Pro rata goes to Delaware, Nvidia circles Reflection, a record 42 decacorns, and why the B2B apocalypse never ended for anyone under $1B in ARR
A nuclear startup took its earliest institutional backer to court over a spring-back clause. Meta put its first model at the frontier. A 20-person team raised $870 million less than a month after launching. Anthropic cut Haiku pricing by 90%. And Bending Spoons started the Airtable layoffs that were obvious from the day the deal was announced in August. Harry Stebbings, Rory O’Driscoll and I covered all of it this week.
Pro rata goes to Delaware, Nvidia circles Reflection, a record 42 decacorns, and why the B2B apocalypse never ended for anyone under $1B in ARR
Meta is back at the frontier, a 20-person team raised $870 million less than a month after launch, and Anthropic made its small model 90% cheaper. A nuclear startup also sued its own early investor, and Bending Spoons started cutting at Airtable. Harry Stebbings, Rory O’Driscoll and I went through all of it.
#1. Valar Atomics Sued Day One Ventures Over Pro Rata in Its $1B Series B
Valar is raising roughly $1B at about $6B, led by Sequoia. CEO Isaiah Taylor says Valar asked major investors to take zero allocation so a spring-back pro rata clause wouldn’t trigger, and everyone agreed except Day One, which put $1M into the seed and $15M into the A and owns under 5%. Day One GP Masha Bucher says it offered to give up most of its allocation, and Valar demanded it give up its rights in every future round. Valar then sued in Delaware for a declaratory judgment. Day One did not sue Valar, despite what X said.
Jason’s read: Everyone’s a little wrong. The big funds put “everyone waives” in every term sheet, and letting one existing investor in as a “strategic” was going to annoy Day One. But once the founder decides, you raise it once and let it go. It’s not your company and it’s not your cap table.
I learned this the hard way. I was the only investor in a company for years, then got cut to zero in a round where Harry got an allocation. I said something, and it bent the relationship with the founder. Harry didn’t know until this episode.
Rory’s read: Ask what you’d actually win. Day One’s 5% is already worth around $300M. They’re fighting to put maybe $10M to $15M more in at $6B. “Both people can be intellectually wrong but pragmatically right.”
Harry’s read: The tide is turning on pro rata. Neo says it stopped taking pro rata rights this year and cited this exact dispute. The pushback: Neo doesn’t need follow-on, and the megafunds asking seed investors to waive never waive their own.
#2. Meta Shipped Muse Wonder, Its First Frontier Model
Meta’s Superintelligence Lab released Muse Wonder and Muse Wonder Full, scoring comparably to Fable 5.1 and GPT-6 Astra on Meta’s own tests. Wonder costs $4/$20 per million tokens on the API. Wonder Full, the one that earned the frontier scores, is only available inside Muse Code.
Jason’s read: I can’t put the headline model in my own app, so for builders it’s mostly a performative eval. The bigger story is the Muse consumer app. There’s a real chance Meta has a monster AI product on its hands.
Rory’s read: Meta was at the front on open weights, faded, and is back near the front with closed models. “If we’re going to spend $100 billion, we at least be able to charge for the damn thing.”
Jason on the people: Zuck’s had a lonely stretch. Finding a true right hand is hard. Alexandr Wang may turn out to be the WhatsApp of acqui-hires.
#3. Nvidia Is in Talks to Buy or Deepen Its Stake in Reflection AI
The FT reported Nvidia is in early talks to acquire Reflection or invest more, possibly as an acqui-hire to avoid regulatory review. Nvidia has already invested $800M, and Reflection last raised at a $25B pre-money. Harry put the rumored price around $30B. Nvidia agreed to buy Hugging Face for $12.9B in September.
Rory’s read: Nvidia wants more of the AI dollar going to chips and less to model companies, so it loves open models. “We’re the best of friends… but we’re watching you closely.” On Reflection’s board, he’d take the bid.
Jason’s read: I’ve been on the other side of this. Saying no is scary when Nvidia tells you, smiling, it’ll fund your competitor instead. Give it 18 months and Nvidia’s open model could be the default.
#4. Jev Maker TypeSafe Raised $870M at $7.5B, 24 Days After Launch
a16z led an $870M Series A at $7.5B, with Sequoia and DCVC. Jev launched September 15. It returns calibrated decisions instead of text, the company has about 20 employees, and it says a third of the Fortune 500 use it.
I did a parody of the megafund investment committee: buy every share, block the pro rata, mark it to $15B, tell the LPs we’re geniuses. Rory: “accurate as hell.”
Jason’s actual read: I use Jev in production. It’s fast, cheap and right maybe 70% to 80% of the time on carefully prompted decisions. If a team doing nothing else gets it to very high accuracy, it wins, and I doubt Cloudflare or Amazon keep investing in a niche line item. Cautiously bullish.
Rory’s read: $200B of frontier inference, 20% decision workloads, prices down 80%, Jev takes half. That’s a $4B revenue company worth about $50B, and he thinks each of those assumptions is reasonable. Competition is the risk.
Harry’s read: Every deal at these prices ends with “worst case, a great acquirer buys it.” I hear it every single time. Not everything gets bought.
#5. A Record 42 Decacorns So Far in 2026
PitchBook counts 42 new decacorns globally in 2026, with three months left. The previous record was 34 in 2021.
Rory’s read: Public markets have produced about 20 tech decacorns a decade for five decades. This year produced 42 in ten months. Either this is the most transformative event since the Industrial Revolution, “or we’re getting a wee bit ahead of ourselves.”
Jason’s read: We’re all betting it’s real. A big fund deploying in 12 months needs four $25B-plus exits, and there may only be about 70 of those ever. Andreessen was mocked for overpaying for winners and was right.
#6. Isomorphic Labs Is in Talks at a $40B to $50B Valuation
The DeepMind drug-discovery spinout is in early talks at $40B or more, five months after a $2.1B Series B led by Thrive.
Jason’s read: Generational talent gets generational pricing. I only recently bought AI drug discovery. GLP-1 data changed my mind. Getting the next one out five years faster is a bigger market than inference.
Rory’s read: This is AI that makes money from discovery, not replacing labor. You want royalties on a $200B drug. The hard part is capturing value before phase one, two and three.
#7. Mecka Raised a $60M Series B From Sequoia for Robotics Training Data
Mecka pays people to wear sensors doing everyday tasks and sells the data to humanoid trainers. It reported $100M-plus in run-rate revenue in June and expects $300M by year end, with 40 to 60 people.
Rory’s read: Same pattern as LLM data vendors, with one gap: the robots aren’t deployed yet. One logistics company counted 42 humanoid PoCs and zero deployments.
Jason’s read: I’d rather own this than any robot maker. One founder I met did $30M in 60 days in niche training data and is turning away $270M of demand.
Rory’s pushback: That’s like picking Scale over Anthropic. If you can pick the winner, pick the winner.
#8. Anthropic Cut Haiku Pricing by 90% With Haiku 5.5
Haiku 5.5 costs $0.10/$0.50 per million tokens for prompts under 100K, down from $1/$5. Anthropic says 90% of prior Haiku requests fall in that tier.
Jason’s read: This is the first thing in AI that’s actually been deflationary for me. Haiku 5.5 is nearly Sonnet for a lot of use cases at a tenth the price. Sonnet 5.5 is twice as fast in my testing. Evals don’t talk enough about speed. No more excuses at board meetings.
Rory’s read: Cut price 90% and “we’ll use more” doesn’t cut it. You need 10x more usage or revenue falls.
I think we’ll get the 10x. You stop sampling and run it on every row of the database.
#9. Bending Spoons Is Laying Off a “Large Part” of Airtable
Bending Spoons bought Airtable in August for $1.285B in enterprise value. This week it said a large part of the team is being laid off. Evernote and Vimeo went the same way.
Jason’s read: Nobody talked about the jobs when the deal was announced. If you sell to Bending Spoons, you know your team is gone. I wouldn’t do it with a cash-flow positive company and hundreds of millions in the bank.
Rory’s read: Legally, the board’s job is shareholder value. And sometimes resolution is better for employees than a two-year death march. Take the deal, but negotiate employee terms up front.
Jason on Cloudflare: Cloudflare was crushing it and still cut 20% in a day because it was too slow. It reaccelerated. Netflix at 5% and HubSpot at 7% are too small.
#10. Bumble Is Going Private for $1.3B
Ithaca Holdings, led by Ravi Gupta and John Hegeman, is paying $6.05 a share, a 138% premium. Bumble was worth about $14B at its 2021 IPO.
Rory’s read: Half of couples meet on apps, and the category is worth a few billion. VCs have always hated the churn.
Jason’s read: Having built an AI matching product, I see a bull case. Use the data to get people five great dates instead of 20 bad ones, then charge 5x more.
Harry’s read: AI avatars that date for you. Rory: “a wonderfully antisocial way to date.”
#11. The B2B Apocalypse Only Ended for Companies Over $1B in Revenue
Software stocks bottomed in March and April. By Jason’s count, the Nasdaq is up about 22% this year, his SaaS AI index about 28%, and the WisdomTree cloud index about 25%.
Jason’s read: It only passed for companies with massive scale. Everyone else is worse off than in March, because no one wants to buy software. A $500M ARR IPO may not be safe anymore. You might need $1.5B to $2B.
Harry’s read: Pre-AI companies can still get a good outcome if they go AI-first. 20VC portfolio company TechWolf, an HR software company that rebuilt itself around AI, just agreed to sell to SAP in what it says is the largest venture-backed software acquisition in Belgian history.
Rory’s read: It’s a mature, consolidating industry now. There are 200 to 300 midsize companies and big tech buys maybe 10 a year. Most will stay midsize.
Quotable Moments
Jason Lemkin
“You just got to get in line and move on. It’s not your company. It is not your cap table.”
“These are really the first thing that’s actually been deflationary with AI. No matter how much we talk about token prices falling, we’ve been consuming so many more tokens, we’re not getting the benefit of it.”
“Everyone without massive scale is living a worse apocalypse than in March or April, because no one wants to buy software. Everyone’s breathing a sigh of relief. I don’t think we should be.”
Rory O’Driscoll
“Both people can be intellectually wrong but pragmatically right.”
“Either this is the most economically transformative event since the Industrial Revolution, or we’re getting a wee bit ahead of ourselves.”
“When you cut prices by 90%, ‘we’re going to use more’ doesn’t cut it. We’ve got to use 10 times more.”
Harry Stebbings
“I love the salaciousness of this market. I remember when it was quite boring.”
“What worries me in every justification for a lot of these prices is, ‘the downside protection is that it’ll just get bought by a great acquirer.’ I just hear this every single time.”
“AI avatars that represent you, and then you can go out and meet a hundred different people without actually spending any time meeting anyone.”
