Nov 27, 2025 · 1h 32m · news
Anthropic Raises $30B from Microsoft & NVIDIA & NVIDIA’s Core Business Faces TPU Threat · 20VC with Harry Stebbings
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Host Harry Stebbings and tech experts Rory O'Driscoll and Jason Lemkin analyze the rapidly shifting dynamics of the AI sector, discussing massive funding rounds, emerging hardware threats to Nvidia, and the complex valuation disparities between AI startups and tech incumbents.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 10% of the talking time here. How this is scored →
Harry as informed peer 4.4 Guest teaching 4.9 Guest disagreement 3.5 Harry pushing back 4.0
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:11:06 Jason calling GEO tools fraudulent snake oil
Jason aggressively attacks the entire Generative Engine Optimization category, claiming startups demanding credit cards upfront without free trials are selling snake oil and bordering on fraud.
Hardest push from Harry ▶ 1:12:00 Harry placing a $5,000 bet to counter Jason's framingHarry directly refuses Jason's dismissive framing of the GEO sector, backing his portfolio company Peak.ai by proposing a $5,000 personal bet on its success.
Biggest teaching moment ▶ 6:04 Rory detailing the microeconomics of custom TPU developmentRory takes Harry's high-level question about chip verticalization and delivers an authoritative economic breakdown of Google's $90B CapEx and Nvidia's 75% gross margins.
Harry holds his own ▶ 15:07 Harry articulating specialized data requirements in data labelingHarry demonstrates deep domain domain expertise gathered from interviewing founders, explaining verticalized data nuances so convincingly that Rory acknowledges the point as super helpful and insightful.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Welcoming Guests to the Upgraded 20VC Studio | 3 | 4 | 3 | 2 | Harry sets the context regarding Anthropic's funding deal with Microsoft and Nvidia, while Rory playfully corrects Harry's omission of Microsoft's broader strategic positioning with OpenAI. | |
| The TPU Threat: Custom Chips vs. Nvidia's Dominance | 4 | 7 | 2 | 4 | Harry asks a strong question about verticalization and custom chips, leading Rory to deliver an in-depth breakdown of Nvidia's customer concentration and CapEx math. | |
| Customer Concentration Risk & The AI Data Labeling Market | 8 | 3 | 2 | 4 | Harry demonstrates strong subject-matter expertise on the data labeling sector, explaining his thesis on specialized vertical requirements which earns Rory's explicit praise. | |
| Nvidia's Valuation: P/E Ratios and Compute Demand | 4 | 5 | 2 | 4 | Harry pushes the panel directly on Nvidia's valuation, prompting Rory to break down end-demand compute cycles and P/E comparisons against retail giants like Costco. | |
| Analyzing "War Mode" and Hyper-Aggressive Corporate Culture | 4 | 5 | 6 | 4 | Jason takes an aggressive stance on hyper-aggressive corporate culture and pushing management past traditional limits, while Rory pushes back on military war metaphors. | |
| The Enterprise AI Race: Sierra's Valuation & Customer Support | 5 | 5 | 4 | 6 | Harry puts Rory on the spot with an uncomfortable comparison between Intercom's valuation and Sierra's 100x ARR valuation, forcing Rory to defend his portfolio investment. | |
| The "Cement Shoes" Dilemma: AI Incumbents vs. Startups | 3 | 5 | 6 | 3 | Jason characterizes incumbent user bases as heavy cement shoes due to technical debt, which Rory forcefully rejects as ignoring core data advantages. | |
| Customer Segmentation and Churn in AI Startups | 4 | 5 | 2 | 3 | Jason breaks down customer segmentation dynamics in fast-growing AI startups like Lovable, explaining why low-end churn is standard. | |
| Public vs. Private AI Valuations and Growth Multiples | 4 | 6 | 3 | 4 | The panel compares public multiples like Wix against private AI valuations, with Jason highlighting how brutal public markets are to low-growth SaaS. | |
| Generative Engine Optimization: Actionable AI vs. Snake Oil | 6 | 5 | 7 | 7 | Jason dismisses Generative Engine Optimization as snake oil and fraud, prompting Harry to strongly defend the sector and challenge Jason with a $5,000 bet on Peak.ai. | |
| The Stagnant IPO Market, Wiz, and Regulatory Obstacles | 4 | 5 | 3 | 3 | Jason and Rory discuss the lackluster IPO environment using Figma's trading performance as a benchmark, alongside antitrust regulatory friction. | |
| Venture Capital Markdowns and 2021 Unicorn Valuations | 4 | 5 | 2 | 3 | Rory explains why 2021-era tech valuations must be systematically marked down unless re-accelerated by core AI progress. | |
| Quick Fire Would You Rather: Wix vs. Lovable | 4 | 4 | 3 | 5 | Harry conducts a quick-fire segment asking the guests to pick between investing in Wix or Lovable, driving a debate on founder retention incentives. |
Statements from this episode (47)
Rory O'Driscoll: Nvidia's P/E ratio is lower than Costco's
“The P.E. On NVIDIA, to give that sound bite that I love, it's lower than the P.E. On Costco, right?”
Jason Lemkin: Startup leaders cannot push their teams too hard
“I do not think you can push your team too hard. I think you should push them as hard as the business needs to go, and if they leave, it's great.”
Jason Lemkin: Salesforce Agentforce could render AI startups obsolete within two years
“Salesforce could be the next Google. It could be in 18 to 24 months. We could be like, oh my god, Agent Force crushed it. All these startups, we didn't need them because Agent Force is so good.”
Rory O'Driscoll: Big Tech extracts an excessive portion of global profits
“The great American tech Mag-Savans suck more of the world's profit dollars out of the rest of humanity.”
Rory O'Driscoll: Anthropic plans to build its own physical data centers
“Kind of, you didn't mention it, but in the same announcement, I think, Anthropic also said they're going to break ground on a physical data center. So that's another one of the model companies saying it's not enough to rely on compute from your service provide…”
Rory O'Driscoll: Bull markets ignore risk discipline until bear markets arrive
“In a bull market, nobody cares about everything. And then in a bear market, everybody discovers why you were meant to care. And now we're in the don't care part of the trade.”
Rory O'Driscoll: Microsoft-Anthropic-Nvidia compute deal has better fundamentals than peers
“Of all the round tripping deals, I would argue a Microsoft, Entropic, Nvidia deal probably has better principles and upside than most. I mean, some of the other round tripping deals look like they're already, you know, a bit shaky.”
Jason Lemkin: Seamless switching between TPUs and GPUs undercuts Nvidia's moat
“The idea that NVIDIA is unstoppable because of the software and hardware connection, because we have to have GPUs. I know there's a lot of truth to that, but literally as an end user, I went right back and forth to them today. No, no issue. TPUs, GPUs, LLMs, a…”
Rory O'Driscoll: Four to five customers generate 70-80% of Nvidia's revenue
“The odd thing about the Nvidia business, unlike most other businesses, is four or five of their customers account for 80% of the revenue, something like that, 70, 80%.”
Rory O'Driscoll: Google is spending $90 billion on CapEx this year
“Google is spending ninety billion this year on CapEx.”
Rory O'Driscoll: Nvidia backs neoclouds because they cannot build custom silicon
“First of all, there's a reason why they sponsor the Core Weaves and the next generation of NeoClouds, because they're like, those guys ain't going to build their own chip.”
Rory O'Driscoll: Nvidia customers will not abandon CUDA due to switching costs
“For most users, because they have such dominance, such validation of the CUDA software layer, for most customers, it's going to be too much brain debt to switch from the cheap, you know, the GPU you know and love to something new, right? Because there's probab…”
Jason Lemkin: Investors ignore systemic AI risks due to Nvidia's strong numbers
“I actually think going to the opening of the show, I actually think we're just ignoring the risk because the NVIDIA's numbers are just too good. No, that's what we're all doing. We're ignoring it. This is systemic risk.”
Rory O'Driscoll: I was wrong to fear AI data labeling startups
“I mean, I was, I will admit I was scared of the data labeling companies and I was wrong. There was a period of a couple of years where they clearly worked and worked really well.”
Rory O'Driscoll: Undifferentiated startups must differentiate or exit before growth slows
“The earlier you are with The more hyper growth you have ahead of you, the easier it is to be undifferentiated. And by the time the growth slows down, you better either be differentiated or as Jason was, you better be exited.”
Rory O'Driscoll: Microsoft compute spending will grow 30% next year
“Is Microsoft going to increase another 30% next year? Is Antropic going to invest in compute? And right now, all those answers to those questions are yes.”
Jason Lemkin: Hedge fund backed my VC fund solely for startup data
“I mean, I remember my first fund, I had a hedge fund invest. The only reason they invested was to get that data from the portfolio.”
Jason Lemkin: Tech companies accomplish nothing unless operating in hyper-aggressive mode
“Across every professional experience I've had from startup to scale up to VP at a fortune, 500 tech company, nothing happens when you're not in hyper aggressive mode.”
Jason Lemkin: Google AI products give users no reason to switch
“I actually prefer AIO reviews because I don't have to leave Google. That's a great product. I mean we mean, we're so is Gemini three pro. These are great products. There's no, you're not giving me any reason to leave. They're great. They're truly great softwar…”
Rory O'Driscoll: ChatGPT will succeed without destroying Google Search
“The likely outcome for me is ChatGP continues to build this compelling consumer product, can command differentiated market, while at the same time Google search does not quote unquote go to zero.”
Rory O'Driscoll: OpenAI lost an entire team following its 2023 board crisis
“In the last two years since the ChatGPT, since the OpenAI board change, you have lost an entire team at OpenAI.”
Jason Lemkin: Overpushing key employees forces them to build rival competitors
“You'd rather not create an anthropic when your team leaves too, let alone thinking machines and all that. You don't want to force two of your best people out and they build your top competitor.”
Jason Lemkin: Enterprise AI customer support startups are massively overselling their capabilities
“The whole category, the enterprise side of it is massively overselling what they can deliver today.”
Rory O'Driscoll: LLM-powered support tools double call resolution rates to 60%
“Before LLMs, and I had an investment in a company that was pre-LLM customer support. Our resolution rate was around 23%. In other words, we could solve 23% of calls, maybe up to 30. With LLMs, with these new modern things, you can solve 60%, which means on a n…”
Rory O'Driscoll: Sierra must target $200B labor market to justify $10B valuation
“The only way this math works is if you eat a huge slug of the labor and because you eat the labor, instead of the, you know, twenty billion dollar software market for services, you eat the two hundred billion dollar a year services market for customer support …”
Rory O'Driscoll: Enterprise change management will limit Sierra's growth speed
“And I think the physics of diffusing this technology into the enterprise will be the rate limiter on how fast this company can grow. It won't be raw demand. It won't be the talent Of the CEO. It won't be the product. It will literally be, can you really roll o…”
Rory O'Driscoll: Intercom's AI product growth rate is comparable to Sierra's
“I think they've done an amazing job in a way that very few SaaS companies have of transitioning to an AI first world, and the growth rate of their AI product is frankly comparable to Sierra's”
Jason Lemkin: Portfolio firm sees 100% AI growth alongside flat legacy revenue
“And I was literally with a portfolio company. It's got. Forty-five million of AI revenue growing about a hundred percent. Okay. Now listen, that's not a lawful or, but it's pretty good. Okay. And then it's got fifty million of pre AI revenue growing zero.”
Jason Lemkin: Base Zendesk AI achieves 20% ticket automation without custom training
“The base version of Zendesk, which you get with AI, it's about 20% automated. That's about as well as you can do without training.”
Jason Lemkin: Most B2B software unicorns will fail transitioning to AI
“This is why I think most of our unicorns will fail. Because you've got to be as good or better than Owen to make this transition. Most of them are going to fail. When I look at so many of these B to B unicorns, they just, it's so hard to do both.”
Jason Lemkin: AI music startup Suno retains only 20% of customers annually
“They just said they have, they only retained 20% of their customers in a year”
Jason Lemkin: AI startup retention will improve as enterprise customer bases grow
“As these guys grow, their NRR will go up and the churn will go down, just because for classic enterprise reasons, those big customers are going to be stickier.”
Jason Lemkin: Low-growth software companies are worth only 3x to 5x ARR
“Generally, once you descend into low growth and that's different for public companies and private companies, you're worth, I mean, if you're worth anything, you might be worth nothing as a startup, you're worth in the three to five X ARR range”
Jason Lemkin: Public software multiples jump sharply above 20% and 30% growth
“If you're growing less than 20% of a cherry pick basket, 5.1 X AR, Wix is lower. It's not in my basket. 20 to 30, 11.8 X AR. And then the 30% plus, which is Rubrik, Palantir and Figma, average 23.7 X AR, but Palantir and Rubrik push that up.”
Rory O'Driscoll: Tech investors prioritizing profitability over growth after 2022 were wrong
“I think markets pay up for growth, and it's why all these people who in 2022 were like, just get profitable, they're wrong.”
Jason Lemkin: Generative Engine Optimization AI software is snake oil
“I view the pro what there is a category of snake oil AI of which geo is one of them. Okay. And I'll tell you why it's snake oil. And it's gonna die in 26 or 27.”
Jason Lemkin: SaaStr overall traffic grew 50% while SEO traffic dropped 8%
“I get about five million to our blog, ok? About five million views a year from SEO, ok? Our traffic is up 50% this year, and our SEO is down eight percent.”
Jason Lemkin: AI marketing tools will reach $1B ARR in 36 months
“As soon as they do, this will be a billion dollar ARR opportunity in 36 months. But the snake oil isn't going to do it.”
Rory O'Driscoll: Advertisers will shift budgets to LLM optimization tools
“If all the people are showing up In the LLMs, then all the advertisers are going to want to show up in the LLMs. And if you can help them get there, you can probably clip some of that money.”
Jason Lemkin: Figma's public valuation is 30-40% below Adobe's acquisition offer
“I think it's like 30 or 40% lower than Adobe deal adjusted for time dilution and risk, 30 to 40% lower.”
Jason Lemkin: The tech IPO market will not reopen next year
“I don't think we're going to see it. I don't think we earned it this year. I don't think we have so many. You're better than Figma. You're better than Netscope. It's a high bar.”
Rory O'Driscoll: Non-AI pre-2022 unicorns will likely never go public
“And the scary thing, going back to linking it to the AI thing, is with every year that goes by, the probability is that the next crop of IPOs will be AI first companies from 2022, or pre-AI, pre-twenty-two companies that have clawed their way into AI land. And…”
Rory O'Driscoll: Scale Venture Partners has marked down 2021 peak valuations
“I don't, it's very hard to imagine any 20, 21 mark that's been maintained that hasn't been validated or anywhere close to it, and we wouldn't and haven't.”
Jason Lemkin: Most teen-percentage software growth is driven by price increases
“Because if you're growing 15% or so, or in the teens, probably the majority is from price increases in today. That's, you know, it's fake growth. It is price increases or stuff jammed down the channel, and maybe it's a couple percent from new logos, but if you…”
Rory O'Driscoll: Cutting R&D in downturns prevents long-term software growth
“If you cut R and D in the downturn, you're probably screwed. If you don't have that compelling second product that you built in 22, 23, 24, you know, you're just in a tough place.”