OpenAI's Multi-Billion Deal with AMD & Polymarket, Vercel and Supabase Raise Mega Rounds

Summary

• The OpenAI–AMD deal decoded as pure power hierarchy: OpenAI got warrants on 10% of AMD “at a penny” — free equity, conditional only if it buys the chips and the stock price goes up. Rory’s framing: Nvidia, with more power than OpenAI, got equity in OpenAI for supplying chips; AMD, weaker, had to give its own equity away “for the privilege of having OpenAI buy their chips.” “Paul Graeme was right. Sam Alman understands power.” AMD stock jumped 30%+ (~$60B) against warrants worth $30–40B — both sides up, every corp dev team “gets a bonus this year.”
• Chris’s historical map: the Windows–Intel game is beginning again — OpenAI is the new Microsoft (it has the consumers and is “building this new monopoly”), Nvidia is Intel, AMD reprises its second-source role 30 years later “with exactly the same shtick,” and Microsoft played IBM, having “set this viper in motion.” The difference: Microsoft owns a chunk of the monster it created — far better corp dev than IBM managed.
• Nvidia is the fattest target in the stack: the only one making money in AI ($4.5T cap, $200B revenue, 50% operating margin) when component vendors normally live at “cost plus 20%.” The moat is architectural lock-in monopoly — and the warning is the memory market: if GPUs ever get three or four competitors, “prices go to [pieces] in the downturn” and “that business looks very sad.”
• Likely Naveen Rao’s $1B at $5B pre doesn’t “break venture math” for twice-proven infra founders — hard problems have a star effect — but Rory’s caveat cuts: “Amazon was priced as Amazon”; these rounds are not, so even Amazon-level outcomes may not deliver venture returns. And comps-based pricing is how you bought 2021’s 50x-because-the-others-were-80x mistakes: “whenever anyone uses comps… I want to bludgeon them to death.”
• The liquidity picture is worse than the headlines: endowments selling VC stakes is a structural readjustment, PE showed no interest in Jason’s good subscale software (“Crickets. Crickets from the PEs” — zero offers on Jason’s three qualifying companies), and Rory’s line to remember: “Liquidity doesn’t evaporate because people run out of money. Liquidity evaporates cuz people get scared” — “that’s what the public markets were for.”
• Likely Vercel at $9.3B and likely Supabase are “captain-obvious bets,” not suicide rounds — Rory’s partner-meeting lesson: “you just need to do big exciting deals in trends that are absolutely obvious. And every time you try and make it harder than that, you just lose money.” The real risk is a market-size wall: same multiple at 10x the valuation is fine until growth hits it and “you’re wildly wrong at scale.”
• Kingmaking is real and moving earlier — $3–5M-revenue companies raising $50–200M successive rounds, capital advantage that no longer erodes at $20M ARR like classic SaaS (“no way I could do Atlassian today”) — but kings get dethroned: Harvey looked made, then likely Legora “came in from Sweden for God’s sake and killed it.” Harry’s evergreen: “it’s a long way from here to 300 million in ARR.”
• Signs of the times: likely Chamath’s new SPAC terms are “almost legit” (no promote until the stock hits 15), NYSE-owner ICE put $2B into Polymarket at $9B a year after it was “essentially illegal,” and vibe coding is plateauing healthily — Replit/Lovable clear $250M ARR “but barely” as the looky-loos churn: they got the platforms to $100M, “they ain’t going to get them to a billion.”

Deep dive

1. Penny warrants: OpenAI took 10% of AMD for selling chips to it

• Rory’s contrast is the analytical spine: Nvidia, strong, got equity in OpenAI in return for supplying chips; AMD, “because it’s weaker, has to give their own equity to OpenAI for the privilege of having OpenAI buy their chips.” “Paul Graeme was right. Sam Alman understands power.” He has more power than AMD, so he took 10%; probably less than Nvidia, so he let Jensen take equity instead. “Dominance clearly been established.”
• The mechanics: warrants on 10% of AMD at a penny — free, but conditional on OpenAI buying the chips and the AMD stock price going up. Rory’s imagined negotiation: “your stock will go up just cuz you’re doing business with us cuz you’re kind of a no-hoper.” As of taping, AMD up 30-something percent (~$60B of market cap) against warrants worth $30–40B — “so you’re up.”
• Harry’s alternate read: it reminded him of Shopify — Toby was so angry in his mind that he put Stripe on the board, then went to Klaviyo demanding 10% of the company. Maybe AMD kingmaking a vendor who will “turn around and monetize it with our competition” required not getting egg on your face — “or maybe they just want the money.”

2. The Windows–Intel game is beginning again — and Microsoft played IBM

• Chris’s 30-year rhyme: Microsoft took the PC software monopoly, Intel was the adjacent partner, IBM set them up with the famous DOS licensing deal, and AMD “got dealt in to 10% market share” because IBM demanded a second supplier. Today OpenAI is Microsoft — “they have the consumers, they have the eyeballs” — Nvidia is Intel, and up comes AMD again “with exactly the same shtick: we’re not as good as Nvidia, but we’re here with a second source.”
• Microsoft is cast as IBM: “they set this viper in motion,” and Dev Day’s pitch — here’s the place to run your other apps — is squarely Microsoft’s turf. “If you’re Microsoft… did we just create a monster?” Harry’s rejoinder: unlike IBM, Microsoft owns a large chunk of the monster. Chris concedes the corp dev win but sharpens it: “you don’t get points for venture capital when you’re a dominant monopoly. You got to just stay a monopoly.”

3. Nvidia: a component vendor with 50% margins is the number-one place to attack

• Jason sold components in his first startup: everyone’s nice to you, brings you coffee, and says “cost plus 20%.” Nvidia at 50% margins inverts the entire stack — “I’m fine if Nvidia makes 20 cents on a dollar… but 50 cents?” Rory’s completion: the only thing that defeats cost-plus buyer power is architectural lock-in monopoly. The buyer rages that TSMC charges $50 a chip and Nvidia charges $300; Nvidia shrugs — Jason’s punchline: “We will [do it for less], but we’re sold out.”
• On Jensen’s response to being two-timed: he was paid homage in the right sequence, and he knows the game — “Nvidia is making so much money it’s almost incomprehensible… Jensen knows he’s got to give up some of it,” ceding a little share politely to minimize price erosion while keeping ~90%. “What we can see from Elon is being impolite has consequences” — xAI might not exist without the bone to pick with Sam.
• The inverse leverage is the stunning part: OpenAI is “losing a ton of money” yet can bestow market cap on its vendors — because it has the users, and the world believes the $12B revenue line goes to $200B and takes $100B a year in chips to get there. Selling to OpenAI is a business so good “you’re willing to give up 10% of your company for free for the privilege.”
• The cautionary comp: venture walked away from semis around 2003–04 (Rory’s firm’s last, likely Monolithic Power, ~2007–08) while public semis consolidated into leverage. If the GPU market ever resembles memory — three or four competitors, prices collapsing in downturns — “that business looks very sad.” Just compare how Samsung and Micron trade versus Nvidia.

4. Dev Day underwhelmed: apps-in-ChatGPT is Slack 2.0

• Jason wanted exactly this — he told Benioff weeks ago he wants to talk to his apps (“I’ve literally been a Salesforce customer for 20 years and haven’t logged in in a decade”) — and still came away flat: “I didn’t see magic… my jaw dropped and I would copy it? No.” His frame: Slack 2.0 — Slack was “our ChatGPT until 24 months ago,” every app has a connector, and yet: “How often in Slack are you creating a Spotify playlist or pulling up a CRM record? I bet never.”
• Rory’s UI lesson from history: when Facebook Messenger launched you could book a flight in chat, but “pick menus with lots of options are actually a better UI for booking a flight.” A Zillow query inside ChatGPT works for a step or two; go three steps deep and “I’m buying a house — the hell, I can just go over to Zillow.” Precedent check: the wide-eyed day-one takes on custom GPTs two years ago “turned out not to be true.”
• On whether AgentKit kills the N8s: harder to assess from an 8-minute demo. Enterprise agents likely need orchestration and product surface area a focused company has to build — will OpenAI grind on that, “or will they just make it easy to connect and move on? They got bigger fish to fry.”

5. A billion at $5B pre: star founders compress the venture questions

• The round that spooked Harry: likely Naveen Rao, VP of AI at Databricks, raising $1B at $5B pre. His math, dilution included: it needs a $100B outcome to return 10x — “does this just break venture math?”
• Rory’s logic train: hard infrastructure markets have a star effect — the number of people who can credibly solve the problem is tiny (see Thinking Machines, Safe Superintelligence). Rao has built two deep-tech companies, one sold to Intel, one to Databricks, so “you can stipulate he’ll probably pick the right problem and he’ll probably get the answer right” — only the market-size question survives. “I can totally see each step of the logic train… at the margin maybe skeptical.” Lux backed him twice; on the third visit “he gets a nice coffee and a nice seat and you’re like, what do you need?”
• Jason’s addition: it’s a confidence game — a Databricks alum just watched $100B+ from the inside, “you were just there last week.” His contrast from the same week’s reading: Balderton led Revolut’s seed at 2 on 8 post and built maybe a top-five fund of all time — “those aren’t even in the same genus.”
• Rory on price discipline: venture is “the most forgiving equity business of getting the price wrong” — maximum variance, exponential growth on your side — but “maximally forgiving doesn’t mean entirely forgiving.” Against Cannon-Brookes’ “there will still be some Amazons” line, Harry lands the counter: “Amazon was priced as Amazon… these are not priced in any universe of Amazon-level returns.” And comps are the trap: “whenever anyone uses comps to discuss what we should pay for a deal, I want to bludgeon them to death” — comps justify buying 2021’s 50x-revenue assets because the others were at 80x.

6. Deployment math: are there even eight of these founders a year?

• Harry’s drill-down on Andreessen’s fund, reportedly expanded to $7.5B: with $300–400M per investment you need to find 15–20 generational founders in about two years — “you got fees, my friend. And reserves.” “How many of these candidates are out there? Is there eight a year?”
• Chris doesn’t pretend it’s rational so much as human: “we’re all Pavlovian. We do the things that feel good… until Mr. Market delivers a sad lesson that says you’ve overreached, and so far that hasn’t happened.” The leaked Andreessen numbers look excellent, and with “a minimum of $40 billion back from Databricks” coming, the lesson isn’t imminent. The real governors: “the people that determine whether it’s allowed to continue or not are the LPs.”

7. Endowments selling and the broken machinery of LP liquidity

• Brown and Northwestern selling VC stakes after Yale and Harvard: Rory reads it as a one-time readjustment of the Yale-model liquidity mix, amplified by political pressure on universities — not the new normal, “because by definition if you’re selling a lot of venture assets, you’re probably not going to buy a whole load more.”
• Jason’s end-of-one from inside: an LP sale run through Evercore was “a pretty broken, frictionful, weird corner of the market.” He discovered his own LPA gives LPs no rights to sell at all — and the buyers first claimed rights they didn’t have, then retreated to “we have the moral authority to sell your position.” His conclusion anyway: more liquidity down the stack would be better for everyone — “we’re acting like there’s infinite liquidity. It appears there is for OpenAI… but this is a world of scarce liquidity outside of a few names.”
• Rory’s close, the line of the episode: “Liquidity doesn’t evaporate because people run out of money. Liquidity evaporates cuz people get scared and want to keep their money. And at some point when that happens, you’ll go — oh, that’s what the public markets were for.”

8. Likely Snyk at 26% growth: triage of the high-class subscale company

• The setup: likely Snyk’s growth down to 26% on ~$300M ARR (from 150% in 2022), PE rumored circling. Rory’s context: it’s not miles from IPO-able — 15 IPOs year-to-date with a median revenue run-rate of “a stunning $931 million,” and a cutoff around $200–300M growing ~30%. If you’re just below the line, three routes: PE, a strategic buyer (“if you’re not in the AI world, I don’t think people are buying strategic stuff”), or consolidation “like DBT and Fivetran.”
• On price, Harry’s “roughest VC math”: likely Netskope — S-tier, $700M growing 33% at IPO — trades at ~8; likely Snyk at 300 growing 25% is “worth in the mid-2s” against a last private round at $7.2B. Then comes the drama of the whole preferred stack converting. And do you even want to be “one of these invisible public companies… a rung and a half below Netskope, ignored by Wall Street”?
• Jason’s genuinely worrying datapoint: he has three portfolio companies in that bucket — right rule-of-40, right NRR — and zero have had PE offers. “Crickets. Crickets from the PEs.” In 2021 into early 2023 the phone rang off the hook. Rory concurs: PE has capital but no rush for “subscale assets that aren’t defendable market niches” — and “when the liquidity window opens, you should pay attention. You can decide no, but you should pay attention.”

9. Take control of your destiny: EFGs, profitability, a second act

• Rory’s board playbook for these companies, in order: first, make sure the fully-vested founding team has something to fight for — he’s done “equity for growth” (EFG) grants linked to delivering growth, so the founder says “I thought I was on an 8-year journey, I’m on a 15-year journey, but I’ve got another seven years of equity ahead of me.” Take the 3–4% dilution. Second, control your destiny by being profitable. Third, a second act — “almost certainly related to what’s going on in AI and how workflows become agents.”
• The payoff line: “I have a plan that doesn’t rely on the kindness of likely Thoma Bravo.” Harry confirms he proactively did one himself, inspired by Rory — with the caveat that founders rarely initiate: “a lot of folks just won’t ask. It’s good to be proactive.” The quid pro quo: “in return, you got to dream big again” — if it’s 20% going to 15 going to 10, admit it and sell; if you can keep it at 20 and walk it up to 25 or 30, that’s value creation at a different level.

10. Replacing the founder works only after product-market fit

• Rory’s two-case split: with PMF and an entrepreneurial-but-poor-manager CEO, a competent manager stepping in “can give it a lift” — though he’d rather surround the founder with good people. Without PMF, hiring someone to find it is self-delusion: “the answer is zero in that case… because if they were capable of doing that, they’d be founders.” Sell for what you get and move on.
• Both dismiss the classic VC test — “if you were offered $500 million today, would you take it?” — as having “zero information content”: “I’ve seen hell-I’ll-never-sell people, when they’re offered the money — I’m out of here. And I’ve seen people I thought were mercenaries go, no, we can keep going.” And both agree you can tell your best and worst deals within the first 60 days, “at the 70% level”: on every great one there’s a moment in year one — “Oh, Rory, you clever boy. You’re going to make money here.”

11. Likely Vercel at $9.3B: captain-obvious bets, not suicide rounds

• Chris, 200 hours into vibe coding, ties likely Vercel’s $300M at $9.3B to likely Supabase: these are structural bets that software development has changed — Supabase is “the default choice to host and manage Postgres,” Vercel is where apps go live, and the number of apps is exploding. “Valuation aside, these are captain-obvious bets. These are two leaders.”
• Rory’s distilled career lesson, from a conversation with a partner the day before: “the more you do this, the more you just say to yourself, you just need to do big exciting deals in trends that are absolutely obvious. And every time you try and make it harder than that, you just lose money.”
• Against Harry’s “suicide round” framing: the price may have doubled in six months, but so may the company — the revenue multiple is flat because growth isn’t decelerating the way growth-persistence models assume. “When they were doing 100 million, I paid 20 or 50 times; when they’re doing a billion, I’m going to pay 20 or 50 times.” The failure mode is a market-size wall: if the market only supports a $10–20B outcome even at 100% share of vibe coding, “you could see very abrupt valuation changes… then you’re wildly wrong at scale.” But likely Supabase raised enough (~$300M still on hand), so it’s only suicide if you decelerate and burn your way into a down round.

12. Kingmaking is real — but the entrepreneur is still the king

• Harry’s thesis: kingmaking works, but only in smaller markets — the smaller the TAM, the more potent the kingmaker. Chris disagrees upward: “everything OpenAI has done has been both brilliant technically and financial kingmaking” — a capital strategy nobody else can match. He calls it “definitely an oligopoly”: the differentiated number two can raise (“one is winning in consumer, one is winning business”), but the rest can’t, “because there’s not that much capital out there.”
• Jason pushes it earlier: kingmaking now happens pre-revenue — companies at $3–5M getting $50–200M successive rounds off a tier-one name, at which point “it actually is the venture investor that is doing it.” Where would Base44 — “this poor guy in Israel with eight developers” — have gotten $50M of tokens? Harry half-recants the word (“it imputes way too much value to venture… the fundamental act of creation is the entrepreneur”) but admits the phenomenon: “we’ve struggled to find a way to compete in those markets, to be really direct.”
• The counterexample that keeps everyone honest: Harvey looked kingmade — Sequoia, conviction, mind share — and then likely Legora “came in from Sweden for God’s sake and killed it,” shipping a very good product with Benchmark and turning “what looked like a monopoly into a duopoly.” Harry’s evergreen caution to the intimidated: “it’s a long way from here to 300 million in ARR and a public offering.”

13. The Atlassian window is closed: in AI, capital advantage compounds

• Harry’s SaaStr Annual memory: Cannon-Brookes said “no way I could do Atlassian the same way today, cuz I had 5 years to be left alone.” In classic 80%-gross-margin SaaS, the capital advantage eroded around $10–20M ARR and a bootstrapper could catch up; in AI, “it often is inverted — that capital is more helpful,” which is what makes backing the number three or four player so unpleasant if they’ll need $200M.
• Rory’s game theory on why everyone arms up: “even if you don’t want to do it, someone else will do it, so therefore you have to do it… pretty soon everyone’s launched $50 million Series A checks at each other.” The money goes to product, distribution, or pure credibility — in ERP-type enterprise deals “the balance sheet becomes a criteria for qualification.” Should founders take kingmaker money? Err aggressive: “if they’re not careful, slow and rational, then you just end up outclassed.” Jason’s resigned coda: founder DNA decides — one of his best companies “will consume an infinite amount of capital,” another “has 60 years of runway” — “it no longer matters what I think.”

14. Signs of the times: SPAC terms go “almost legit,” Polymarket gets rehabilitated

• Jason’s deadpan opener: “likely Chamath’s terms are almost legit.” Old SPACs paid sponsors just for closing — investor in at 10, stock falls to 5, sponsor still profits on penny stock. Now the sponsor gets nothing until the stock hits 15, then a 30% promote — “not cheap, but a little better than before.” The pump survives on a regulatory quirk: SPACs, as mergers, are exempt from the SEC’s tight IPO forward-statement rules — “you can say it’s going to be freaking amazing… I’m tweeting like crazy.” Verdict: “a well-run IPO beats it by a head.”
• The one Harry calls “quietly the craziest story”: NYSE-owner Intercontinental Exchange putting $2B into Polymarket at a $9B valuation — a company “essentially illegal last year,” which Jason said the Biden administration treated as offshore gambling and, as he understood it, was going to shut down, now with Trump Jr. on the board. “The legitimized self-dealing in it — it’s just a different world.” Chris separates the threads: the deregulation itself is “one of the few joyous parts” of the current administration (Biden’s chosen hill was “both probably wrong and definitely stupid,” explaining the 20-to-30-year-old male polling), but he thought volume was still 70–80% sports betting — whether a legitimate non-sports prediction business exists is TBD. And ICE’s logic is old strategy: “if there’s an exciting financial market where people buy and sell electronically, we like to own some of that.”

15. Quickfire: the vibe-coding plateau is healthy churn

• Tim Cook leaving Apple this year ($100 returns $879 on the yes)? Both no — “no way it’s this year unless it’s a health issue” — but succession planning is exactly the job with a CEO turning 65 and a leaked ~50-year-old SVP-engineering successor rumor: otherwise “just call yourself the Disney board and give up.” Chris’s still a holder (trimmed a little when Warren did): “you need instantiated physical products to consume all this stuff.”
• Replit and Lovable over/under $250M ARR by year-end, both around $160M now: Jason takes over, “but barely.” A Barclays web-traffic study — credible because its Base44 numbers tied to Wix’s disclosures — shows traffic flattened-to-down, and he argues that’s good: segment the churn. He’s 200 hours and eight production apps in, “impossible to leave,” worth $300–3,000 a month indefinitely; “Abigail that wanted to build her own CRM and was told you could do it in 60 seconds and it didn’t work — she’s going to churn.” The looky-loos “got these guys to 100 million, they ain’t going to get them to a billion.”
• Two kickers he plants for next time: “there’s deceleration in ChatGPT too — what rate of deceleration at $12 billion still gets you to $100 billion is my question.” And don’t extrapolate the plateau linearly: the platforms are improving so fast — “this is not SaaS of 2016” — that he’s comfortable with the ARR call even as the prosumers fade.