This incident highlights how agentic systems can fail when tool permissions and API controls are overly broad [X, Techloy].

AI data centers turn power, water, and grids into investor risks: AI’s rapid data center expansion is driving significant demand for electricity and cooling, leading to concerns about emissions, water usage, and grid strain. This trend also creates investment opportunities in grid infrastructure, efficiency, storage, cooling, and water management [IFAmagazine].

EU finds Meta in breach of Digital Services Act over child safety failures: The European Commission has issued a preliminary finding that Meta breached the Digital Services Act (DSA) due to weak age verification and inadequate reporting tools, failing to block users under 13 on Instagram and Facebook. If confirmed, fines could reach 6% of global annual revenue, potentially close to $10B [European Commission]. This ruling comes shortly after Meta increased its capital expenditure guidance to $145B and announced 8,000 layoffs, underscoring the contrast between its aggressive AI expansion and growing regulatory exposure [Fortune].

Big tech Q1 earnings - AI demand lifts cloud, but capex bills keep rising: Google, Amazon, Microsoft, and Meta simultaneously reported Q1 earnings, with AI-driven cloud demand emerging as a clear winner. Google Cloud grew 63%, and its backlog doubled to $460B; AWS accelerated to 28% growth. However, both Meta and Google raised capital expenditure guidance to $145B and $190B, respectively, with much of the compute demand likely originating from AI labs rather than broader enterprise adoption. Following these reports, Google saw a +7% increase and Amazon a +3% increase after hours, while Meta was -6.6% and Microsoft remained flat [The Information 1, 2, 3, 4].

Microsoft commits $18B to expand AI and cloud infrastructure in Australia: Microsoft will invest $18B to enhance Azure capacity, cybersecurity, and AI training in Australia, strengthening its ties with the government. This initiative supports Canberra’s efforts to attract hyperscaler investment amid intense global competition for AI infrastructure [CNBC].

Space-based solar power - the AI energy crisis may have finally made it viable: Reduced launch costs and a 14–17% annual growth in AI data center demand through 2030 may finally make space-based solar power economically viable. Companies to watch include Aetherflux (laser transmission), Virtus Solis, Solestial, Star Catcher, and Reflect Orbital. Success in this area could transform both the energy and space sectors [CLNM Capital at NYSE Space Summit 2026].

OpenAI ends Microsoft exclusivity and brings models to AWS: OpenAI will make its models and Codex agent available on Amazon Web Services, following a loosening of its exclusivity agreement with Microsoft. This move expands enterprise access and signals a multi-cloud strategy as OpenAI intensifies its commercial push [OpenAI, CNBC].

Lilly and Profluent sign $2.25B AI gene-editing pact: Eli Lilly and Profluent Bio have partnered to develop AI-designed DNA-editing tools for hard-to-treat genetic diseases. The deal grants Lilly licensing rights and reflects its broader strategy in AI-driven drug discovery and biotech collaborations [BioSpace].

Biohub launches $500M open-data push for AI biology: Biohub has initiated a five-year, $500M program to build open datasets and tools for predictive cell models. This effort aims to accelerate drug discovery and disease understanding by integrating AI, biology, and large-scale global data sharing [Biohub].

MIT and IBM expand lab to advance AI and quantum computing: MIT and IBM have launched a new research lab focused on AI, algorithms, and quantum computing. The partnership aims to develop hybrid computing methods, new quantum algorithms, and practical AI systems for scientific and enterprise applications [MIT News].

New chip design improves data center power efficiency: Researchers have developed a chip using piezoelectric components to convert power more efficiently, achieving 96.2% efficiency in tests. This approach could reduce energy loss and size in data centers, though commercialization is still in early stages [UC San Diego].

Goldman blocks Anthropic Claude for Hong Kong bankers: Goldman Sachs has restricted its Hong Kong staff’s access to Anthropic’s Claude, citing increased scrutiny of AI data and cyber risks. Gemini and ChatGPT remain available, while regulators are also reviewing Anthropic’s newer model, Mythos [Reuters].

Greg Brockman provides a rare first-person account of OpenAI’s internal history, the 72 hours following Sam Altman’s firing, and the company’s transition from nonprofit ideals to scaled commercial AI. The interview also covers OpenAI’s current compute strategy, the future of ChatGPT and AGI, and AI’s potential impact on jobs [Farnam Street].

“I sought good judgment mostly by collecting instances of bad judgment, then pondering ways to avoid such outcomes.” - Charlie Munger

That’s it for this week.

The CLNM Capital