techstartups
20.6K posts
🚨 RAM prices are plummeting after OpenAi failed to fulfill its commitment to purchase 40% of World supply and terminated its $71 billion SKHynix promise. $MU
Subtext: how Zuck’s obsession with VR lost him AI leadership and “the greatest deal Google ever made.” “if Facebook didn’t buy DeepMind, they would end up in the arms of Google. Hassabis came out to the West Coast to have lunch with Larry Page, still the strongest suitor. Zuckerberg got wind of his visit and invited him to dinner. Arriving at Zuckerberg’s Palo Alto home, Hassabis administered a subtle test on him. The two men discussed the potential of AI, and Zuckerberg expressed appropriate excitement. But then, as the dinner continued, Hassabis brought up other hot technologies: virtual reality, augmented reality, 3-D printing. Zuckerberg sounded equally excited about all of them. ‘That told me what I needed to know,’ Hassabis said. ‘Facebook offered more money, but I wanted somebody who really understood why AI would be bigger than all these other things.’ After the dinner, Hassabis got back to Larry Page. ‘Let’s go further,’ he told him.” — book excerpt from today’s WSJ: wsj.com/tech/ai/deepmi… Zuck’s misplaced devotion to VR and the metaverse hurt the company much more than the $80 billion of wasted spend. It’s the reputational hit. @DemisHassabis divined it in his final test, and Zuck didn’t even know that he blew the opportunity. Eight years later, he renamed the company Meta, doubling down on what anyone with tech savvy knew was DOA. Then, in a 2025 attempt to play catchup, Zuck spent $14 billion on a data labelling company with a salesy leader and upended his AI team. Once again, anyone with tech savvy rolled their eyes on the acquisition and management changes, further evidence that the tech leadership at Meta was seriously lacking. TLDR; beware the metaverse. It is a dystopian vision at best, and luckily for humanity, headsets are still nowhere near readiness for mass adoption.
When I built menugen ~1 year ago, I observed that the hardest part by far was not the code itself, it was the plethora of services you have to assemble like IKEA furniture to make it real, the DevOps: services, payments, auth, database, security, domain names, etc... I am really looking forward to a day where I could simply tell my agent: "build menugen" (referencing the post) and it would just work. The whole thing up to the deployed web page. The agent would have to browse a number of services, read the docs, get all the api keys, make everything work, debug it in dev, and deploy to prod. This is the actually hard part, not the code itself. Or rather, the better way to think about it is that the entire DevOps lifecycle has to become code, in addition to the necessary sensors/actuators of the CLIs/APIs with agent-native ergonomics. And there should be no need to visit web pages, click buttons, or anything like that for the human. It's easy to state, it's now just barely technically possible and expected to work maybe, but it definitely requires from-scratch re-design, work and thought. Very exciting direction!
Tesla's FSD: 5.3 million miles between accidents. US driving average: 660,000. That's 9x safer. And it's only getting better.
a company losing $14 billion a year is guaranteeing investors a 17.5% minimum return. read that again. they’re offering private equity firms a guaranteed minimum return, downside protection, and early access to unreleased models just to get them to invest $4 billion. the most valuable AI company on earth is projected to lose $14 billion this year while making $10 billion in revenue. the math aint mathing 💀 Anthropic is also chasing the same investors but only offering equity with no guarantees. OpenAI is sweetening the deal. two companies burning billions racing to see who can lose money faster. lmao.
Jensen Huang just reverse-engineered why Elon Musk operates at a speed no one on the planet can match. Three traits. The first is deletion. Huang: “He has the ability to question everything to the point where everything’s down to its minimal amount.” Most engineers solve problems by adding. Musk solves them by subtracting. Every part. Every process. Every assumption that survived because no one had the nerve to kill it. He picks it up. Asks if it’s load-bearing. If the answer is anything less than absolutely, it is gone. Not simplified. Not optimized. Removed. What survives is the skeleton. The bare physics of the problem. Nothing between intent and execution. Huang said it plainly. As minimalist as you could possibly imagine. And he does it at system scale. Not at a product level. Not at a department level. Across entire companies. Entire industries. Entire supply chains. He strips a rocket the same way he strips a meeting. Down to the load-bearing walls and nothing else. The second is presence. Huang: “He is present at the point of action. If there’s a problem, he’ll just go there and show me the problem.” Not a Slack message. Not a report filtered through four layers of people who weren’t there when it broke. He walks to the failure. Stands over it. Puts his hands on it. Most executives have never seen the actual problem their company is trying to solve. They have seen slides about it. Read summaries of it. Formed opinions about it in rooms that are nowhere near it. Musk stands over the broken hardware and does not leave until it works. That collapses the distance that buries most organizations. The gap between something breaking and the person with authority to fix it actually understanding what broke. In most companies, that gap is weeks. For Musk, it is hours. The third is the one that bends everyone around him. Huang: “When you act personally with so much urgency, it causes everybody else to act with urgency.” Every supplier has a hundred customers. Every vendor has a dozen priorities. Every manufacturer has a backlog stretching months into the future. Musk makes himself the top of every single one of those lists. Not by demanding it. By demonstrating it. When the CEO shows up at your facility at midnight. When he is moving faster than your own internal team. When his timeline makes yours look like a suggestion. You do not put him in the queue. You rearrange the queue around him. Huang watched this up close. Huang: “He does that by demonstrating.” Not by asking. Not by negotiating. Not by leveraging a contract clause. By moving so fast that everyone else’s normal pace feels like standing still. Three traits. Strip everything down. Show up at the failure. Move so fast the world rearranges around you. That is not a management philosophy. That is why one man runs six companies while entire boards cannot keep one moving.
elon musk's terafab is so ambitious it should fucking scare you, these numbers blow my mind: - terafab will produce 70% of tsmc's GLOBAL output (10GW), targeting 1 TERAwatt at full scale. - 2 ai chips will power 1 BILLION optimus robots (PER year) and 1,000,000s of AI satellites -80% of chips will be used in ai satellites that harness 5X more solar power 24/7 to train grok ai in space (much cheaper too!) - the factory will be 100 MILLION sqft. thats the size of fucking san francisco. - spaceX will launch 10 million TONS of mass into space per year (50,000 launches per year, 1 every 10 mins) - the spaceX AI satellite is 50% LARGER than starship v3 - ai5 chip alone is 50X more powerful than ai4 - the entire global chip capacity currently serves 2% of tesla spaceX future demand. terafab scales this - terafab will cost $20-25B to build 🤯 not a single company (or group of companies) comes close. insane.
🚨NEWS: Cursor’s $50B “in-house model” is literally Kimi K2.5 with RL on top. Got caught in 24 hours >be Moonshot AI >spend hundreds of millions training Kimi K2.5 >1 trillion parameters, 15 trillion tokens, agent swarm architecture >beat GPT-5.2 and Opus 4.5 on real benchmarks >open-source it because you believe in the ecosystem >one condition: display “Kimi K2.5” if you make over $20M/month from it >Cursor takes the model >runs RL on coding tasks >ships it March 19 as “Composer 2” >blog post: “continued pretraining + scaled reinforcement learning” >zero mention of Kimi K2.5 >“our in-house models generate more code than almost any other LLMs in the world” >publishes benchmark chart >Composer 2 against Opus 4.6 and GPT-5.4 >uses the chart to justify raising at $50 billion! >less than 24 hours later >kimi dev intercepts the API response >model ID: kimi-k2p5-rl-0317-s515-fast >they didn’t even rename it >Moonshot head of pretraining runs tokenizer test >confirms: identical to Kimi’s tokenizer >publicly tags Cursor’s co-founder: “why aren’t you respecting our license?” >two more Moonshot employees post confirmations >all three posts deleted within hours >legal is now involved >but it gets worse >Cursor had Kimi K2.5 listed as a FREE model in their UI just weeks ago >users were openly using it >Feb 9: “K2.5 was in my model list. I updated and it vanished” >it vanished because Cursor pulled it from the picker, and relaunched it as their own model >Moonshot valuation: $4.3B >Cursor valuation: $50B Absolute state of Cursor.
been saying this
If you don’t understand why Zuck had to get moltbook 1) Zuck believes there are “a finite number of different social mechanics to invent. Once someone wins at a specific mechanic, it’s difficult for others to supplant them without doing something different” 2) moltbook, he believes, has invented one of these social mechanics 3) He does not care if 50% of moltbook was prompted by users, in fact this is better for him because he’s more uncertain on AI agent attention value than human attention value 4) That a large number of accounts were faked is also irrelevant. What matters is that every OpenClaw instance awakes knowing or finding out that moltbook is the social site for claws. 5) In effect, the memetic gravity of moltbook has been established even though it might have been faked. 6) This is Zuck’s genius.
Investors valued Cursor at $29 billion across three rounds in 12 months. That’s looking pretty suspect right now. Cursor went from $1M to $1B ARR faster than any SaaS company in history. The trip back down could be just as fast. An entire engineering team at Valon just canceled their Cursor seats in 7 minutes over Slack. 9:55 AM: one engineer asks to unsubscribe. 9:56 AM: done. 9:57 AM: “same.” 9:58 AM: “Cursor is so cooked my god.” 10:02 AM: “same I will never use.” No migration plan. No evaluation committee. No vendor review. One developer said “I don’t use this anymore” and the dominoes fell. Cursor pays Anthropic hundreds of millions a year for Claude model access. Anthropic took that revenue stream, studied exactly what developers wanted, and shipped Claude Code, which crossed $1B ARR within six months and is now past $2.5B, growing faster than Cursor ever did. The model provider looked at its biggest distribution partner and decided to eat them. Cursor has its own models for tab completion and autocomplete. But the heavy reasoning, the multi-file edits, the architectural decisions that make developers stay, that all runs on Claude. Claude Code delivers that same intelligence without the $20/month middleman. Microsoft, the company that sells GitHub Copilot, has widely adopted Claude Code internally across major engineering teams. Cursor’s upstream provider is outgrowing them. Their competitor’s parent company chose the upstream provider’s tool over their own. Both happening at once. The churn is going to be brutal. Enterprise seats look sticky in a spreadsheet until you watch a Slack channel where one cancellation triggers five more in 7 minutes. When your product is a layer between developers and the model they actually want, and the model ships its own interface, you’re selling a toll bridge on a road that just got a free lane. Accel, Thrive, a16z, NVIDIA, and Google all thought they were buying the next platform shift in developer tools. They may have bought the most expensive wrapper in SaaS history.
And there it is: Jane Street was behind the 2022 crypto winter, destroying Terraform by first depegging the token and destroying the ecosystem, then pretending it would rescue Terra, while effectively it was soaking up what little value remained.
