Predictions for 2026: Top Buy & Biggest Short | Why Salesforce Could Win & NVIDIA’s Challenges

Summary

• Founder and product of the year converge on Anthropic: Jason’s argument is that Claude 3.5 → 3.7 → 4 is the invisible engine of the entire app layer — “without this, we have no vibe coding… no Lovable, no Replit, no Cursor that really works,” and Cursor is “still 90% powered by Anthropic.” Rory, excluding Elon the way you’d retire a likely Ballon d’Or winner, agrees Dario “played a very steady hand” — growth faster than OpenAI and valuation convergence “even after OpenAI’s potential new round at 800” — while both dismiss his mass-unemployment warnings as “nonsense.”
• The stock bet both panelists actually like for 2026 is Salesforce: all-time-low ~5–5.5x revenues with 30% cash flows, where Agentforce co-attach to even 20% of customers lifts the stock without valuation risk. Jason: “Another $100,000 a year, but our agent will automatically go after all the customers your team didn’t follow up with? Sign me the f— up.” Rory: “You’re less likely to get a 3x, but more likely a 20% lift… it could go down 20%, but it’s not going down 60.”
• On the Mag-cap buy/short, the panel splits: Rory buys Google and shorts Nvidia (Nvidia may need “continued acceleration of the capex cycle”; Google has upside if the tide goes out), while Jason calls that a fool’s trade for 2026 specifically — “you just can’t make semiconductors from scratch that compete with Nvidia in weeks… there is no Base44 of GPUs” — and names Amazon the underperformer for the same not-enough-time reason. Rory’s concession: if OpenAI lands another $100bn, “bets are on for another 12 months… the tell’s not there” for timing a short.
• Jason’s IPO call is unusually precise: “four IPOs backloaded next year — SpaceX goes out first (Rory bets on June; Rory says $1.12T rather than $1.5T), Canva comes out of nowhere second, Databricks in the back half because it’s just time — it’s just the Series M — Anthropic at the end of the year.” OpenAI “probably should have gone first, but it’s burning too much” — mid-‘27. Rory’s structural worry: floating $50bn of a $1T company leaves “$950 billion of stock available to trade” with no in-the-money retail believers.
• The biggest 2025 surprise, per Jason: “for the moment there’s no ceiling on venture, which changes all the math” — trillion-dollar IPOs coming, Anthropic at $80–100M “the deal of the century,” Lovable two-to-eight billion in weeks, and a likely Chris Sacca/“lowercase” deal at Fuse Energy at $5bn. The corollary everyone now admits on circular deals: “We’ve all stopped caring… if it marks up my fund, great” — against Rory’s warning that financial innovation “works amazingly and it overshoots,” and marks aren’t real until $500bn+ deals price in liquid public markets.
• Fund of the year doubles as a recalibration of what seed returns even mean: Rory picks Index (“exits are the coin of the realm” — Wiz, Figma seed, Revolut at $75bn), Jason picks Neo on “aesthetics,” and Harry’s Hummingbird — a $100M fund returning $800M — gets Jason’s deadpan verdict that 8x is no longer elite: “in the age of AI, if you’re a seed fund and you bought 10% of Cursor,” an 8x looks like a rounding error.
• The 2025 monetization lesson: Adobe is the “worst offender” ("$5 billion of AI-influenced revenue… it don’t count unless it’s net new bookings"), the copilot was “the cynical whiteboard of January 2025 that failed up and down the B2B board” starting with Microsoft, and Notion is the rare company that earned a 2x price ($10 → $20/seat driving 50% growth at $600M) — Jason’s worry for GitLab, Atlassian, Monday, Bill, HubSpot is “not that they won’t be amazing AI products next year — it’s that they can’t charge more for it.”
• The consensus 2026 macro-political call: if unemployment rises 2–3 points for any reason — tariffs, the cycle — AI takes the blame because its executives “have admitted to the f—ing crime,” producing “a tech lash that makes what we’re dealing with now an understatement.” Harry: “Society will be terrified of AI” — and the operative investor logic, said literally by Elon: “If the robots are going to take over society, I want to be sure that I own the robots.”

Deep dive

1. Founder of the year is Dario — because Claude 3.5→4 made the app layer work

• Jason answers the founder question with a product: Claude “3.5 or 3.7.” “Without this, we have no vibe coding… no Lovable, no Replit, no Cursor that really works, no Gamma” — products that “existed for years” and “did not work” until Claude 3.5 late last year, “weren’t great until 3.7, and then 4 this year. And 4 changed our lives.” Cursor, he claims, is “still 90% powered by Anthropic.” Do you give the CEO credit for the product of the year? “Maybe not. I’m going to anyway.”
• Rory — grumpy by declaration (“I hate these events… I’ll be grumpy the whole time”) — lands in the same place after ruling Elon out the way the likely Ballon d’Or gets boring (“founder of the decade… it’s going to be Elon”). His case is Dario as executive, not pontificator: “he played a very steady hand,” went for “the sensible party,” and delivered growth faster than OpenAI with valuation convergence “even after OpenAI’s potential new round at 800.” “If you owned those stocks at the start of the year, this is the one you’d feel most excited about.”
• Both discount the doom talk: Jason doesn’t “love all those warnings about how 90% of the world’s going to be unemployed in several weeks”; Rory: “I do not think we’ll all be unemployed on Friday week. That’s all nonsense.”
• The additions: Harry’s Gwyn Shotwell, navigating “pretty challenging times geopolitically with Elon’s brand” while steering toward “the biggest IPO of all time” — “she may be the winner next year.” Alex Wang takes the Machiavellian award: “a 4x TVPI founder in terms of world impact, but a 20x in terms of returns — leave the keys with the rest of the team, done to perfection.” And Vlad at Robinhood — stock up 220%, nine products doing over $100M in revenue: “I wish half my founders could execute half this playbook.”

2. Fund of the year: Index’s boring perfection vs Neo’s aesthetics — and why an 8x fund no longer impresses

• Rory’s pick is Index, on the only metric he accepts — “exits are the coin of the realm”: early in Wiz (announced in March, not yet closed), seed investor in Figma, early position in Revolut, which just raised at $75bn. “Index just took that same old model and executed to perfection. I like that, because I’m a boring kind of guy.”
• Jason votes on “the aesthetics of venture”: Neo — first money into Cursor, plus an unclear company reference, run by someone who “24 months ago seemed to be having weird arguments with Gary Tan on Twitter” and then “literally hustled his way into Cursor, hanging out at MIT giving programming tests himself.” It also marks the rebirth of the accelerator — Neo, South Park Commons, HF0 — a year after “it seemed like YC had won the accelerator race.” His second vote: Creandum — Trade Republic at $15bn and another deal he says might not be a Revolut — “a phenomenal, premier dominant brand rivaling Index and Accel (spoken ‘Excel’) in a way very few have done in the last few years.”
• Harry’s seed pick — Hummingbird, a $100M fund returning $800M, first-ever investor in Lovable, very early in Kraken, plus a $3bn exit this year — triggers the episode’s best exchange. Jason, deadpan: “I don’t feel like I would get full kudos from my LPs for that performance today… I would have to do 8x to even get to speak at their AGM.” His serious point: “in the age of AI, if you’re a seed fund and you bought 10% of Cursor,” an 8x fund looks like a rounding error. “If this was the first pod of this series, I wouldn’t have said it.”

3. Lots of strategies work in venture — Benchmark, Thrive, Founders Fund, Google, Elad

• Harry’s Series A winner is Benchmark’s likely 21st fund — likely Manus, Sierra, likely Fireworks, likely Lora, likely Cerebras, and likely Lang Chain — and Rory turns it into the episode’s investing thesis: instead of super-sizing funds “to make the math work,” they “stuck to their knitting.” “X and Y strategy both work if done well and neither work if done badly… turns out there’s lots of ways to make money. You just have to be good at doing it.” (Hummingbird’s version, by their own statements: “picking psychologically damaged human beings and backing them to the hilt.”)
• Growth fund of the year is Josh Kushner’s Thrive — fund-size and vehicle expansion, the holding vehicle, the rollup play, a central OpenAI position plus Cursor, Databricks, Carvana, Revolut. Runner-up: Founders Fund deploying $3bn in under 12 months — “as an LP: what the f—, my deployment period has gone way down… ah, it’s SpaceX, likely Anduril and Stripe. Fine. Keep going.”
• Jason’s corporate investor of the year is Google: almost 10% of SpaceX, “perhaps 14% of Anthropic depending on dilution,” and the vast majority of likely Waymo — “double-digit ownerships, even if done in a circular way in one of them.”
• On individuals: Elad Gil, for raising a $3bn solo fund on “a perception of taste and style such that you become a preferred investor” (Harry’s footnote: “he does have 18 people”). Harry adds Lee Marie at Kleiner Perkins — Windsurf and Chronosphere, $2.5bn and $3bn, both liquid this year. Rory approves via his 20-year “great exit database” discipline: “it’s a raw capital dollar business… cold hard cash in early 2026 is an excellent outcome for all involved.”

4. Breakout companies: OpenEvidence’s perfect two-sided hit, Databricks’ wave, ElevenLabs’ $30-vs-$2,000 math

• Rory’s breakout is OpenEvidence: about 60% of academic-paper search traffic is doctors, and the product went “from nothing to 500,000 doctors out of about a million in the US in the space of one year.” The monetization is the beautiful part: a doctor types “my patient has this obscure disease” — and the buyer of that real estate is “the drug company who makes the $40,000 orphan drug that only has 10,000 customers in the US, and now you’ve put up your hand and said my patient is one of them.”
• Jason’s is Databricks, because riding the AI torrent “was all of our jobs” this year: Snowflake and Databricks both sit at $5bn, but Databricks is “growing and accelerating at 55%” while Snowflake’s AI revenue is small and its agentic strategy just starting. “If you didn’t ride that wave this year as a founder, you get a D-minus.”
• Harry names ElevenLabs — scaled to $400M ARR while standing off OpenAI — and Jason supplies the specimen: SaaStr London’s voice-of-God intros for ~200 speakers used to take two weeks and ~$2,000; Amelia retrained her voice in two minutes the morning of day one and did them all perfectly for $30. “That is why these apps are blowing up — it’s not the little step function we got before AI.”
• Rory is disappointed nobody said Lovable or Replit; Jason’s hedged rebuttal: personally “it was the breakout tool of the year — I spent 200 hours” — but “no one built a game-changing B2B app with nine figures of revenue on Lovable or Replit. I do believe it will come… they’re just at the very beginning of the journey.”

5. 2025’s surprises: talent wars, no ceiling on venture, and nobody caring about circular deals

• Rory’s shock was the talent wars: Meta “willing to just give a dude a hundred million bucks to show up,” buying a company for $14bn for the people and letting it “be a husk,” joking about buying Nat and Daniel’s venture fund — plus the whole Google–Windsurf saga. In retrospect the logic holds — “if you’re spending $73 billion on capex, spending $5 billion to make sure that the people using the capex know what they’re doing probably makes sense” — but the shredding of social conventions in six months blew him away.
• Jason’s: “for the moment, there’s no ceiling on venture, which changes all the math and calculations” — a trillion-dollar IPO next year, maybe two; getting into Anthropic at $80–100M “the deal of the century from a risk-reward outcome”; Lovable from two to eight or six (billion) in weeks. The old game — enter at 10–12 and let Bessemer’s Byron Deeter mark you up at 50 — is inverted: “now the sucker is wasting all the energy to have smaller ownership anywhere on the journey.” Exhibit: likely Chris Sacca (the captions say “lowercase”) — once sniping deals at $3–5M pre — leading a $5bn Fuse Energy deal.
• Harry’s: the bizarre Windsurf three days (“are they being bought? Are they not?”) and Nvidia’s $100bn into OpenAI kicking off the circular deals — “and all of us giving up on caring. If it drives the stock price up, great. If it marks up my fund, great.” Rory’s warning is worth keeping: “You won’t care until you do — and you’ll look back and go, the first ones were sensible.” Financial innovation “works amazingly and it overshoots”; the marks won’t be truly known until $500bn+ deals find an equilibrium price in liquid public markets.
• Jason’s quieter second surprise, as a vibe coder: autonomy plus reasoning. He started 2025 “a want-to-believer but skeptic” who couldn’t get anything to work; now — “we haven’t missed the boat as investors, because what agents can do, we just started. Outside of parts of coding, we just started.”

6. The 2026 stock game: Jason’s momentum six vs the Salesforce value bet

• Rory refuses to predict, with receipts: 2025’s actual top five tech stocks were Planet Labs, Bloom Energy, Opendoor, Oklo (likely — “you’ve got fission”) and Seagate — “utterly idiosyncratic, fairly unthematic… if you can barely nail the past, the best performing stock of ‘26 is a bridge too far.”
• Jason bets “$10 grand” that three of his big-six B2B names repeat in the top six: Palantir, Cloudflare, likely MongoDB (the raw is bleeped), Shopify, CrowdStrike, Snowflake — because the ones reaccelerating into AI tailwinds hold “for at least two, three, four quarters.” Precedent: AppLovin up 700% in 2024 and still up 124% this year; Palantir up 340% in 2024.
• Rory’s translation: that’s a momentum bet that works “until the AI lift starts to recede… at some point someone’s going to look up and say, oh my god, 70x, I should panic — it’s pennies in front of steamrollers.”
• The bet they converge on is Salesforce, “trading at all-time lows on a revenue multiple — five, five and a half times revenues” with 30% cash flows. Jason: half the customer base wants an agentic product that works tomorrow — “another $100,000 a year, but our agent will automatically go after all the customers your team didn’t follow up with? Sign me the f— up.” Rory: get Agentforce co-attach to 20% of customers and “you’re less likely to get a 3x, but more likely a 20% lift, and you’re not taking the valuation risk… it could go down 20%, but it’s not going down 60.”

7. The AI-revenue honesty test: Adobe fails, Notion earned the 2x, copilots flunked

• Jason’s worst offender is Adobe: “$5 billion of quote AI-influenced revenue… let’s be kind, I’m sure it’s true — but it don’t count unless it’s net new bookings.” Growing 8% while calling half your revenue AI-influenced “is table stakes.”
• His bottom six — GitLab, Atlassian, Adobe, Monday, Bill, HubSpot — share one worry: “it’s not that they won’t be amazing AI products next year. It’s that they can’t charge more for it.” Rory, an ex-Bill board member, pushes back that Bill has real AI room (invoice recognition, RAMP-style financial management); Jason, “a happy 100-NPS Bill customer,” holds the line: the question is whether they can dramatically lift ACVs. “I don’t know.”
• Notion is the counterexample: if the numbers are right, 50% growth at $600M, which Jason says only pencils if users moved from the $10 to the $20 seat — “I might be wrong, but I just don’t see any other way they did this… so many B2B companies want to double their pricing for AI; Notion might be one of the very few that earned it.” His own tell: a portfolio company shared a Notion board without the AI — “I can’t use Notion without AI. It is worth the 2x.”
• The biggest failing of 2025 was the copilot: “the worst of all worlds… the cynical whiteboard of January 2025 that failed up and down the B2B board” — starting, both agree, with Microsoft (“there wasn’t enough value”). Harry adds the bundling trap: attribute core-product dollars to AI now and you risk “a nasty renewal cycle” later — and if you bundle AI free without expanding revenue, “congratulations, you’ve reduced your operating margins by 10%.”

8. Buy Google, short Nvidia — or “you’re a fool to invest against Nvidia for 2026”

• From Google/Meta/Amazon/Microsoft/Apple/Nvidia, Rory’s pair is instant: buy Google, short Nvidia. Nvidia may need “continued acceleration of the capex cycle” at a hefty price; Google “still has its business and can gradually roll out AI in a more measured fashion,” with upside “if the capex tide goes out.” Microsoft, Apple and Amazon are “meh in the middle” — Apple to the downside: all-time-high multiple, sub-10% growth, no catalyst. His actions match: “I sold some of my Apple and I didn’t sell any of my Google.”
• Jason’s counter is pure timing: “It is not enough time for all this competition to come online… you just can’t make semiconductors from scratch that compete with Nvidia in weeks. There’s no Base44 of GPUs.” Everyone designing out their most expensive component — Amazon’s chips, Google’s TPUs, OpenAI’s own team — matters for 2027, not 2026. His twin call: Amazon, up only 2% this year, “will be the underperformer next year for the same reason.”
• Pressed by Rory on the real Nvidia risk — share erosion or demand erosion — Jason sees only one: OpenAI failing to fund its chip purchases. “But good news is Nvidia is going to give them the hundred billion. So that risk is off the table.” OpenAI’s stated position is that compute growth is one-to-one correlated with revenue growth — “whether it’s correlation or causation will be the debate. The market is going to solve for that problem in 2026. Everyone should feel pretty good about their 401ks.”
• Rory concedes the mechanism: “as long as the protagonists who believe it keep getting money, it will keep going” — watch the Oracle CDS (per likely Tomasz), and if OpenAI lands another $100bn “bets are on for another 12 months. Don’t try and make a clever market-timing position yet — the tell’s not there.” His standing rule on shorts: “You can be right conceptually and wrong in timing, and then you’re definitionally wrong… buying puts is the coward’s way of shorting — an even shittier way to make money.”

9. The 2026 IPO slate: four, backloaded — and a trillion-dollar banking problem

• Jason’s call, in order: “I think we see four IPOs backloaded next year. SpaceX goes out first” — Rory puts money on June and says $1.12T rather than $1.5T — “Canva comes out of nowhere second, because it’s not that AI-first… the numbers are there, do it now or risk looking obsolete. Databricks does it in the back half because it’s just time — it’s just the Series M. Anthropic does it at the end of the year.” OpenAI “probably should have gone first, but it’s burning too much” — sometime mid-‘27. Self-aware hedge: “I’ve proven with Harry I’m always too optimistic by a quarter or two, or sometimes two years.”
• Rory bets narrower — Anthropic and SpaceX out, Databricks/OpenAI/Stripe not — and names the problem bankers are “quietly wrestling with”: take a company public at ~$1T, float $50bn (5%), and six months later “$950 billion of stock is available to trade,” much of it held 10–15 years. Story stocks like Tesla and Palantir had retail believers who bought early and are in the money; a trillion-dollar debut has none, and “institutions are ruthless — when they hit their price target, they’ll be sellers.”
• His aside worth keeping: rereading the appendix to the Ashley Vance’s 2015 Musk biography, where Elon lays out why SpaceX shouldn’t go public — “one of the smartest first-principles thinkers on the planet had a really good argument against being public. If he’s going out, it must be purely cost of capital.”

10. The 2026 techlash: “they’ve admitted to the crime”

• The final question — does real AI-driven unemployment show up in the federal numbers by end-2026? — gets Rory’s sharpest reframe: it doesn’t matter whether it’s true. Politics doesn’t run subgroup regressions on BLS data; voters will know two things — unemployment ticked up, and “those guys building AI are saying that they’re causing it. They’ve admitted the f—ing crime. So if unemployment goes up for any reason — tariffs, random bank events, a big company goes bust — the drumbeat against AI will go up.”
• The next speaker converts mid-sentence (“I didn’t see that clearly until I started to speak”) and raises: “if unemployment ratchets up even two or three points, regardless of the reason — could just be the business cycle — you’ll see a tech lash that makes what we’re dealing with now an understatement.” And if hard numbers do land — “we lost half a million jobs last month in November 2026 due to AI” at the top of the NYT and WSJ — “it will become every single dinner table conversation. Society will be terrified of AI.” Rory’s hedge on the other quadrant: it may be “not as impactful in the short term as we think — but because we’ve confessed to the crime, we’re going to take the heat.”
• The gallows close: the three of them up on likely Elysium, and the objective someone once described to Rory — “If the robots are going to take over society, I want to be sure that I own the robots” — which, Jason notes, Elon has said literally: “I wish I could slow down robots and AI, but because I can’t, I’m all in.”

Verification Notes

• The raw captions do not establish the speaker of the final unemployment question or the immediately following agreement; labels are marked [Speaker?].