Benchmark Breakfast #2 | November 12, 2026

Anthropic

Across the shows: 117 speakers in other episodes, 161 bullish and 93 bearish calls, past 30 days

• Ev Randle · BenchmarkBullish

  Anthropic's final IPO pricing will be raised above the initial range, to at least a $1.5T valuation.

  The initial range of $70 to $78 per share implies roughly $1.3T to $1.4T. Companies usually start with a range they can raise once demand builds and the offering is oversubscribed. Everett expects the range to be increased and the final price to land at least $1.5T, possibly $1.6T, with the aim of popping after listing. He notes the high end of the range would raise about $60B of cash.

  “my guess is that they end up increasing the range, and the final price ends up at least 1.5, if not 1.6.”at 1:28

• Jack Altman · BenchmarkBullish

  Anthropic's IPO should go well and price in the expected range, with hedge funds as eager buyers.

  The offering is large but not so big that enough capital won't be available. Hedge funds love the frontier labs and Anthropic in particular, so they will be the buyers and holders of the stock. Jack expects the IPO to price in roughly the range discussed and to go fine, with plenty of demand.

  “my guess is it goes well, honestly. My guess is it comes out in this kind of price range. There's going to be a lot of buyers.”at 5:32

Sequoia Capital

Across the shows: 3 speakers in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Ev Randle · BenchmarkBullish

  Sequoia's restructuring gives it a consistent identity, with fund size, strategy and people aligned.

  Everett argues that organizational choices matter only insofar as they align with the firm's identity and fund size, and that there is no absolute right or wrong structure. Sequoia had previously seemed unclear about whether it was a global behemoth, a scarce boutique, or a slightly fancier version of other firms. The recent changes, he says, show a clear identity: a tight-knit SWAT team investing globally, defined by scarcity, selection, and a very high bar. He concludes that for the first time in years Sequoia's fund size, strategy, and people are aligned, and that misalignment is usually when things go badly.

  “And so I think for the first time in the last few years, in my mind, Sequoia has a very consistent identity strategy, fund size, everything. Like, everything's nice and aligned.”at 58:36

• Jack Altman · BenchmarkBullish

  Sequoia's restructuring was smart: a bulk-then-cut move that simplifies the firm back to its core.

  Jack says he initially read the Sequoia changes as smart, and he compares them to YC's expansion and later cutting under Gary Tan. He argues that organizations need both bulking and cutting moments, trying many things and then simplifying back to the core, as the best founders do with shipping and cutting features. He says Sequoia moved along the spectrum toward simplicity, in contrast to Andreessen Horowitz's maximalism and Benchmark's focus. He also cautions that the real risk lies in failing to make the transitions between expansion and contraction when the firm's identity is uncertain.

  “Those are good changes. Those are smart. Makes sense. ... generally speaking, I think you need these moments where you do a bunch of things, you try stuff, and then you simplify back to the core.”at 59:07

NVIDIA

Across the shows: 115 speakers in other episodes, 99 bullish and 69 bearish calls, past 30 days

• Jack Altman · BenchmarkBearish

  Nvidia will find it harder to ship chips to China and to be vocally pro-China and pro-open source.

  Jensen Huang has argued for exporting chips to China, for open source, and for using Chinese models to strengthen an American AI epicenter. The spy allegations make it harder for chips to reach Singapore and China. Jack also thinks Jensen will find it hard to be publicly supportive of China and open source in the same way. Combined with anti-distillation pressure from the labs, this could change the landscape for Nvidia's China business. Nvidia's stock is already down about 30% from its peak.

  “it'll be hard for chips to make it to Singapore and to China and all of that. And I also think it's going to be hard for Jensen to be vocally supportive of China and open source in the same way.”at 11:41

Venture capital & startups

• Jack Altman · Benchmark

  Largest VC firms will raise roughly $50-60B in the next one to four months.

  Jack and Everett tallied the funds they had heard were raising, including reports of $5B, $11B and $16B raises, and reached about $60B from five or six firms, probably not counting everything. All of these firms raised their last funds only about 12 months earlier, which Jack says is the typical gap. Jack notes that if the funds deploy over a more typical two to two-and-a-half years, the outlays are very large, and he asks where the money comes from and where it goes. He says the firms are probably right to raise now because the money is still available, and that the market's tough times will also bring opportunity.

  “we believe that something like 50,000,000,000 to $60,000,000,000 of venture capital is going be raised by the largest firms over the next, I don't know, one to four months.”at 41:15

• Ev Randle · Benchmark

  Outside frontier labs, putting in huge venture capital has not yet proven high money-on-money returns.

  Everett argues that the frontier AI labs were a one-time subsidy that let the mega funds deploy huge amounts of capital and earn great returns. He says the labs were uniquely able to absorb tens of billions of dollars and still generate strong returns. Everywhere else, he says, the ability to put in a lot of capital and produce high money-on-money returns is unproven. He suggests the many firms now raising huge funds will have to grapple with this question, and that the labs' success may not be repeatable.

  “the Frontier Labs were sort of a one time subsidy on this phenomenon of immense ability to suck in capital and produce great returns. And everywhere else that you can put in a lot of capital is completely unproven in terms of its ability to actually produce high money on money returns.”at 43:22

• Jack Altman · Benchmark

  Outside SpaceX, Anthropic and OpenAI, most great startups do not need anywhere near this much capital.

  Jack argues that outside the largest capital consumers, the rest of the venture market does not need that much money. He gives Databricks and Stripe as examples of great businesses, and says they do not need anything like $150B. The point undercuts the case for the very large venture funds, since most of their capital cannot be absorbed productively by the companies they back.

  “anthropic open AI, the rest just don't need that much capital. ... how much money do Databricks and Stripe really need? It's not that it's not 150,000,000,000”at 43:43

• Jack Altman · Benchmark

  Mega VC funds are squeezing midsize funds as LP capital concentrates in the biggest firms.

  Jack argues that there is always room for small early-only funds, but that midsize funds are being squeezed by LPs as the largest firms grow. The mega funds are pushing up against the total demand for venture investment, so they absorb the capital that would otherwise go to midsize managers. Everett agrees, calling the mega funds suffocating squeezes on midsize funds. Jack says historically the well-known firms raised less than LP demand, which allowed a broad ecosystem to form, and that if the funds keep growing, they will take all of the available dollars.

  “there's always room for a small early only fund. But I think there is a group of funds that's midsize that I do think gets a lot of LP squeeze as a result of these other funds just being so big that they truly are starting to push up against the total desire for venture capital investment.”at 47:21

• Ev Randle · Benchmark

  Private markets follow public markets with a three-to-six-month lag, in a traffic-wave pattern.

  Everett describes a consistent pattern in which private markets trail public markets by about three to six months and move less sharply, since they are more fragmented and have no stock chart. When public investors are hit, crossover investors who also do pre-IPO deals become less willing to pay for pre-IPO rounds, so those valuations fall. That then pressures the earlier rounds, since founders and investors can no longer count on markups. The shock works its way back toward Series A, so investors fear a logjam and become cautious.

  “there's a pretty consistent pattern in private markets, which is that they follow public markets with some delay, call it three to six months.”at 49:05

• Jack Altman · Benchmark

  Seed valuations are now permanently expensive, as seed has become a safe haven for capital.

  Jack argues that the logic of the private traffic jam led investors to treat seed as always safe, since a seed round is the cheapest round and always has a safe haven unless the company fails outright. Large firms now write seed checks that do not change their economics, so the check size no longer constrains price. Founders also skip angel rounds and go straight to Series A, which means fewer early-stage rounds are available at lower prices. He concludes that seed has moved into unexplored territory and that seed valuations will stay permanently high.

  “the knock on effect to that has been that seeds have just become permanently expensive. ... Seeds are now permanently expensive. Because you have this oscillating traffic jam ahead of you.”at 51:21

AI infrastructure & data centers

• Ev Randle · Benchmark

  AI ecosystem revenue cannot justify the planned data center capex, so the buildout and its estimates will shrink.

  Estimated 2027 data center capacity outside China has fallen from about 70 GW to 40–50 GW, a cut of roughly 40% at the low end. That cut lowers CapEx numbers and questions the return on the spending. Revenue from the AI ecosystem is not large enough or growing fast enough to justify the capex planned for the next two years. Companies with weaker balance sheets, such as neoclouds financing data centers with debt, are most exposed if capital markets tighten and the ROI is questioned.

  “when you get down to it, the revenues from the AI ecosystem just aren't big enough at this point, and they're not growing fast enough at this point to justify the amount of CapEx that's being contemplated for the next two years.”at 21:41

• Jack Altman · Benchmark

  The data center buildout will eventually overshoot and halt, and price per gigawatt cannot keep rising.

  Jack says the odds of a massive buildout ending smoothly are effectively zero. At some point everyone will stop and admit they have overbuilt, and spending will fall sharply. He adds that the price per gigawatt cannot keep rising from $30B to $200B. Ground-level information suggests strength over the next 12 to 24 months, but the trend cannot continue forever and will need to find an equilibrium.

  “There will come a moment when everybody screeches to a halt, and they say, we have just... Like, we gotta stop. We've overbuilt.”at 27:35

• Jack Altman · Benchmark

  Americans broadly dislike AI and data centers, and tech has not yet connected with ordinary voters.

  People resent AI billionaires and trillion-dollar companies dropping large, ugly data centers into their towns during a huge wealth creation event and telling them these are their jobs now. In Texas and other places, this sentiment is strong. Tech made progress connecting with DC but has done a poor job connecting with ordinary America, though it has started trying harder in recent months.

  “We're like, America hates AI, big time. And America hates data centers.”at 31:38

• Ev Randle · Benchmark

  Data centers can be symbiotic with host communities when done right, as Meta's Louisiana project shows.

  Everett argues that the backlash against data centers in places like Texas is overstated and that host communities can benefit. He cites Meta's Louisiana Hyperion project, which he says roughly doubled the county's local tax receipts and raised teacher bonuses by about 400%, often to the level of their salaries. He says local concerns about air and water pollution can be addressed. He also says the shift to a role reversal in which cities now bid for data center developers, as they once bid for Amazon HQ2, is an okay thing because it lets communities extract value from the build-out. He concludes the relationship can be very symbiotic if it is done right.

  “that project alone doubled the local tax receipts for that county and gave teachers bonuses that were equal to their salary in many cases... So when it's done right, I think these can actually be extremely symbiotic.”at 37:04

Open-source AI models

• Ev Randle · Benchmark

  Losing Jensen Huang as Nvidia CEO would be a major headwind for open source and startups.

  Jensen has been the key figure arming open-source challengers to the frontier labs and has funded many US open-source startups. Nvidia is structured around him, and it would be hard to imagine it without him, so it would slow down and execute less aggressively. Everett says the spy allegations have cost him the ability to be vocally pro-China, pro-open source, and pro-chip exports, even though Ben Thompson gives rational reasons for selling chips into China to build on CUDA.

  “if Jensen is removed from the CEO role, that is a huge headwind. Again, not just to China... Not just to Chinese open source. Bad for open source writ large.”at 13:47

• Ev Randle · Benchmark

  American open source has become immensely critical as Chinese open source becomes harder to rely on.

  American open-source models have trailed Chinese ones for many reasons, including less aggressive distillation. If Chinese open source becomes harder to access and a more tenuous relationship, the strategic need for American open source rises sharply. The episode has lit a match under this, and the strategic imperative had already been growing before these events.

  “If America cares about open source at all, if Chinese open source is going to be harder and harder and be more tenuous as a relationship and all these things, American open source has just become immensely critical.”at 16:13

• Jack Altman · Benchmark

  On net, the Nvidia news is positive for American open source, even as closed labs push their own case.

  Frontier labs will likely push the message that closed source is the safer, cheaper option while the situation is unclear. Still, many people care about open source and correctly believe it makes sense. Jensen is not the only advocate, so the forces behind open source will largely offset the losses and leave American open source a net positive.

  “I would still on net take the side that this is positive for American open source, though. There's too many people. It's not just Jensen.”at 16:47

Frontier AI labs & model race

• Ev Randle · Benchmark

  AI labs' low switching costs make their revenue volatile, so they will trade at a discount to software.

  Customers in the API business can switch providers easily. OpenAI cut prices and took share from Anthropic for a few months over the summer, and that slowed Anthropic's growth. Everett argues this makes lab growth and stock prices inherently volatile. Unlike SaaS, which had steady sequential growth from sales-capacity models, labs will not show that stability, so they will likely trade at a discount to pure software unless it changes.

  “it's very clear that, especially in the API business, the switching costs are not high.”at 8:39

• Jack Altman · Benchmark

  AI lab stock volatility is rational because their possible outcomes are far wider than a stable company's.

  Coca-Cola's operating range is narrow, so its stock trades in a tight distribution. For a lab like Anthropic, the outcomes are not a small band around a likely revenue number. It could be 5 or 9, or something terrible could happen, so the range of outcomes is wide. Daily news can push the stock sharply one way or the other, so the volatility is a rational response to that uncertainty.

  “the volatility is sort of rational. And it's rational because you think about Coca Cola stock. It's like the bands of what that company is going to do are very narrow.”at 9:34

AI: overall outlook

• Ev Randle · Benchmark

  A major AI capability breakthrough is fairly likely within 12 to 24 months, adding to the next leg of the trade.

  Everett says many people in the AI investing circle believe there is a solid chance of a major unlock in intelligence over 12 to 24 months, though he doesn't know its form. The bull case depends on that unlock, and it would be clear when it arrives. He admits that with the stocks down, this feels like hopium, but thinks the rapid pace of model improvement makes a breakthrough quite possible.

  “there's a pretty solid chance that over the next twelve to twenty four months, we don't know what it is. But there will be some massive unlock in intelligence.”at 28:58

• Jack Altman · Benchmark

  The 36% US versus 93% China AI-optimism gap reflects a narrative problem, not real differences in AI's impact.

  Jack cites a Gallup poll in which 93% of Chinese respondents said AI will mostly help the country versus 36% of Americans, which is near the bottom of the countries surveyed. He argues the gap cannot be explained by real differences in how AI will affect the two countries' lives, so it comes from sentiment and narrative. He thinks the US public's dislike is probably rooted in real concerns, and that AI can be a real problem for people. But he says the gap points to a large narrative problem, because the public's experience of AI's benefits does not match its fears.

  “There is no real reason for that discrepancy between the two. It is not going to impact the lives of one country that much differently than the other. That is sentiment and narrative.”at 38:30

AI pricing & unit economics

• Jack Altman · Benchmark

  Token buyers will eventually rationalize AI spend, so consumption grows but not at an exponential rate.

  Within penetrated accounts, AI consumption will show exponential growth at first. Then buyers reach saturation for their tasks, decide the last model is good enough, and move to cheaper models. Spending then flattens instead of eating the whole margin. Growth continues because AI is still far from penetrating the rest of the economy and each new frontier model can drive another spending step-up. Jack expects that to happen fast by normal standards, but diffusion will take time and won't follow an infinite exponential path.

  “at some point, then, what you see is the buyers of tokens get to a place where they decide, Okay, we're ready to rationalize costs. We have saturated the task.”at 23:57

Robotics & autonomous vehicles

• Jack Altman · Benchmark

  A lab demo of perfect robot control will come and cause a major stock rally, as scaled models reach robotics.

  Under the bitter-lesson view, enough compute and model size applied to robotics could produce good robots and a very large S-curve, possibly bigger than coding. Jack expects one day a lab such as Tesla, OpenAI or Anthropic will demonstrate perfect control of a robot. That would drive a huge upward move in the stocks, and similar surprises will cause shocks both up and down before then.

  “there will one day, there will be a demo where one of the labs will show that they have perfect control of a robot.”at 30:03

US elections & politics

• Ev Randle · Benchmark

  A Democratic House, Senate and presidency in 2028 could undo Silicon Valley's gains in Washington.

  The current administration has had the most Silicon Valley representatives in the White House and government, but DC polarization is at a very high level, making the relationship knife-edge rather than collaborative. Everett worries that if Democrats control everything, Congress could haul tech figures in front of committees over every government deal, including those by defense tech startups, which would undo progress made so far.

  “you almost wonder, well, if the House, Senate, and presidency are blue... what does that mean for all the progress Silicon Valley has made in DC?”at 33:53

AI & jobs

• Jack Altman · Benchmark

  The fear of AI taking jobs is a narrative problem; automation is not remotely starting to do that yet.

  Jack says the industry has been telling people that AI will take their jobs and that data centers will automate away the work they have spent 20 years learning. He argues this framing is why people push back. He says that automation of that kind is not even remotely starting to happen. He treats the job-loss fear as a narrative problem rather than a current economic reality, which helps explain the poor public sentiment on AI.

  “We've gone around telling people, we're going to take all their jobs. ... It's not even remotely starting to do that.”at 39:28

US–China AI & chip race

• Ev Randle · Benchmark

  US public hostility to AI and data centers is a long-run structural disadvantage against China in AI.

  Everett argues that leading in AI has become a major geopolitical imperative, and that the US is at a structural disadvantage against China over the long run. He points out that China already has more solar and power generation than the US and much better public perception of AI. He says China benefits from the US public's hostility to AI and data centers, and from American political distraction, including in the Middle East. He concludes that this is a worrying dynamic for US competitiveness.

  “a massive, massive imperative from a geopolitical standpoint to have the best AI models. ... China already has more solar and power generation than we do. They already have much better public perception.”at 40:11