Feds Lift Anthropic Export Restrictions
1. Key Themes
AI Infrastructure Cost Compression Is Buying Labs Strategic Runway
OpenAI's software efficiency gains are reducing hardware dependency dramatically — a signal that the AI capex arms race may be more manageable than feared.
"OpenAI cut inference costs for guest ChatGPT users by more than 50%, dropping the number of Nvidia GPUs needed to serve them to a few hundred and underscoring how software optimization could buy AI labs breathing room amid slow data-center buildouts."
Reinforcement Learning Is Moving From Games to High-Stakes Financial Markets
The same AI training technique that beat humans at poker is now being deployed at scale in equity markets — with verified results and institutional backing.
"The common denominator between poker and Wall Street is that they are well suited for reinforcement learning, an AI training technique where self-learning models are incentivized by rewards… EquiLibre's algorithms have been trading billions in daily volume across the S&P 500 and Nasdaq… with 'a perfect record of zero negative months since inception.'"
Industry Consortium Stablecoins Are a Structural Threat to Circle and Tether's Business Model
Over 100 major financial institutions are joining forces to launch a shared stablecoin, directly attacking the incumbent duopoly's economics.
"Visa, Stripe, Coinbase, Mastercard, BlackRock, and more than 100 other companies are joining an industry consortium called Open Standard to launch Open USD, a stablecoin launching later this year that will directly challenge the economics behind dominant stablecoin issuers like Circle and Tether."
Agentic AI Is Disrupting Travel Distribution, Threatening Booking Incumbents
AI-native booking layers are emerging as a new intermediary that could redirect consumer demand away from established platforms.
"Hotels, travel agencies, and aggregators are racing to build loyalty programs and AI-native booking tools as agentic chatbots threaten to become a new travel-planning layer that could steer customers away from Expedia, Booking.com, and other incumbents."
AI Adoption Correlates With Headcount Growth, Not Reduction
Contrary to the dominant narrative that AI destroys jobs, a large-scale study finds the opposite among heavy AI adopters.
"A study by Ramp and Revelio Labs of nearly 22,000 companies found that heavy AI adopters — defined as firms spending about $30 per employee per month on AI for 3 months — grew headcount 10.2% overall and 12% in entry-level roles, countering the argument that AI kills jobs."
2. Contrarian Perspectives
AI doesn't kill jobs — it correlates with hiring, especially at entry level. The default assumption in most boardrooms and policy circles is that AI adoption leads to workforce reduction. The Ramp/Revelio Labs data directly contradicts this at scale. Heavy AI adopters across nearly 22,000 companies grew headcount 10.2% overall and 12% in entry-level roles. This has meaningful implications for hiring strategy, investor diligence on workforce risk, and policy debates.
Software optimization — not just hardware buildout — may be the decisive AI moat. The prevailing investment thesis has been that whoever builds the most data centers wins. But OpenAI's >50% inference cost reduction, achieved through software alone and dropping GPU needs "to a few hundred," suggests that algorithmic efficiency could be a more durable and capital-efficient competitive advantage than raw compute scale.
EquiLibre explicitly rejects the "finance firm" label despite operating in finance. Most quant shops optimize for AUM and trading revenue. EquiLibre's investor, Creandum VP Cameron Sellers, noted the company "explicitly defines itself as 'a lab first, not a finance firm.'" This identity positions them to attract research talent and build transferable IP rather than becoming a single-strategy fund — a potentially more defensible and scalable business model.
3. Companies Identified
Anthropic AI model developer Why mentioned: The U.S. Department of Commerce lifted export licensing requirements on its most advanced models — Mythos and Fable — opening up global access.
"The U.S. Department of Commerce has lifted a requirement that Anthropic obtain a license before exporting its Mythos and Fable models abroad, a restriction that effectively cut off public access to what are widely considered the most advanced AI models released to date."
EquiLibre Technologies Prague-based AI trading lab, founded by former DeepMind researchers Why mentioned: Valued at $500M after a record Series A from Creandum; applying reinforcement learning from poker AI to live equity and crypto markets.
"EquiLibre's algorithms have been trading billions in daily volume across the S&P 500 and Nasdaq… with 'a perfect record of zero negative months since inception.'"
Realta Fusion Private fusion energy company Why mentioned: Claimed to be the first private company to publicly demonstrate direct electricity generation from a fusion reaction.
"Realta Fusion says it became the first private company to publicly show direct electricity generation from a fusion reaction, using its Wisconsin test reactor to produce usable power without first converting heat into steam."
Etched San Jose AI inference chip startup Why mentioned: Emerged from stealth with a working chip, $1B+ in signed contracts, and $800M raised at a $5B valuation — a major signal in the custom silicon for AI space.
"Etched emerged from stealth with a working chip, $1+ billion in signed customer contracts, and $800 million raised across multiple rounds, including a $500 million financing at a $5 billion post-money valuation."
Bending Spoons Milan-based software acquirer/revamper Why mentioned: IPO priced above range at $18.4B market cap, validating the buy-and-revamp aging software brand model.
"Bending Spoons priced its Nasdaq IPO above range, raising $1.68 billion and giving the Milan company that buys and revamps aging software brands like Vimeo, WeTransfer, Evernote, and AOL an $18.4 billion market value."
Cognite Oslo industrial AI software company Why mentioned: Acquired by Schneider Electric for $3.1B in cash — a major exit validating industrial data/AI infrastructure.
"Schneider Electric is acquiring Cognite, a nine-year-old Oslo industrial AI software company whose tools organize factory, energy, and infrastructure data for analytics and AI workflows, for $3.1 billion in cash."
Higharc Durham, NC AI software for homebuilders Why mentioned: Raised a $95M Series C — a signal of sustained enterprise investment in construction tech AI.
"Higharc develops AI software for homebuilders to automate design, estimating, sales, and construction workflows."
Dominion Dynamics Ottawa defense-tech startup Why mentioned: Raised $97.9M Series A for command-and-control software and autonomous drone systems in extreme environments — reflects continued surge in defense tech investment.
Open Standard / Open USD Industry stablecoin consortium Why mentioned: The mechanism through which Visa, Stripe, Coinbase, Mastercard, BlackRock, and 100+ others are challenging Circle and Tether.
Pie New York AI search visibility startup Why mentioned: Raised $19.5M Series A from Lightspeed to help small businesses appear in AI search results — an emerging category as traditional SEO erodes.
Stathera Montreal MEMS silicon oscillator company Why mentioned: Raised $55M Series B; their chips synchronize data transfers in GPU clusters — a deep infrastructure play in the AI compute stack.
4. People Identified
Martin Schmid CEO, EquiLibre Technologies; former DeepMind researcher Why mentioned: Led the application of poker-derived reinforcement learning AI to financial markets.
"The nice thing about trading and markets is that the scoring is super simple: how much money did the agent make?"
Cameron Sellers VP, Creandum Why mentioned: Led what Creandum described as its largest-ever single investment into EquiLibre Technologies.
"The potential total addressable market of trading in the financial markets is one of the biggest on earth, and there are countless funds over the years that have generated quantums of profit that make most venture-backed successes look small."
Marc Andreessen Co-founder, Andreessen Horowitz Why mentioned: Joined the Pentagon's Defense Policy Board, raising conflict-of-interest questions given a16z's major stakes in defense-tech companies.
"Marc Andreessen joined the Pentagon's Defense Policy Board, raising questions about potential conflicts of interest given Andreessen Horowitz's major stakes in defense-tech companies like Anduril, Skydio, Shield AI, Saronic, and Flock Safety."
Vint Cerf Chief Internet Evangelist, Google; "father of the internet" Why mentioned: Stepping down from Google, closing one of the most consequential careers in tech history.
"Vint Cerf, often called one of the 'fathers of the internet,' is stepping down next week from his role as Google's chief internet evangelist, closing out one of the most influential careers in tech history."
Roberto Serrano Economics professor, Brown University Why mentioned: Documented a mass AI-cheating incident — 40 students scored 100 on a take-home exam, average was 96; the same class averaged 48 on an in-person final.
"At least 50 students cheated with AI on a take-home mathematical economics midterm, with 40 students scoring 100 and the class averaging a whopping 96."
Tarek Mansour CEO, Kalshi Why mentioned: Was in acquisition talks with Mark Zuckerberg before Meta chose to build its own prediction market product instead.
"Mark Zuckerberg met with Kalshi CEO Tarek Mansour last year about a possible acquisition before Meta decided to build Arena, a prediction-market app using play money instead of real bets."
5. Operating Insights
Use AI tools to grow headcount — not replace it — as a competitive signal. The Ramp/Revelio Labs data suggests that aggressive AI adoption (defined at just ~$30/employee/month for 3 months) is correlated with faster hiring, particularly at junior levels. Operators who frame AI as a growth accelerator rather than a headcount reducer may attract better talent and scale faster than competitors hesitating on adoption.
Define your company as a lab, not just a revenue-generating vertical. EquiLibre's deliberate self-positioning as "a lab first, not a finance firm" — even while trading billions in daily volume — enables them to attract top research talent and build transferable technology. Operators in AI-adjacent fields should consider whether a research identity could yield better long-term defensibility than pure commercial focus.
Build for AI search visibility now, before incumbents lock it up. The emergence of Pie (raising $19.5M to help SMBs appear in AI search results) signals that AI-native distribution is becoming a real go-to-market moat. Operators should audit their discoverability in LLM outputs — the window to establish presence before AI search consolidates may be short.
6. Overlooked Insights
Amazon dropped a $40M film about Sam Altman after investing $50B in OpenAI — revealing how financial exposure shapes editorial decisions. Neon acquired the film Artificial after Amazon abandoned it. The implication is that a $50B strategic investment effectively created a content conflict of interest at a major media distributor — a preview of how AI investment concentration could have downstream effects on journalism, entertainment, and information ecosystems.
"Neon…has purchased Artificial, Luca Guadagnino's $40 million film about Sam Altman's firing and rehiring at OpenAI, after Amazon dropped the project following its $50 billion investment in the company."
MEMS silicon oscillators for GPU clusters are an under-watched AI infrastructure play. Stathera's $55M raise for chips that synchronize data transfers in GPU clusters and network switches is a deep picks-and-shovels bet. As GPU clusters scale, timing and synchronization hardware becomes a critical — and easily overlooked — bottleneck in AI infrastructure.
"Stathera develops MEMS silicon oscillators that replace quartz crystals in electronics and synchronize data transfers in GPU clusters and network switches."