Save your energy for that professional meaningful connections and your emotion for that great date, on BIMP.
Join our private launch here with your email only.
https://t.co/mCZLxqiV3s
.@asenkut on the duct-taped servers that let early @Google outscale everyone:
"Larry designed these crazy servers at 10 PM one night.
We all thought, 'We have all this money. Why don't we go buy servers?'
But Larry said, 'Why would I buy servers? It's a huge waste of money. They're inefficient. They're very expensive. They're not designed to do what I want.
I'm gonna buy these cheap Intel CPUs and a crap ton of disk. We're gonna literally duct tape it.'
He said, 'Sanjay is gonna run software.' Sanjay, who just started a new company with Jeff Dean, basically wrote the precursor of AWS. Just like S3. He built it inside Google 10, 15 years before Amazon did it.
We used such cheap chips and disks that it failed all the time. It pissed people off.
There was literally someone whose job was to write babysitter software that knew when things were failing.
Then when it failed, someone would literally run to the data center to plug in and replace 10 new disks.
That's early Google. Nobody rational would ever do that.
But Google scaled way faster and way more economically than people imagined because they did things like this."
Letting internal teams build their own tools is becoming de rigueur at AI-native companies like @figma. This is a preview of what F1000 companies will routinely be doing in a couple of years.
Teams building their own tools will be one of the biggest headwinds to AI-native companies trying to sell workflows to enterprises. Systems of record should be more insulated, because teams want to build their workflows on top of a robust foundation.
Get more such insights weekly in Signalful (@signalfulnews) Best of the Week. Link below to get it delivered to your inbox every Friday.
Big week on 20VC with @HarryStebbings and @rodriscoll.
Sequoia closed London and killed Scouts and Arc, Jensen resigned the day after we recorded, Anthropic's IPO book is reportedly 6x oversubscribed, and agents kept eating seat-based B2B.
My takes:
#1. The biggest Sequoia news wasn't London. It was winding down the evergreen fund.
Holding generational assets for decades was the best idea in venture. It launched right before stocks fell 30%, and LPs don't want to be forced to hold.
#2. Sequoia can cover Europe from the Bay above pre-seed
The best European founders are already here every month. Mati from ElevenLabs is at every dinner I go to. Pre-seed still needs someone local doing the hands-on work.
#3. On Tuesday I said Jensen needed a pass. Rory said you can hope for a pass but can't assume one. Rory was right.
He resigned Wednesday, and the CFO is now interim CEO. China comes out the biggest loser here. Selling GPUs there, or parking them just over the border, just got a lot harder.
#4. The AI trade doesn't crack on supply
Not on data centers, not on AMD taking share. It cracks only if demand cracks, and you'll see that in OpenAI and Anthropic ARR first. The $2.5T NASDAQ drop was hedge fund pods getting stopped out.
#5. I called Anthropic's IPO fatigued. A 6x book says otherwise
The $70 to $78 range isn't the price. It could still go up before pricing on Nov 18. Either way, you'd rather be Anthropic than OpenAI right now: public, with the float and roughly $200B in cash.
#6. OpenAI reaccelerated to 70% annualized revenue growth in Q3, and enterprise is now the majority of revenue
It's a two-horse race, and the next question is how price-sensitive token demand is. If you cut price 20% and usage goes up 21%, revenue grows. If usage goes up 15%, it shrinks.
#7. IC this week was Zipline at $21.5B, fal at $16B and Revolut at $115B. I'd do all three.
If I have to pick, it's fal. Sometimes in venture you have to be a tourist and ride the wave. Harry and Rory both took Revolut, which is set up to become Europe's biggest bank.
#8. Stolen AI API keys are the crime of the yea
Bot swarms are hitting everyone, and my own Claude limit had defaulted to $100K. It's better than stealing Rolexes off wrists in London. And customers can't leave their inference provider, because there's nowhere else with the capacity.
#9. Dialpad buying Ada for about $430M in stock isn't an exit
What depresses me more is that so many pre-AI companies still ship at 2021 speed. If your 2027 plan is "build an AI-native team next year," that was the 2025 plan.
#10. Cloudflare grew 39% on agents while Atlassian fell on seat growth
Agents don't care about your UX. If it's not an agentic beneficiary, I don't want to hear about it. Figma below $10B is the cautionary tale: turn down $20B, execute brilliantly for three years, and you can still end up worth half.
Rory's best line: "You could work for the next three years, do an amazing job, grow your damn company 40% a year, get the damn thing public. And at the end of that, you're going to be worth half what you are today. No one ever said capitalism was fair or easy."
https://t.co/ebh8FmvJ8H
Skyfall launched Steve, an AI "CEO" running 3 software businesses they acquired. Now they're putting $2M toward buying 5 more startups for it to run.
I love this as an experiment. P&L as the scoreboard is about as honest an eval as it gets.
My question as an investor: can an agent do the messy part? Getting the first customers, reading the room on a call, knowing when a customer is lying to be nice haha.
Maybe I'm wrong. Show me the numbers in 12 months and I'll happily eat my words :)