20VC: Cognition vs Factory: Vinod Khosla Creates a Storm | OpenAI Nears $70B Run Rate: Anthropic Under Threat | ElevenLabs Doubles Its Valuation to $22B & Salesforce Buys Listen Labs for $2B

OpenAI

Across the shows: 13 speakers in other episodes, 4 bullish and 15 bearish calls, past 30 days

• Jason Lemkin · SaaStrBullish

  OpenAI has rebounded to be competitive, with teams switching to its models for better and cheaper output.

  Jason argues OpenAI was counted out over the summer, when consumer was seen as the wrong path, but its team has now built models that are truly competitive. He cites the $70B run rate, the pace of the turnaround, and the fact that businesses are switching their models for real because the new models are better and cheaper. He also points to Astra, Sol and Loon being used across his portfolio companies.

  “the OpenAI team got back in the game and made something that is truly, epically competitive to the point where people are switching out their models for real because it's better and cheaper.”at 5:15

• Dev Ittycheria · MongoDBBullish

  ChatGPT gives OpenAI huge distribution, but enterprise contracts and workflows may be its durable moat.

  Dev says OpenAI has shown that a billion weekly ChatGPT users is a great distribution channel. But the potential moat is the enterprise business, because enterprises must sign contracts, train thousands of employees and build workflows. Once those are in place it is hard to switch providers, so OpenAI's enterprise position may be durable even if its consumer lead is contested.

  “when you have essentially a billion people using ChatGPT a week, that's a great distribution channel. ... Potentially where the moat is where the enterprise business is. You have to sign contracts.”at 6:38

• Rory O'Driscoll · Scale Venture PartnersBullish

  A 70% quarter-on-quarter revenue jump would be a major reacceleration that rescues OpenAI's narrative.

  Rory tweeted that OpenAI's GAAP revenue grew only 18% quarter-on-quarter from Q1 to Q2, which implied near-100% year-on-year growth. That was a sharp deceleration at a time when Anthropic was growing about 10x, so OpenAI looked to be losing the race. If Q3 GAAP revenue really grew 70% quarter-on-quarter, or even 60%, that would be a massive reacceleration. It would also matter because OpenAI has made huge forward commitments to compute, data centres and land in Virginia, which only work if it grows at hyperspeed. He stresses the 70% figure is unverified and that the next question is whether Anthropic's growth came at its expense.

  “if GAAP revenue grows 70% Q on Q or even 60%, up from 18% Q1 to Q2, that's a massive re acceleration.”at 9:06

• Harry Stebbings · 20VCBearish

  OpenAI's Dots has been very unimpressive; its demo was poor, tech-centric and out of touch.

  He thinks the Dots demo was incredibly poor, very tech centric, and it asked how to book flights for enterprise usage, which he found out of touch. He contrasts it with Meta's Muse launch, which had a phenomenal reception, and calls Dots very unimpressive. The consumer-facing element of Dots was underwhelming.

  “And I respectfully think DOS has been very unimpressive. I thought that demo was incredibly poor, very tech centric.”at 1:01:10

• Jason Lemkin · SaaStrBullish

  OpenAI's Dots could be a big win as a persistent Codex agent; judge it after 30 days.

  Dots integrates with Codex and can work as a twenty-four-hour coding agent, which Replit and Lovable did well early. Claude Code and Codex can't stay running all day, which is why OpenClaw took off. If Dots is the persistent extension of Codex on the OpenAI stack, it almost can't lose unless development stops. He discounts the weak consumer launch, arguing it's a developer-focused product, and wants to see what developers build over thirty days.

  “I want to give it thirty days and see what developers do for real, right, with the persistent agent optimized around codecs”at 1:02:59

• Rory O'Driscoll · Scale Venture PartnersBearish

  OpenAI struggles to be developer-, enterprise- and consumer-forward at once; doing one thing well is hard.

  He argues that doing one thing well is hard and doing two well at the same time is super hard. OpenAI tries to be developer-forward, enterprise-forward and consumer-forward at once, and he thinks it struggles with that all the time. The product surface area is large, and that is hard to manage. He draws a parallel to Meta, where a former colleague runs its enterprise side.

  “I think that doing one thing well is hard. Doing two things well at the same time is super hard. And I think OpenAI struggles with that all the time.”at 1:04:29

Anthropic

Across the shows: 13 speakers in other episodes, 8 bullish and 30 bearish calls, past 30 days

• Jason Lemkin · SaaStrBearish

  Teams are moving workloads from Anthropic to OpenAI's Codex, so Anthropic's share is eroding.

  Jason says that every team he speaks to has moved at least a meaningful portion of its usage from Anthropic to Codex, and that people praise it. He ties this to OpenAI's recovery, arguing that people now switch models for real because the alternative is better and cheaper. He therefore doubts Anthropic's Q3 numbers will be blowout, since its lead is no longer assumed.

  “every team that I speak to has at least deviated some meaningful portion from Anthropic to Codex, everyone says how amazing it is.”at 5:15

• Dev Ittycheria · MongoDBUncertain

  Anthropic's lead is not secure; developers switch tools readily, so the Q3 numbers matter.

  Dev has sold to developers for over twelve years and finds little loyalty: developers quickly switch tools or use several at once and chase the newest option. He questions whether Anthropic can keep producing blowout numbers when the market assumes it always will, and asks what Claude or Anthropic will ship next as alternatives multiply. He treats the Q3 number as a key test of whether the lead holds.

  “there's not a lot of loyalty. Developers are very quick to use... To switch from one tool to another or, frankly, use multiple tools.”at 6:10

• Rory O'Driscoll · Scale Venture PartnersUncertain

  Cheaper open models and tighter token budgets may chip away at Anthropic's growth rate.

  Rory says enterprises have so far used frontier models out of inertia, with bills not yet huge. As numbers grow, CFOs will push CIOs to cut token bills, moving non-frontier work to cheaper open-weight models and routing some work away from Anthropic. He argues this is a chipping-away effect on Anthropic's revenue, and asks whether it can keep its unparalleled growth rate in that context.

  “Can they still maintain that unparalleled growth rate in the context of this kind of chipping away?”at 28:22

ElevenLabs

• Jason Lemkin · SaaStrBullish

  ElevenLabs is a top voice-AI asset with strong growth and margins; his fund should go up to 10%.

  Jason argues ElevenLabs has doubled its valuation to $22B while fending off intense competition, and its lead keeps widening. Voice is one of the most important underpinnings of agentic applications. He points to its impressive growth rate, surprisingly strong margins, and sub-millisecond voice response as a key differentiator. He thinks it could approach $1B in revenue by the time the friends-and-family round closes at a $50B valuation, and argues this is not JEV-level risk. He says his fund should consider up to 10% of the fund at $22B.

  “The rate of growth and surprisingly the margins are very impressive... This one could be approaching a billion in revenue.”at 34:30

• Jason Lemkin · SaaStrBullish

  ElevenLabs' enterprise voice deals are non-fungible and the team is hard to catch; his fund put 10% in.

  He is shocked at ElevenLabs' deal size, with modest companies paying hundreds of thousands of dollars a year happily. Reference checks said customers love it and feel they get a good deal, so he argues models are not fungible for this use case today. He is impressed with the team and doesn't think anyone will catch them today, though he says twelve-month outlook is hard to predict. His fund put about 10% of its capital into the company in a single round.

  “I'm shocked at the deal size of [ElevenLabs]. ... So I don't think all models are fungible. ... this is not a fungible use case today. Twelve months, it's hard to predict, but I'm impressed with the team and I don't think anybody's gonna catch them here today.”at 35:45

Meta Platforms

Across the shows: 9 speakers in other episodes, 13 bullish and 3 bearish calls, past 30 days

• Harry Stebbings · 20VCBullish

  Meta's Muse was a phenomenal launch with a phenomenal reception, the strongest AI launch of the moment.

  He says Muse was a phenomenal launch and has had a phenomenal reception, which he thinks isn't discussed enough. He contrasts it with the underwhelming reception of OpenAI's Dots. The point is that the launch's quality and reception were strong.

  “I think an additional that just isn't discussed enough is Muse was a phenomenal launch and has been a phenomenal reception.”at 1:01:10

• Rory O'Driscoll · Scale Venture PartnersBullish

  Meta has nailed Muse; its shipping, compute and execution make it as compelling as any startup.

  He had been cynical about Meta's AI efforts, but Muse changed that. Meta's team put their heads down and shipped product, launching it full-on and resourcing it with compute. He calls Muse frontier and state of the art and as compelling as any startup. He is willing to celebrate the good, given the executives' execution.

  “They ship the killer product. They launched it full on. They resourced it with compute. They kicked ass.”at 1:01:35

Reflection AI

• Dev Ittycheria · MongoDBBullish

  Reflection's US open-weight model could give enterprises near-frontier quality at a fraction of the cost.

  Dev sees Reflection's Beam as a noteworthy announcement. A US open-source model near frontier intelligence would be three to four times more cost-effective. Enterprises in regulated or sensitive industries hesitate to use Chinese open-source models, so a US option fills a demand gap. He discloses that he has a tie through Sequoia, which invests in Reflection.

  “One, you get a US open source model that's near frontier intelligence. That's three to four times more cost effective, I e, cheaper.”at 26:15

Listen Labs

• Rory O'Driscoll · Scale Venture PartnersBullish

  LLMs make market research far better; Listen Labs was an early winner executing well on AI surveys.

  Market research is a large, fragmented category with an existing big spend, and only three last-generation survey companies reached high-single to low-double-digit billion-dollar outcomes. LLMs do it much better because they can adjust questions based on what respondents say, unlike canned surveys. Voice is a solved LLM modality, and Listen Labs was one of the early contenders that executed well and got real revenue.

  “LLMs just do it so much better because what are you doing in market research? You're kind of asking questions of people and aggregating and assimilating the answers.”at 38:05

Salesforce

Across the shows: 1 speaker in other episodes, 1 bullish and 0 bearish calls, past 30 days

• Jason Lemkin · SaaStrUncertain

  Salesforce's $2B Listen Labs deal is hard to justify given only about $20M revenue.

  Salesforce is nearing a $60B revenue target, and he doesn't see how a roughly $20M revenue AI survey business moves the needle for it. He says it's not a bad idea, and he's a fan of many Salesforce acquisitions, but this one he doesn't get. To move the needle, a deal has to be big or it should be a tuck-in. He suspects it may be forgotten in a couple of years.

  “This one I don't really get. ... I just don't know how 20,000,000 of revenue of next generation AI survey moves the needle. And I'm not saying it doesn't.”at 39:01

Vercel

• Jason Lemkin · SaaStrBullish

  Vercel is a major beneficiary as agents pick vendors; agent decisions are a force of nature.

  Vercel reached $600M ARR with agents driving 50% of new business, up from 3% at the start of the year. Vercel was founded in 2020 with the right infrastructure before agents existed, and nobody predicted agents would spin up applications on their own. He says when agents pick you it is a force of nature. He also notes that his own agents have strong opinions and that expansion revenue is astronomical.

  “when agents pick you, it's a force of nature right now. ... Vercel is a major beneficiary of this.”at 47:36

Resend

• Jason Lemkin · SaaStrBullish

  Agents are steering vendor choice; Resend's MCP calls grew from about 106,000 to 3 million in months.

  Resend was the first agentic purchase his company made. His agent recommended it over SendGrid, which had deprecated its free tier and was hard to set up. Resend's MCP calls rose from 106,000 in April to 3 million in September, which he cites as evidence that agents are steering purchases to products.

  “Resend April 106,000 MCP calls to September 3,000,000.”at 50:07

NVIDIA

Across the shows: 14 speakers in other episodes, 13 bullish and 3 bearish calls, past 30 days

• Rory O'Driscoll · Scale Venture PartnersNeutral

  Nvidia's Groq license-and-hire suit likely settles, adding about $200M, since such deals dodge FTC review.

  The employees claim Groq's board sold its core technology and top 200 staff to Nvidia through an $11B license plus about $3B in stock, leaving common holders hollowed out. He says the structures exist mainly because the FTC makes M&A slow, so these deals are done overnight as a workaround. He expects a small class and a settlement, and he bets Nvidia will pay an extra $200M. He calls it a small price for having avoided a long regulatory process.

  “Grock was NVIDIA, wasn't it? Right? I'm willing to bet. Let's say they have to settle and they have to chuck in an extra $200,000,000.”at 1:00:25

Aura

Across the shows: 1 speaker in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Jason Lemkin · SaaStr

  Pulling Aura's IPO is terrible for venture, despite pricing concerns, for a company growing 74%.

  He thinks the Aura IPO being pulled is terrible news for venture. He acknowledges that the price could be argued as overpriced and that Forerunner selling its entire stake weighed on pricing. He is still bummed because a company growing 74% with an attractive recurring revenue base couldn't get out. He says liquidity is easy in the market, with M&A everywhere, but IPO exits matter.

  “I thought this was terrible news for venture, terrible news for venture because you could argue this one is overpriced.”at 1:05:55

Enterprise AI adoption & AI software

• Jason Lemkin · SaaStr

  Clever AI apps built on others' platforms are features, so more will sell to larger companies.

  He draws a line between companies building a feature masquerading as a company and those building a franchise. He thinks many AI application companies will sell because a clever product on top of someone else's platform is frankly a feature. Such companies can use the acquirer's cash and distribution, so he expects more deals like Salesforce's Listen Labs purchase. Durable companies are the ones building data loops.

  “a clever product on top of someone else's platform is frankly a feature. And if they can use the distribution of that larger company, have the cash and distribution, they're going to take advantage of it.”at 39:01

• Dev Ittycheria · MongoDB

  Durable AI app companies build data loops that copycats can't match; thin wrappers risk being features.

  The more durable companies create a data loop: usage creates data no one else has, the data makes the product better, and the better product attracts more usage. Competitors can copy features but not the accumulated interactions. The key test for standalone AI app companies is whether they are feature masking as a company or building a franchise. In his view, the Listen Labs founders chose an offer that made sense for them.

  “the usage creates data that no one else has, the data makes the product better, the product attracts more usage, and you kind of create that virtuous cycle. So if someone tries to copy you, they can copy the features, but they don't have all the interactions that you have.”at 40:12

• Dev Ittycheria · MongoDB

  Agents will steer B2B and e-commerce buying; vendors must educate agents or lose share.

  For 25 years companies optimized for page one of Google, and now they must ensure agents can find them, understand what they sell, and have current information. Missing from an AI-generated answer is like being invisible on Google. Agents now make buying decisions in e-commerce and retail, so sellers must educate them or their business goes to zero. Agents also favour incumbents because they have more data, so startups must provide documentation and APIs.

  “If you're an e commerce site or you're a retail company trying to sell through e commerce and agents are now making the buying decisions, you gotta figure out how to educate those agents, otherwise your business is going to zero.”at 48:38

AI pricing & unit economics

• Dev Ittycheria · MongoDB

  Falling AI token costs will drive broader enterprise deployment, so the market is not zero-sum.

  Dev argues that enterprise capability and usage are still far apart and that skills are scarce. He expects Jevons paradox to apply, so as costs come down, enterprises get comfortable deploying AI for more use cases. He sees demand for tokens as more than companies can imagine, and says he doesn't see the market as zero-sum among frontier and open-source providers.

  “as these costs come down, people get more and more comfortable deploying it for more and more use cases across the enterprise.”at 29:24

Open-source AI models

• Jason Lemkin · SaaStr

  US models could reach about half of open-source token share within a year, though parity is uncertain.

  Jason argues developers will switch models more easily than they switch databases, though it is still real work to qualify prompts and redo workflows. He thinks sovereignty and US basing matter a lot to enterprises. He says he is not sure the parity is really there, but if US models are roughly three months behind the frontier, the share could be half of open-source tokens within a year.

  “I'm not saying it will happen because I don't know that... I'm skeptical the parity is really there... it could be half in a year.”at 31:20

IPOs, M&A & deal activity

• Jason Lemkin · SaaStr

  License-and-hire deals will die because of double taxation and legal risk over shareholder treatment.

  He thinks some of these license-and-hire deals are really mergers in substance, and that a lawsuit may succeed even if another fails. These structures already carry heavy double taxation. Adding legal risk and carve-backs for employees left behind, as at Windsurf, will end these deals if the legal risk is significant. He notes that licensing and not-an-acquisition clauses don't change what the deal really is.

  “I think this is a mergers and substance. I think these deals will die because they already have double taxation.”at 59:17