This week, AI investing sits at the intersection of policy, hardware realignment, and new autonomy infrastructure. U.S. and EU regulators are testing how far they can shape AI economics, from federal preemption in Washington to a proposed cultural levy in Europe. At the same time, Arm’s direct entry into AI data‑center CPUs and helium supply disruption sharpen the hardware bottleneck story, while “agent utilities” and spatial data platforms quietly build the rails for always‑on AI systems.
Watch the 2-min video overview:
🌍 Landscape
White House plan for AI policy
The White House has released an AI policy blueprint calling for federal preemption of state AI laws. The framework states training AI on copyrighted material is legal but defers to judicial interpretation, proposes regulatory sandboxes, and opposes creating new federal oversight bodies. Over 50 Republican state legislators opposed the approach arguing it shields large tech companies. Despite a 2026 target, congressional consensus remains elusive, making passage uncertain. [NYT]
Mistral's proposed cultural content levy highlights EU's implementation challenge
Mistral CEO proposes 1-1.5% revenue levy on all AI providers selling in Europe (led by US hyperscalers) to fund creators and clarify training data rules [Le Monde, FT] — echoing the Masters of Digital Forum on Feb. 26: "the problem lies not in regulation but in ineffective implementation” [CLNM newsletter].
Arm launches AGI CPU and enters AI chip market
Arm has launched the AGI CPU, its first in-house processor built for AI agent infrastructure — marking a shift from pure IP licensing to direct chip sales. The move places Arm in direct competition with customers like Nvidia, Intel and Qualcomm, while reinforcing its growing influence in the data center market through energy-efficient, high-density designs. [Arm, Forbes]
Arm AGI CPU. Arm Holdings
📈 Tailwinds
Clean energy surge powers China’s battery giants: After strikes on Iran, investors shifted from fossil fuels to clean energy, with China’s battery giants capturing most of the upside; CATL, BYD and Sungrow added over $70B in market as grid storage became a security hedge [FT].
Agent infra as a new ‘utilities’ layer: OpenClaw adoption (GitHub stars exponential, token consumption spiking) signals real demand for always-on agents. Early value accrues to network and delivery providers (Content Delivery Networks like AKAM, FSLY, NET) that host always-on agents. [Citrini Research]
Source: Citrini Research
The "Pokémon approach” on spatial data: Startup Niantic Spatial crowdsourced 30B images from Pokémon Go players over a decade and now uses this dataset to power precise Visual Positioning Systems, helping Coco Robotics’ 1,000 delivery bots navigate city streets where GPS fails [Fortune]. Mapped 3D worlds can further unlock AR glasses, self-driving cars, digital twins, etc. favoring firms with proprietary sensor datasets.
📉 Headwinds
Hormuz He2 shock: The Strait of Hormuz closure has cut off roughly a third of global helium supply, creating an estimated 15% net shortage that threatens advanced AI chip production, with leading Asian fabs most exposed. US and EU chipmakers face delayed deliveries. [Scientific American]
World production of helium gas in 2024 (according to U.S. Geological Survey, Mineral Commodity Summaries, Jan 2016 & 2025, https://doi.org/10.3133/mcs2025).
Software companies lose value as AI takes over jobs: Salesforce, Atlassian and HubSpot stocks dropped sharply after news that AmazonWS is building AI to automatically handle sales and business development tasks [The Information].
Super Micro crashes on alleged Nvidia chip smuggling to China: Super Micro shares dropped 33% after a U.S. indictment alleged a co‑founder and two associates illegally exported Nvidia AI chips to China, highlighting enforcement risks around U.S. export controls and AI hardware supply chains. [CNBC, Fortune]
🚀 Deals & Capital Flows
Bezos seeks $100B fund for AI chip, defense & aerospace manufacturing likely channeling them into his secretive Project Prometheus AI startup. Recent travels indicate that the fundraising targets Middle East and Singapore. [WSJ, Forbes]
Halter's “cowgorithm" collars target $2B valuation: New Zealand’s Halter, maker of AI-powered solar collars for virtual cattle fencing and health monitoring, is raising a new round doubling its valuation with Peter Thiel's Founders Fund leading. [Bloomberg]
Lace raises $40M for AI chip fab breakthrough: Barcelona startup's nanotech prints circuits 10x smaller than current methods, solving next-gen AI chip bottlenecks. Atomico leads Series A. [Reuters, TechStartups]
OpenAI acquires Astral to supercharge Codex dev tools: OpenAI is buying Astral, creators of open-source Python tools (uv, Ruff, ty) used by millions. The deal integrates these into Codex (now with 2M weekly users) to expand AI from code generation to full development workflows like planning, testing, and maintenance. [OpenAI]
+ News on AI
MIT AI system reconstructs 3D objects through walls using wireless signals
MIT researchers developed Wave-Former, an AI system using wireless signal reflections to reconstruct 3D objects hidden behind walls or inside containers without cameras — achieving higher accuracy than prior methods with a single sensor. The breakthrough solves a critical constraint for autonomous systems: spatial awareness in GPS-denied, camera-free environments. Applications span warehouse robotics (inventory tracking without visual surveillance), factory floor parametrics (real-time process monitoring through opaque containers), and autonomous inspection systems. This infrastructure layer becomes essential as autonomous AI labs expand beyond biotech into materials science and manufacturing, where parametric monitoring without privacy-invasive imaging reshapes both efficiency and regulatory compliance. [MIT News].
Image credits: MIT News
Lab study reveals AI agent security failures
Researchers tested autonomous AI agents in a live lab: agents leaked data, obeyed strangers, ran destructive commands, and lied about task status despite real system damage. Vulnerabilities demand urgent fixes. [arXiv]
Crypto bets on AI agents as key to mass adoption
Crypto firms are repositioning as infrastructure for AI agent payments, citing x402's 107M transactions and McKinsey's $3–5T agentic commerce projection by 2030. Skeptics note current volume is negligible and trust infrastructure remains unbuilt [Forbes]
What to Watch
Will Congress pass federal preemption of state AI laws? Either outcome favors Big Tech — federal preemption means one compliance regime; fragmented state laws favor companies with resources to navigate 50+ jurisdictions. Regulatory risk for smaller AI startups rises if one national standard is written around Big Tech inputs.
EU revenue levy test: If adopted, watch whether U.S. hyperscalers exit European markets or absorb the cost. Either signals how much moat they've built. European startups win only if levy funds credible alternatives.
Agentic Utilities test: Watch whether hyperscalers build dedicated ‘agent infra’ or bolt-on to existing CDNs. Either validates agentic demand. Winners emerge only if B2A APIs create credible network effects.
🎓 Webinars & Events
📅 April 6-10, 2026 | 🌐 Virtual (free)
Data Science Dojo's virtual event on governance, enterprise adoption, and security for autonomous agents. Speakers include CrewAI, Fiddler AI, LlamaIndex, Google DeepMind, Microsoft, Docker, and Oracle. Register
Valence Thoughts
It ain‘t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so. - Mark Twain
That’s it for this week.
The CLNM Capital