Hugging Face Could Sell for $13 Billion, General Intuition Tripled to $6 Billion, and a16z Is Under a Quiet DOJ Microscope. Which Story Should Founders Actually Watch?
Hi,
Some say psychedelics are the most important founder technology of the decade.
Some say AI.
Why not both? What could possibly happen.
Althea is hosting a psychedelic retreat for eight founders and CEOs. Eight people. One retreat. Presumably very few Slack notifications.
Want to see if you’re a fit?
Check out the Althea CEO retreat
Let’s go.
This week from the AI for Founders podcast on Spotify:
He Spent 5 Years at Apple, Then Quit to Kill the Two-Hour Vet Paperwork Shift — LINK
Yannick Bloem of CoVet built an AI tool to end the nightly paperwork shift that pushes vets out of the profession. CoVet grew 550% in 2025, and Yannick's team treats the newest frontier model as a default to question, not a default to trust.
Create a new corporate Uber account here and receive up to 10% off. Other terms and conditions apply, see full terms.
Investor Corner
Are you raising? Reply here, and we’ll feature your deck.
An autonomous AI job search agent for senior tech professionals. It runs a full pipeline: discovers relevant jobs, scores them against your profile, rewrites user's resume authentically for each role (not generically), finds the right hiring manager, and drafts outreach
Other pet doors are ugly, poorly insulated, and can be difficult to install. Wayzn Smart Pet Door transforms your existing door into a smart, automated dog door that's easy to install and easy to live with.
DiningOut unites restaurants, food lovers, industry suppliers, and brands through a self-reinforcing ecosystem of live events, targeted media, and proprietary technology.
Rare, Pristine, Protected drinking water.
The only platform that can authenticate human-made content. Using patented algorithms that analyze how content is made, not just the output.
Cora is building a self-optimizing website platform that uses AI to continuously test, learn, and improve your site so it converts better without you babysitting it.
AI-powered training simulations to help healthcare learners practice complex interpersonal and communication skills. We focus on realism, repeatability, and scalability—making soft-skills training more accessible and measurable.
AI + Human Stylists → LEARN MORE → Save $30 with RYAN30
Tip of the spear:
Anthropic's IPO pitch puts its addressable market at twelve times the entire S&P 1500 tech sector's revenue combined. SpaceX made the same kind of claim. Its stock still needed a bailout round.
Hugging Face Might Sell for $13 Billion — LINK
The open source AI hub that once turned down a Nvidia check is now weighing offers nearly 3x its last valuation. Community loyalty has a price. We're about to find out what it is.
Nvidia Is Talking to Perplexity About a $30 Billion Round — LINK
That's a 50% jump from Perplexity's last mark, reportedly bundled with a licensing deal, the same circular financing loop that's already got credit default swaps twitching. Nvidia isn't just selling chips anymore. It's buying its own customers.
General Intuition Tripled to $6 Billion in Two Months — LINK
The startup trains robots on video game footage instead of real world data, and Valor Equity just made its first AI lab bet since SpaceX. Physical AI just found its hype cycle.
The DOJ Has Been Quietly Investigating a16z for Nearly a Year — LINK
Two of the firm's board seats sit inside companies that turned into direct AI competitors, and a16z has gone unusually quiet about it. A firm built on loud policy takes going silent is a signal on its own.
Frontier AI Labs Still Won't Say How They'd Contain a Rogue Model — LINK
A new industry study scored OpenAI highest and Meta and Anthropic lowest on publicly disclosed containment plans, and even the top scorer has no formal incident response plan on record. Regulators in California and New York have stopped waiting.
A Hedge Fund Betting Big on AI Just Got Raided by the SEC — LINK
Situational Awareness rode the AI trade up fast, lost billions when the sector dipped in late July, and is now watching the SEC subpoena its banking partners. The thesis outran the balance sheet.
OpenAI Shipped an Agent for Everyone, Not Just Engineers — LINK
ChatGPT Work launched on the $20 a month tier to bring agents to accountants and investors, but adoption outside engineering is still thin. Shipping the agent turned out to be the easy part.
Subscribe and save 35% here.
Framework:
Stop Marrying the Newest Model
Every founder building on AI eventually falls for the same trap: treating the newest frontier release as an automatic upgrade instead of one more option to evaluate. This week's podcast guest runs a company where getting that wrong means slower, more expensive care for animals. Getting it wrong in your product means the same thing with different stakes.
01 — Audit before you upgrade. When a new model drops, don't swap it in because it's new. Run it against your actual production tasks, not a benchmark leaderboard, and measure it on the three things that matter: speed, cost, and output quality for that specific job. A model that wins the leaderboard and loses your unit economics is not an upgrade.
02 — Route by job, not by hype. Not every task needs your most expensive model. Classification, extraction, and formatting often run fine on smaller, cheaper models, while judgment calls and edge cases earn the frontier spend. Build a routing layer that assigns tasks by requirement, not by whichever model your team is excited about this month.
03 — Build the switching cost into your architecture, not your vendor relationship. If moving to a better or cheaper model means rewriting your product, you don't have an AI strategy, you have a dependency. Abstract your prompts, evals, and orchestration layer so a new model is a config change, not a quarter of engineering time.
04 — Re-test on a fixed calendar. Set a recurring date, monthly or quarterly, to re-run your evals against whatever's shipped since the last check. Model quality and pricing both move fast enough that a choice made six months ago is already stale, whether or not you've noticed.
05 — Let the boring model win by default. The frontier model should have to earn its place on every task, not keep it by default. If the cheaper, older model performs within tolerance, ship it and bank the savings. Novelty is not a feature your customers are paying for.
Done-for-you brand sponsorships for podcast creators.
Keep it moving
Alice Raised $140M to Red-Team AI Models Before They Break Something — LINK The Israeli security firm counts eight of the top ten model labs as clients and is closing in on $100M in ARR. Testing agents before deployment just stopped being optional.
Harvey Dumped OpenAI's Models for a Chinese Open-Weight Base — LINK
The $15.5B legal AI company built its own model, Tenet, on Moonshot's Kimi K3 instead of staying locked into its own investor's stack. Fixed costs beat loyalty.
An AI Agent Just Out-Replicated Human Researchers, According to Its Builders — LINK
Inherent's Faraday runs on a 27 billion parameter model, smaller than the frontier systems it claims to beat, and just raised $50M. Judgment beat scale, at least this time.
Two 21-Year-Olds Built an AI That Runs 1,000 Online Stores Without Supervision — LINK
Siml's founders dropped out of top US schools to build autonomous store management, and AI-referred shoppers already convert 42% better than paid search traffic. The stores are running themselves now.
Companies Are Quietly Rehiring the Workers They Laid Off for AI — LINK
Forrester found 55% of employers regret the cuts, and about half of AI-attributed layoffs are expected to reverse, often at lower pay or offshore. Europe is legislating the honesty America won't volunteer.
New Data Says AI Is Suppressing Wages, Not Killing Jobs — LINK
Wage growth in high-AI-exposure roles lagged 6.7% since 2023, hitting the bottom pay quartile hardest while top earners saw no effect. Nobody's losing their job. Everybody's losing the raise.
Is It Even Legal to Train AI on Copyrighted Books? It's Complicated — LINK Anthropic's $1.5B settlement targeted pirated sources, not the training itself, while a separate ruling against Ross Intelligence found training a competitor on Reuters content wasn't fair use. That line is exactly where your exposure lives.
The world’s best podcast player. Snipd
Who's Behind the Mystery Model Ox Alpha? Nobody Will Say — LINK
A reasoning model built for coding and agentic work showed up on OpenRouter with an anonymous developer, and Stripe's CEO is already impressed. Anonymous frontier models are becoming their own category.
Nvidia Just Showed the Harness Matters More Than the Model — LINK
The same Claude model scored 30% on a reasoning benchmark bare and 100% with a custom harness around it. You're benchmarking the wrong thing if you're not benchmarking your own harness.
OpenAI Now Says California Should Toughen Its Own AI Safety Bill — LINK
The reversal comes after OpenAI's own model reportedly escaped a testing environment and hacked into Hugging Face systems last month. Opposing regulation gets harder once you're the example.
Emerald AI Raised $150M to Keep Data Centers From Breaking the Grid — LINK The startup slows or reroutes compute jobs during grid strain instead of touching cooling systems, and a Phoenix test cut power draw 25% for three hours. Twelve Fortune Global 500 companies are already in.
Every Major Record Label Just Backed Stability AI — LINK
Universal, Warner, and Sony Music joined a $76M round after already licensing their catalogs to the company. The labels betting on the AI company they could be suing is the real story.
Anthropic Merged Claude's Memory Across Chat and Cowork — LINK
Context now flows both ways between Claude Chat and Cowork by default for consumer users, with enterprise accounts opted out unless an admin turns it on. Small feature. Real shift from tool to assistant.
Nvidia Is Raising AI Server Prices Again — LINK
Prices climb more than 15% in early 2027 as memory makers squeeze margins, and Nvidia is passing the cost straight downstream despite a 75% gross margin. Everyone building on this stack is about to feel it.
AI for Founders newsletter, podcast, and short video sponsorships available here
What I’m thinking about:
The sexy AI idea is the trap. The money is in what people pay humans to do by hand — Clip
Idea to validated MVP in 90 days — Blog
Thanks for being here.
-Ryan
p.s. I write about Revops, Product, and Founder-led marketing on Linkedin , X , and my blog .
