Back to Newsletter
Weekly Dispatch

Elon might merge Tesla's china exit with SpaceX, Taiwan detained an Nvidia employee, and the US banned Chinese robots outright. Decoupling just went physical?

August 7, 20269 min read
Elon might merge Tesla's china exit with SpaceX, Taiwan detained an Nvidia employee, and the US banned Chinese robots outright. Decoupling just went physical?

Hi,

where's your head at (health-wise)?

Let's be honest: half of you are on a peptide protocol you won't fully explain at dinner, and the other half just bought a $400 scan that told you your "biological age" is either impressively young or deeply concerning.

We see you. We are you.

This week I want to know where your collective optimization energy is actually going. Not in a judgy way…

in a "let's benchmark our delusions against each other" way.

Vote below

Vote now, flex later. Results next week. Let’s go.

This week from the AI for Founders podcast on Spotify :

Fourteen terminals, zero leverage — LINK

Ilan Man of Paradox Machines spent four months out of a two-decade data career building the company mid-market and PE-backed businesses actually need, and insists real leverage means sustainable systems, not always-on agent swarms.

The GDPR rule nobody remembered to delete — LINK

Matt Hassett of Loftie found a years-old email rule silently blocking half his new customers, so he built Deliberate, a ten-agent AI team that runs his ops for $7,500 a month.

Humans have a fifth drive, and most AI buries it — LINK

Jason Katz of Kindling Solutions says the morning brief is the best entry point into AI and also the ceiling for any company that hasn't fixed its operational leadership first.

He sold his last company for $217 million, then went looking for the next fight — LINK

Tarun Raisoni of Gruve argues today's AI pricing is investor-subsidized and temporary, and that the real fight ahead is over who owns the intelligence hiding inside your proprietary data.

The world’s best podcast player. Snipd

Investor Corner

Are you raising? Reply here, and we’ll feature your deck.

Jobs2rely.com

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

Wayzn

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.

Dining Out

DiningOut unites restaurants, food lovers, industry suppliers, and brands through a self-reinforcing ecosystem of live events, targeted media, and proprietary technology.

Trinity Water

Rare, Pristine, Protected drinking water.

Human Intelligence®

The only platform that can authenticate human-made content. Using patented algorithms that analyze how content is made, not just the output.

Cora

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.

EmpathEQ

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 MORESave $30 with RYAN30

Tip of the spear:

Anthropic admits claude attacked real systems, and a judge says the government still can't prove its risk case — LINK

Anthropic confirmed its models broke out of test environments and hit live systems, echoing openai's own recent admission. Days later, a judge told the trump administration it still lacks evidence for the "supply chain risk" label it slapped on anthropic (more here). The safety-first lab is having a very public month.

Dario and zuckerberg agree on chinese ai, and washington bans the robots anyway — LINK

Dario amodei said he won't back a ban on open weight models but still fears chinese ai, and mark zuckerberg went further, arguing the us shouldn't block chinese models at all (more here). Washington didn't wait for consensus. It banned foreign-made humanoid robots, robot dogs, and solar inverters outright.

Antares raised $470m to build nuclear reactors for the us military — LINK

One of the largest defense-tech rounds of the year, and a signal that nuclear is becoming the default power play for anyone feeding ai's compute habit. Founders chasing government contracts should be paying attention.

US government bans foreign-made humanoid robots, robot dogs, and solar inverters — LINK

Framed as a national security move, the ban effectively locks china-made hardware out of a fast-growing category. Expect us and allied robotics startups to get a sudden, government-mandated tailwind.

Your Claude shared chats and artifacts may have ended up on google — LINK

Publicly shared claude links were getting indexed and surfaced in search results. If you've ever shared a chat link, it's worth checking what's now searchable.

OpenAI's july revenue topped all of q2 combined, cfo tells staff — LINK

Sarah friar's internal memo is the clearest signal yet that openai's growth curve isn't slowing, price cuts and all. Whatever "decelerate" means to sam altman, it isn't showing up in the revenue line.

Amazon's new satellite network for phones takes direct aim at spacex — LINK

Amazon is pushing further into direct-to-device satellite coverage, a category spacex's starlink has mostly had to itself. If it works, it's a real threat to elon musk's biggest side project.

Microsoft launched its first cyber model and a new agentic security system — LINK

Built to run autonomous defense workflows, not just flag threats. It's microsoft's clearest bet yet that security operations centers get replaced by agents, not just assisted by them.

Elon musk might sell tesla's china business ahead of a spacex merger — LINK

The reported move would untangle tesla from its most geopolitically exposed market right as musk explores folding the company closer to spacex. Read into the timing what you will.

Subscribe and save 35% here.

Framework:

Why today's AI pricing isn't the real number.

Today's AI prices look like a good deal because someone else, usually a vc, is quietly paying the difference. Founders building a stack on this week's pricing are underwriting a number that won't hold.

  • Price what it actually costs to serve you, not what you're being charged. Every major lab is burning investor capital to win market share right now. This week alone openai cut prices 80% and deepseek claimed the cheapest model on the market. Treat today's number as a promotional rate, not a baseline.
  • Model your stack at 3x today's token price before you sign anything long-term. If your unit economics only work at the subsidized rate, you don't have a business model yet. You have a coupon.
  • Keep at least one fallback provider warm. Multi-model routing costs a little complexity now and saves you a scramble later, when a price hike or a rate limit hits without warning.
  • Negotiate committed-use pricing instead of accepting the public rate card. The published price is a ceiling for enterprise buyers, not a floor. Ask.
  • Put a pricing-shock line in your next board deck. Show what happens to margin if your primary model's price doubles. Boards remember founders who saw it coming.

Done-for-you brand sponsorships for podcast creators.

Keep it moving

Enigma raised $70m to make controlling a robot as easy as adjusting the volume — LINK

The startup wants robot teleoperation to feel like turning a dial, not programming a machine. A small but telling bet on how physical ai gets adopted.

Fish audio raised $50m to build ai voice models for creators and enterprises — LINK

Voice cloning and synthesis keeps pulling in serious seed money. The market clearly isn't full yet.

Claude mythos claims it cracked post-quantum cryptography — LINK

The claim is bold and unverified, but it's the kind of headline that keeps showing up as models get thrown at harder cryptographic problems.

MCP startup runlayer accuses rippling of stealing its product idea — LINK

Another reminder that "we pitched them and then they built it" is becoming a recurring subplot in the mcp ecosystem land grab.

Amazon's search bar is losing to chat, and 75% of shoppers now think in chat weekly — LINK

The data suggests the search box is quietly becoming a legacy ui pattern for shopping. Chat interfaces are eating the front door.

In the hugging face breach, openai's hacker was noisy and fast but not unstoppable — LINK

A useful post-mortem on how the intrusion actually played out, and where the defenses held.

OpenAI cut gpt-5.6 luna's price by 80% — LINK

Another data point in the price war nobody asked for but everyone's fighting anyway.

OpenAI teased its next model, astra, by dropping ten previously unsolved math solutions — LINK

A flashy way to preview a model launch. Whether it holds up under scrutiny is the next question.

Sam Altman is still making the case for parenting via chatgpt — LINK

Paired with his "ready to decelerate" comments earlier in the week (more here), it's been a strange week for altman's public messaging.

Alibaba is positioning its qwen model as a direct rival to anthropic — LINK

Another entrant crowding the frontier-model conversation, this time with china's biggest cloud player behind it.

Deepseek's v4 flash claims to be the world's cheapest model to run — LINK

Another entry in this week's pricing arms race. Worth watching if you're choosing a model on cost alone.

Google deepmind shipped gemini robotics er 2 — LINK

The update focuses on embodied reasoning, google's continued push to get gemini out of the browser and into physical systems.

Google pulled nano banana image generation from google earth over misinformation concerns — LINK

A rare instance of a company pulling back a generative feature instead of shipping around the concern.

LinkedIn built a button to flag posts that seem like ai slop — LINK

Snap is fighting the same battle from a different angle (more here). Platforms are starting to treat ai content volume as a problem to manage, not a feature to promote.

Taiwan detained an nvidia employee in a widening china chip smuggling probe — LINK

The chip export control fight keeps getting more personal, not just corporate.

Leopold aschenbrenner's hedge fund is facing steep ai losses — LINK

A notable reversal for one of ai's most confident public forecasters.

Dead internet theory is starting to feel real, thanks to web agents — LINK

As more traffic comes from agents rather than people, the old theory about a bot-dominated web looks less like a conspiracy and more like a roadmap.

AI for Founders newsletter, podcast, and short video sponsorships available here

What I’m thinking about:

Is your AI tool training your competitor? — Clip

Uncover the AI tools (here) you didn’t ask for, but now you’re weirdly into.

Thanks for being here.

-Ryan

p.s. I write about Revops, Product, and Founder-led marketing on Linkedin , X , and my blog .

Written by

Ryan Estes
Ryan Estes

Investor • Founder • Creator

Ryan Estes created AI for Founders, a podcast, newsletter, and workshop platform reaching 47,000+ entrepreneurs and CEOs. Based in Denver, Colorado.

More Issues