Topics Pioneers Insight Method Author
Back

Jensen Huang probe; Sequoia resets; Anthropic’s $1.4T IPO target

·20VC

Rendering the long image…

Topics US-China AI Rivalry and Geopolitical Controls Early-Stage VC Investing Dynamics Trillion-Dollar AI Capex and Bubble Risks AI Venture Funding and Bubble Risks Battle for Frontier AI Moats Cybersecurity and Automated Penetration in AI Era Autonomous Agents Reshaping Enterprise SaaS

Summary

  • The panel sees the allegations involving Jensen Huang as a threat to China’s access to NVIDIA compute. Harry argues Huang “needs a pass” because of his importance to US AI, while Rory warns that unresolved security and compliance questions make leniency impossible to assume. Jason’s strongest prediction concerns offshore access: putting GPUs just outside China’s borders will become harder, too.
  • Sequoia’s restructuring favors a concentrated partnership over geographic coverage, scouting infrastructure, and complicated holding structures. Jason defends holding generational assets for decades, but Rory argues the Evergreen fund complicated LP liquidity without changing partners’ ability to retain their own shares. London’s departure does not invalidate European companies; the question is whether Sequoia can reach the strongest founders from the US.
  • Rory reads the $2.5 trillion NASDAQ 100 sell-off primarily as a positioning shock rather than evidence of collapsing AI demand. “The AI trade doesn’t crack on the supply side”: his decisive indicators remain OpenAI and Anthropic revenue. Jason nevertheless sees legitimate nervousness around the capital, debt, and financing assumptions supporting the buildout.
  • Anthropic’s IPO divides the panel on initial trading but produces agreement on the strategic value of becoming public. Rory likes the $70–$78 indicative range against $54 in the last private round; Jason predicts flat trading and worries about funding fatigue. Rory conditionally sees $1.1 trillion becoming $1.4 trillion, while both expect public-market financing tools to strengthen a balance sheet Rory puts at $200 billion.
  • OpenAI’s reported rebound strengthens the competitive race while leaving token and agent economics unresolved. A panelist cites 70% annualized-revenue growth in Q3, with enterprise doubling into a majority of revenue. Jason expects continued “ping pong” between developers’ preferred models and admits he underestimated persistent-agent costs; Rory expects eventual cost declines but cannot say whether meaningful relief is two or five years away.
  • Revolut wins the compounding argument, while Jason favors Fal for its immediate exposure to inference demand. Harry chooses Revolut at $115 billion and sees at least 4X potential, with founder dependence the central risk. Jason chooses Fal at $16 billion despite durability concerns; Zipline’s $21.5 billion valuation demands conviction that drone delivery becomes infrastructure, rather than merely a business earning $5–$10 per delivery.
  • Stolen inference keys are becoming a direct financial threat and a catalyst for security spending. Jason reports two Fireworks breaches and describes a $100,000 default spending allowance on his own account as an invitation to rapid losses. Scarce replacement capacity can keep customers attached even after incidents, while CrowdStrike and Palo Alto acquisitions address the emerging agent-security market.
  • Agents are widening the divide between infrastructure beneficiaries and software businesses dependent on human seats. Jason connects Cloudflare’s 39% growth to agent activity and questions whether interface quality protects products agents can bypass. Figma illustrates the valuation problem: roughly 40% growth and strong cash generation coexist with a valuation around half its blocked $20 billion acquisition offer; his proposed response is an aggressive commitment to agentic design.

Deep dive

1. Sequoia’s Evergreen reversal separates investment logic from fund design

  • For Jason, winding down Evergreen matters more than personnel changes: holding generational assets for decades was “the best idea in venture ever.” NVIDIA and Google are his examples of what distributions can surrender. Rory accepts the premise: a diversified portfolio of high-quality Sequoia-backed public companies could compound attractively. His objection concerns the mechanism for making LPs retain it.
  • Rory identifies three problems: implementation in 2022–23 exposed investors immediately to falling stocks; LPs have different liquidity needs; and partners can already retain distributed shares after earning carry. “Trying too hard, too much complexity.” A sound long-term portfolio argument does not mean every LP wants the same holding period.
  • Jason offers Ho Nam’s account of Altos Ventures using continuation funds to let investors exit or remain invested, while acknowledging conflicts around those vehicles. Rory’s pushback concerns choice: once partners have earned their economics, insisting LPs keep holding resembles administering unwanted vitamins. “Damn it, you will hold these stocks ’cause it’s good for you.”

2. A smaller Sequoia prioritizes concentration over coverage

  • Harry reads London’s closure as a coverage decision, citing ElevenLabs, DolarApp, Odoo, and Bolt as evidence that Europe produces valuable assets. Jason agrees that leading founders repeatedly visit the Bay Area, reducing the need for local coverage later on. His qualification matters: “I’m murky on pre-seed,” where proximity and hands-on work may remain essential.
  • Ending Scouts and reducing organizational sprawl fit Jason’s explanation of a tighter partnership. Scouts’ original secrecy and lead were eroded by copying; programs such as Arc require management energy. His broader warning to founders: “the institution may be durable, no particular partner may be durable.”
  • Founders Fund supplies the counterexample to large investing teams: a panelist says roughly three to five principal investors deployed its growth fund in 11 months into companies including Stripe, Anduril, and SpaceX. Rory describes the model as diversified early experiments followed by concentrated growth investing, narrowing approximately 20 investments to seven or eight. Large checks into established winners require conviction more than extensive infrastructure.

3. The Huang allegations put offshore GPU access in the spotlight

  • A panelist introduces the discussion by saying Huang admitted an affair with a suspected Chinese spy and that NVIDIA’s board is investigating. Rory separates the personal allegations from unresolved questions about compliance, security, and lobbying concerning MegaSpeed. He sees a “non-zero chance” that problematic information emerges, while emphasizing Huang’s importance to the company.
  • Harry argues Huang “needs a pass,” comparing his strategic importance with Steve Jobs during the stock-option backdating controversy. Rory resists the analogy: national-security scrutiny could make this substantially harder to contain. His distinction is explicit: “you can hope a pass, but you can’t assume it.”
  • Jason’s commercial concern is the practice of serving China-based models through GPUs hosted in Singapore or Thailand. He predicts tighter restrictions on both direct sales and offshore hosting: “It’s all going to get shut down.” Rory argues Huang already had a rational NVIDIA-specific reason to favor Chinese sales—preventing alternative suppliers from gaining share—without needing outside persuasion.
  • Jason’s governance lesson from two previous CEO investigations is that selecting investigators and defining their remit can shape the outcome. Rory acknowledges that risk but expects external scrutiny to constrain it here. He highlights the board’s long relationships with Huang and says an investigation potentially checked by Congress or the Department of Justice cannot simply rely on loyalty.

4. Rory sees a positioning shock rather than an AI demand break

  • Harry asks whether the NASDAQ 100’s $2.5 trillion one-day loss signals the AI trade cracking. Rory stresses the sequence: stocks initially rose after the Democratic election result became apparent, then fell amid economic and data-center concerns. He suspects overextended hedge-fund positions amplified the move, citing risk limits that remove a trader’s checkbook at a 5% drawdown and employment at 7.5%.
  • “The AI trade doesn’t crack on the supply side,” Rory insists. Data-center delays or NVIDIA losing share to AMD do not invalidate the broader trade if demand survives; OpenAI and Anthropic ARR are his evidence to watch. Individual companies such as CoreWeave could overreach. Jason nevertheless thinks markets have reason to worry about the capital deployment and debt required.

5. Anthropic’s IPO divides trading expectations but strengthens its financing position

  • Jason initially interprets the proposed financing as showing fatigue: approximately $100 billion sought, a roughly $20 billion private placement, and some Google commitments potentially repackaged. He repeatedly qualifies that reading. Rory corrects the timing: the transaction has not reached the stage where oversubscription can be assessed. Jason accepts that it has not priced, while retaining his concern about capital availability.
  • Rory favors the indicative $70–$78 share price against $54 in the last private round. “It is far better in an IPO to start low and go up than to start high and come back.” He explains the roughly $1 trillion versus $1.3–$1.4 trillion valuation descriptions through different treatment of granted but unvested options, rather than simply a falling price.
  • Asked whether $1.1 trillion could become $1.4 trillion, Rory “wouldn’t be surprised,” conditional on a normal market and nothing extraordinary happening. Jason predicts flat trading, acknowledging retail is difficult to forecast. Harry relays an allocator’s concern about “the next 100 billion,” rather than the current $80 billion: the available capital layer may be thin.
  • Rory’s strategic case is stronger than his trading forecast: Anthropic would have $200 billion in cash, public float, and access to converts and follow-on offerings. He imagines a roughly $50 billion convert within six to nine months; Jason suggests $40–$60 billion within four to five months, if available. Both welcome quarterly disclosure. Public status ends what Rory calls the “private begging bowl bullshit.”

6. Model competition accelerates while persistent agents strain economics

  • A panelist cites OpenAI’s report that annualized revenue grew 70% in Q3, enterprise doubled into a majority of revenue, and the company was closing $11 billion at a $1.25 trillion valuation. Rory calls the reacceleration “stunning” if accurate and contrasts it with an earlier 18% quarter. A panelist also highlights Codex’s competitive progress against Claude Code.
  • Jason’s lesson is that developers have little loyalty and model comparisons are easy: the market keeps playing “ping pong.” He expects both companies’ revenue to rise radically over 12 months, but cannot predict their relative positions even 100 days out. A panelist identifies token-price elasticity as the next question: will 10%–20% price cuts stimulate enough additional usage to increase revenue?
  • Jason corrects his own agent-cost assumption. He had imagined $4–$5 monthly running costs, comparable with website-building products; persistent agents can instead cost $20–$30, with users demanding more work. He questions sustainability when compute is constrained and GPU prices are rising. A restricted agent becomes less compelling: “It’s just an automated workflow.”
  • Rory distinguishes short-term scarcity from eventual technological cost declines: “The question is, are you 2 years away or 5?” He does not know. Meta, he thinks, can carry the expense. Jason’s product requirement remains persistence—agents should work whenever needed, “24/7”—which makes the timing of cheaper compute consequential for competitors with smaller balance sheets.

7. Revolut wins the durability case while Fal wins Jason’s tactical vote

  • Zipline’s $1.5 billion raise at $21.5 billion is three times its January price. Jason admits he previously did not understand the African drone-delivery business but now sees an autonomous-delivery winner. Rory likes its distinctive position, Walmart contract, and Dallas deployment, but questions the arithmetic: three million US deliveries at perhaps $5–$10 each require enormous volume. The valuation depends on an infrastructure-scale outcome.
  • Fal doubles to $16 billion on more than $800 million ARR. Jason chooses it because the next two to three years’ demand trends look powerful, speculating they could flip the investment at $40 billion. Rory estimates third-party inference at roughly $5 billion, growing rapidly, but warns $16 billion may be nearer the end than the beginning. Jason’s 6% gross-margin aside is challenged; he admits he does not know.
  • Harry chooses Revolut’s $115 billion secondary valuation, up 53%, and sees at least 4X potential over time, with key-person risk around Nik. Rory’s mechanism is demographic: customers aged 19–25 will eventually need mortgages and more banking products. Today’s money-movement business could become Europe’s largest bank. Jason also considers Revolut and Zipline more enduring than Fal, whose industry may commoditize.

8. Inference theft creates a security market with unusually immediate losses

  • Jason reports two Fireworks breaches within roughly a month and describes bot swarms seeking Anthropic and OpenAI API keys. Stolen access can be monetized as cheap inference. His own Claude account, he says, had a $100,000 default spending allowance he had not noticed: criminals could consume that amount rapidly. “It’s the crime of the year.”
  • The exposure extends beyond one provider. Jason cites Wiz saying nine leading AI developers disclosed swarming incidents in October, and Unit 42 describing bills reaching $1 million before detection. Yet capacity scarcity limits customer alternatives after breaches. “Where’s Harvey gonna go?” Large contracts and difficulty replacing compute complicate the usual assumption that customers can simply switch.
  • Rory views CrowdStrike and Palo Alto’s roughly $400 million acquisitions as classic platform expansion into agent security. Established vendors buy adjacent products as customer needs change. He places Meta and OpenAI’s software-development acquisitions in the same category: the development lifecycle is being rebuilt, creating opportunities to acquire missing capabilities. The renewed venture M&A is welcome.

9. Legacy software needs a different operating tempo

  • On Dialpad buying Ada for $400 million in stock, Rory discloses his Dialpad investment and frames Ada as a strong pre-LLM chatbot facing Sierra and Decagon after a technology transition. Acquisition can make sense even after replatforming. Jason disputes calling it an exit without liquidity, while allowing it may be the best available outcome.
  • Jason says he initially resisted the “AI-native” distinction, believing capable CTOs could adapt. His view has changed because older organizations still cannot match newer teams’ shipping speed. Proposing an internal AI-native team next year repeats a plan they should already have executed. Rory rejects buying old customer contracts as an easy shortcut: inherited problems consume attention, and customers are rarely grateful.
  • Efficiency examples complicate attributing everything to AI: a panelist cites Cloudflare’s earlier 20% layoffs and Freshworks crossing $1 billion with 30% fewer employees. Jason cites Palantir near $2 million revenue per employee and 98%–100% growth. Rory adds that secular demand and $2–$5 million contracts can grow revenue faster than hiring capacity. AI may expose prior overstaffing as well as create new productivity.

10. Agents reward infrastructure and challenge human-seat economics

  • Jason connects Cloudflare’s acceleration from growth in the twenties to 39%, at almost $4 billion revenue, with agent activity; he sees similar agentic spending at Twilio. His proposed investing filter is categorical: “if it’s not an agentic beneficiary, I don’t wanna hear it anymore.” Agents increasingly determine how software gets bought, built, and used.
  • Atlassian’s warning on seat and subscription growth illustrates the opposite exposure. A panelist says the company survived earlier existential fears, but “the new trend is agents, not engineers,” challenging seat licensing. Jason questions interface-based defenses because agents do not value UX like humans. He similarly worries about Monday and HubSpot, correcting “terminal” to mean constrained prospects rather than certain corporate death.
  • Figma, around $10.2 billion after briefly falling below $10 billion, contrasts with its blocked $20 billion acquisition offer in 2022. Harry credits approximately 40% growth and strong free cash flow, attributing the latest hit to additional AI spending and gross-margin pressure. Jason adds the retention problem: depressed equity can create organizational “malaise” despite strong execution.
  • Harry asks what Jason would actually change. His answer is to own agentic design: deliver every page of an application at Notion-level quality, accepting lower margins and cuts elsewhere. He calls current design modes “a C-minus,” leaving room to compete. Harry agrees after watching designers complete work more easily in newer agentic tools than in established products.

11. California’s wealth-tax rejection becomes an argument about retaining builders

  • Rory finds California’s reported rejection of the wealth-tax proposal striking alongside the broader Democratic electoral swing. He attributes it partly to skepticism toward propositions and partly to the argument that departing wealthy residents would remove income-tax revenue. He says a separate measure preventing another attempt also passed. His reading is a preference for a centrist outcome rather than blanket enthusiasm for tech billionaires.
  • Jason adds the campaign-finance qualification: Sergey Brin and allies outspent unions 10 to one. Still, he says two relatively quiet billionaire founders were already planning to leave. His concern is a domino effect among active builders, though “not as big as they said” on Twitter. Even individually rational departures weaken California’s entrepreneurial concentration.
  • Anthropic supplies his wealth-creation example: he cites $41 million per employee on average, $216 billion attributable to current employees, and $160 billion in the pool. He says founders retain approximately 1%-plus each and are selling $264 million apiece. Rory adds that a company need not relocate for employees to make different tax-residency decisions; individual departures can still matter.