Nvidia's Historic Quarter, SaaS Comeback, Bessent vs Druck, America's Debt Crisis, Cancer Vaccine

Summary

• Nvidia posted the most profitable core-business quarter in public-market history and shredded the AI-capex-bubble narrative. Revenue was $96.2B against Wall Street’s $92B, up 106% year over year, with $60B of net profit at 75% gross margins. The panel’s real lead was guidance of 70% growth next year versus the Street’s 45%, a number that “would have been higher, but they said they were supply constrained.” Even after the roughly 9% pop, Nvidia was said to trade at “only… 12 times earnings,” with Polymarket pricing 79% odds that NVDA ends the year as the world’s most valuable company; its current value was cited at $5.5T.
• The “SaaS apocalypse” was overplayed — but mainly for horizontal systems of record. Salesforce ripped 20%+ on an adjusted-EPS blowout ($5.90 versus $3.27) and a $46B full-year guide, vindicating Chamath’s May 15 bottom call (+43% since); revenue was $11.3B, up 11%, and its Anthropic investment helped the reaction. Chamath’s thesis is that agents now need contextual training data held by systems of record, so the “large monoliths” — Salesforce, Workday, Oracle and SAP — “hold an incredibly special place in the ecosystem if they do it right.” Jason and Friedberg describe themselves as short vertical SaaS, which Friedberg says is mainly workflows and processes; Sacks says it is case by case.
• Benioff’s play, decoded by Sacks: risk disintermediation to ride the wave. Salesforce is integrating into Claude, ceding the front end so agents can unlock the platform’s trapped value — “the average user is never going to find all those things, but the agent will.” The generalizable lesson: “simple extrapolations of the future based on these trends don’t work out,” whether “all software is going to zero” or “knowledge workers are all going to lose their jobs.”
• Jensen is “speedrunning open source” — reportedly $12B for Hugging Face plus a roughly $6B Poolside deal — and the panel sees a convergence endgame. After Sam Altman’s AMD announcement and Nvidia’s response around inference chips, Nvidia is moving up the stack. Sacks says Poolside has both a model and an agent harness, though the announcement was unclear; Nvidia could ultimately offer an entire stack. The broader thesis is that customer and supplier roles are melting together: companies will own silicon, models, clouds and data centers, with Nvidia’s neocloud business described as roughly matching the $24B it gets quarterly from hyperscalers.
• America must refinance $10T of debt in the next 12 months at a 5.2% 30-year rate against a 3.4% average cost — and Friedberg sees Druckenmiller’s op-ed as cover for Bessent. Friedberg’s math: each 1% rate rise costs 1.25% of GDP in annual interest, and Bessent has at most roughly $1T of buying capacity against $10T of refinancing, so Congress and the president must address spending. Chamath’s rubric: “yield goes up, trust goes down,” and a 6% 30-year “is the beginning of a death spiral.”
• Chamath forecasts an inflation-driven political reaction between 2026 and 2028, followed by Social Security and state-finance stress around 2030–2032. Friedberg notes that at 2019 spending levels “our federal government would be making a profit paying down the debt”; Sacks blames a tragedy of the commons and poor media coverage for blocking cuts. One panelist’s escape hatch is AI-driven growth: “if we hold back AI… then we’re cooked.”
• The Druckenmiller AI-authorship flap pitted Sacks and the pro-tool argument against Jason’s disclosure standard. Sacks said Druckenmiller’s take was consistent with decades of views and urged people to focus on their portfolios rather than AI-detection artifacts. Jason issued a standing disclosure that his publications, including tweets, may run through AI for fact-checking and editing, but alone objected to undisclosed AI drafting as “the lip-syncing of writing.” Separately, Sacks dismissed the Ratcliffe-Moscow invasion-warning story as neocon “threat inflation” and repeated that Zelensky — out of air defense, exports and money — “should make a deal.”
• Friedberg challenges Moderna’s roughly $20B-to-$60B “cancer vaccine” rerating as an expensive commercialization of a technique rather than a wholly new breakthrough. Personalized neoantigen immunotherapy is “much more of a technique than a drug,” built on decades of research, much of it publicly funded; he contrasts Moderna’s stated $500K price with roughly $50K custom-peptide treatments he says are already available through clinics, including in Montana. He expects cheaper global proliferation, alongside CAR-T ($1M, with “incredible” efficacy in multiple myeloma) and early detection through Grail’s Galleri test.

Deep dive

1. Science’s grant apparatus punishes heterodox thinkers — and that’s why progress stalled

• Friedberg, back from interviewing Eric Weinstein and separately speaking with Kratsios (name unclear in the transcript), frames Weinstein’s outcasting as structural: call string theory unproved and propose an alternative grand-unification theory, and “you don’t get grant funding, you don’t get tenure… you have to follow the mainstream in science or you’re outcast.”
• Chamath’s COVID analogy: “It’s like Jay Bhattacharya being like, ‘Hey, COVID may have come from a lab leak’” — branded misinformation when “everyone was starting with a theory on where COVID came from.”
• Friedberg’s contrast: general relativity “was completely outlandish. But people still studied it, assessed it” — versus today, where “there’s probably a lot of incrementalism in 2026 that didn’t exist in 1926.”

2. China cheers its robots while America catastrophizes — and agents go always-on

• Jason called the robot spectacle a psyop; Chamath’s framing was that the CCP is “brilliant at PR”: “they got a bunch of Chinese people to fill a stadium and cheer for AI and robotics… and then we got Dario saying all jobs are over.”
• Friedberg supplied the numbers: AI optimism in China polls over 80% (“more beneficial than harmful”) versus roughly 30% in the U.S. Chamath’s conclusion was that the U.S. is ahead of China in every category except optimism, which he called the biggest risk to winning the AI race.
• Chamath’s Optimus tease: Elon showed him an unreleased video (“Is this CGI or is this real?” — “No, this is real”) and he called Optimus “the bestselling, I think, product.” Sacks’ caveat stands — the hard part is novel conditions and physical interaction, the whole folding-laundry problem — and Friedberg concedes, “You kind of nailed it. I don’t want to say any more.”
• Sacks has been “Grokbot maxing” to his usage limit: unlike OpenClaw or Hermes desktop harnesses that stopped when the computer slept, “it’s always on, it’s in the cloud.” Multiple agents beat one because “the agents develop more context and expertise” as a swarm. Jason pitched multiplayer agent rooms with humans in the loop: “that will make this thing go so freaking viral.”

3. Nvidia’s quarter shreds the capex-bubble narrative

• The numbers: $96.2B revenue versus the Street’s $92B, up 106% year over year — “a billion a day in revenue” — with $60B of net profit at 75% gross margins. Chamath said that, excluding one-off items, it was the most profit ever generated by a company in history.
• The real lead was guidance of 70% growth next year against expectations of roughly 45%, a number that “would have been higher, but they said they were supply constrained.” The takeaway for the broader market: “this AI capex is going to continue well into the future. It’s got real legs.”
• Despite the roughly 9% pop toward all-time highs and a cited $5.5T valuation, “Nvidia is only trading at 12 times earnings”; competition was identified as the market’s concern. Polymarket put 79% odds on NVDA finishing the year as the world’s most valuable company.

4. Jensen’s open-source land grab and the everyone-does-everything endgame

• Jason highlighted reported deals involving Hugging Face ($12B) and Poolside (roughly $6B), about $20B in total and less than 1% of Nvidia’s market cap. Sacks’ framing was that Sam Altman “poked the tiger” with the AMD deal and that Jensen is now “speedrunning open source” — potentially selling customers “your entire stack.”
• Friedberg questioned whether the Poolside deal was being characterized correctly, saying roughly 100 engineers could be moving to Nvidia to work on an on-premise version of Cursor or Claude Code. Sacks added that Poolside has both a model and a coding-agent harness, but said the announcement was unclear about exactly what Nvidia bought.
• Sacks’ larger thesis is that the delineation between customer and supplier “is melting away.” Hyperscalers are spinning their own silicon, so Nvidia can spin models, host inference and become a cloud provider. Chamath said Nvidia’s neocloud business was roughly equivalent to the $24B it gets from hyperscalers in the quarter. The endgame is that “every company’s going to do everything,” which Sacks said could benefit the broader ecosystem.

5. Chamath’s victory lap: three phases of AI, and why systems of record win

• The May 15 call — Salesforce “meaningfully oversold” — was replayed on-air: CRM was up roughly 43% since Chamath’s bottom call and more than 20% on the print, with revenue of $11.3B up 11%, adjusted EPS of $5.90 versus $3.27 expected, and a full-year guide raised to $46B. Jason also attributed part of the reaction to Salesforce’s Anthropic investment.
• Chamath’s framework: phase one was models (“think of a model as a brain”), phase two was harnesses (“giving a brain a pair of eyes and hands and a notebook for memory and a keyboard”), and the next phase — emerging in the enterprises he works with — is training that autonomous agent “to be a lawyer or be a customer-service rep or be a sales agent.” That requires a ton of contextual information. The people who control that context, the large systems of record, “hold an incredibly special place in the ecosystem if they do it right.”

6. Friedberg’s build-vs-buy conversion — and the vertical-SaaS short that stands

• Friedberg’s story, told on Benioff’s earnings call: his company built a CRM over a weekend with Claude Code and Cursor, then drowned in feature requests, security, access and data-repository work. It realized that ROI lives in software unique to its vertical — plant-breeding tools — rather than “recreating a bunch of software that other people have already made.” Benioff’s relentless sales push (“Are we up yet? Are we up yet?”) closed him.
• The refinement of the meme: “it’s more of a vertical SaaS apocalypse” — horizontal platforms such as CRM, Gmail, Slack and Excel survive, while single-vertical workflow software is the area Friedberg thinks may be vulnerable.
• The live disagreement: Jason asks Sacks, “Are you long vertical SaaS or short it like Friedberg and I are?” Friedberg argues that vertical SaaS generally does not provide a system of record; “their workflows are processes,” and he wonders what survives. Sacks says it is case by case.
• Chamath’s read on the investor or fund manager referenced in the discussion (name unclear) is that the SaaS short may have been a leverage trap: the fund needed liquid shorts against a large net-long position, then worked backward from that constraint to the narrative.

7. Sacks: Benioff risks disintermediation to ride the wave — and beware simple extrapolations

• His long-held line, now vindicated: “I do not believe that core systems of record like CRM are going to get ripped out and replaced with something vibe-coded.” Enterprises want certainty, compliance and professionally managed software with more than 20 years of bug reports, not probabilistic agents guarding the source of truth.
• The Anthropic deal decoded: Salesforce integrates into Claude, letting the AI own the front-end relationship and access all data, workflows and actions. Benioff is “willing to risk disintermediation in order to ride this AI wave because he knows it’s what the customer ultimately wants.”
• From Sacks’ conversation with Benioff that morning: there is enormous trapped value in the platform — “the average user is never going to find all those things, but the agent will.” The agent can propose actions until users “just click the always allow button,” reducing the need for highly paid consultants.
• The prescription for SaaS founders is to build the best agent interface — great APIs and a great CLI — and relax control of the customer relationship. The meta-lesson is that “simple extrapolations of the future based on these trends don’t work out”: “AI agents can code, therefore all software is going to zero” failed, and “AI can do knowledge work, therefore knowledge workers all lose their jobs” ignores how workers can lean into the tools.

8. The $10T refinancing wall

• The setup: the 30-year hit a 19-year high at 5.3%; Bessent doubled long-bond buybacks from $2B to $4B and was reportedly considering more buying to “put the fear of God into traders” shorting the long end. Druckenmiller’s WSJ op-ed opposed manipulating prices instead of fixing the spending that drives yields.
• Friedberg’s math: COVID-era 30-years were around 1.7% (“the federal government should have rolled all of its debt into 30 years”); today the curve runs from 3.8% at one month to 5.2% at 30 years, the average cost of the $40T stack is 3.4%, each 1% rate rise costs 1.25% of GDP in annual interest, and $10T must be refinanced in the next 12 months against at most roughly $1T of buying capacity: “even if he maxed out… that’s only a trillion of buying and then he’s got to turn around and sell 10.”
• Friedberg called Druckenmiller, Bessent and Kevin Warsh a pretty unprecedented “triumvirate” and doubted they were even allowed to talk. In his view, the note provides “coverage for Bessent,” saying the yield curve is not Treasury’s to fix: “He can’t be Atlas… You guys have to act.” The burden sits with Congress and the president.

9. Chamath: yield up, trust down — and hyperscalers as America’s balance sheet

• The tactical mechanics: Bessent’s buying is yield-curve suppression — bid bonds, compress the implied yield, so new issuance marks against 5.1% instead of 5.3%. But the disease is congressional: debt is growing at 7% against 2–4% GDP growth, and “it’s not a Democrat or Republican problem.” Chamath’s rubric for the long bond as a trust gauge is: “yield goes up, trust goes down.”
• The awkward timing: this hits during an AI buildout when, a century ago, the U.S. government was the balance sheet for industrial revolutions. It cannot do that now — “thank God we have companies like Nvidia and Google and Microsoft and Meta and Amazon… these guys are putting the entire U.S. economy on their backs.” Chamath said the president had found roughly $2T through trade deals, but called that only a short-term Band-Aid because nation-state processes are slow and meandering.
• The line in the sand: a 6% 30-year “is the beginning of a death spiral. It’s not going to be immediate, okay? So don’t freak out, but it is the beginning of some extreme pain. And that pain will last years.”

10. Why nobody cuts: tragedy of the commons, the media and the DOGE lesson

• Sacks’ structural diagnosis: 435 House members, 100 senators, no line-item veto, and everyone protecting their programs — “no one person controls the spending.” Trump “did back DOGE to the hilt,” but after Elon reached the point of cutting one agency or subagency, the backlash included accusations that he was killing millions of children and Tesla dealerships “getting firebombed.” Sacks blamed media coverage that “does absolutely nothing to report the story accurately.”
• Nor does Sacks see a ballot-box fix: “the political energy in the Democratic Party is all toward these massive new spending programs that the DSA wants.”
• California high-speed rail became the tell — Sacks called “a trillion dollars for a train from San Francisco to Fresno that costs more than an airline ticket” idiotic and asked, “Why is no one even just saying turn it off?” Brightline in Florida was offered as a private-sector contrast, built at a fraction of the cost.
• Friedberg’s root-cause chain: federally underwritten student loans drove administrative costs up sixfold and tuition up 8% a year, compounding for 30 years; he said similar dynamics occurred in housing and healthcare. “The core root of inflation in this country… is government spending.”

11. The breaking-point timeline: political backlash by 2028, state stress by 2032

• Friedberg’s counterfactual: at 2019 spending levels, “our federal government would be making a profit paying down the debt.” Emergency COVID stimulus was normalized into permanent spending, as happened with some post-2008 stimulus.
• Chamath’s forecast is that inflation and declining affordability become acute between 2026 and 2028, pushing voters toward someone promising free housing, free transportation or other state-provided benefits. Around 2030–2032, he expects Social Security to lack money to pay obligations and states to face bankruptcy or restructuring because of unfunded pensions and debt.
• Do voters connect deficits → Treasury yields → 6.73% mortgages? Friedberg answered, “No. No. But people want to blame someone.” The first human instinct is who, not what’s the solution: “the fire is government spending, and the fuel is more government spending.”
• Friedberg’s rebuttal to a clean political reset is that there is no default button: the debt is too interconnected, and “you would go bankrupt overnight.” Structural change, he said, comes only at acute moments such as “9/11, the GFC, and COVID.”
• The panel’s AI escape hatch was that only exponential growth could create enough economic expansion to grow out of the problem; the warning was, “if we hold back AI by creating some new regulatory apparatus that takes years to approve new model releases, then we’re cooked.” Jason’s coda from Texas was that Austin housing is down 27% from the 2022 peak because leaders allowed more building — “it just takes leadership.”

12. Did Druck write it? The AI-authorship fight

• Detection software pegged the op-ed as roughly 90% AI-written; the Journal said its use was acceptable. Sacks argued that Druckenmiller is an extraordinary investor and urged critics to read the substance: the relevant questions were “how does this affect my portfolio… my 401(k),” not em-dash forensics.
• Sacks’ defense was that Druckenmiller’s take is consistent with decades of positions: “you got to be pretty dumb these days not to use AI to help you write, and he’s definitely not dumb.” If he was plagiarizing, Friedberg said, “he’s plagiarizing himself.”
• Jason then issued a standing disclosure: anything he publishes, including tweets, has probably been through AI for fact-checking, line editing and argument refinement. But his dissent remained that undisclosed AI drafting is “the lip-syncing of writing… like I went to see Adele and she’s lip-syncing”; proofing, research and Grammarly are fine, while full generation without disclosure “feels icky and insincere.” He would “wonder if it’s actually his opinion.”
• Sacks’ reductio was, “Don’t use Excel. Only go to the guys that do hand calculations.” The counterargument was that the disclosure boundary is nebulous for AI just as it is for Photoshop, autotune, synthesizers or calculators; AI can be “like all other tools a magnification of human creativity, ingenuity and potential.”

13. Ratcliffe’s Moscow day-trip and Ukraine’s grim math

• The facts: CIA Director John Ratcliffe made an unannounced same-day trip to Moscow — the first by a sitting director since Bill Burns in November 2021, three months before the invasion. The WSJ reported that he warned Russia against attacking NATO amid intelligence that Putin might test NATO’s resolve “in the next few years”; Polymarket put a ceasefire by year-end at 21%.
• Sacks believed the denials — Trump called in to say the story was untrue, and the Russian readout was similar — because “it wouldn’t make sense for the Russians to invade NATO countries.” He called the invasion-warning story the kind of neocon “threat inflation” heard for years.
• His battlefield assessment was that Russia has air superiority and is making slow but steady progress; Ukraine is out of air defense, Odesa and the Black Sea ports are shut, grain exports have been disrupted, Zelensky needs another $35B, and there are reports of drafting women against Russia’s four-to-five-times larger population. “Zelensky should have listened when President Trump told him, ‘You don’t have the cards’… He should make a deal.”
• Off-docket coda: Chamath thanked Meta for teen Instagram limits — “I’m tired of playing whack-a-mole with my kids” — while Jason said his own children are not allowed social media until 16. Chamath wants TikTok and YouTube to follow: “It’ll allow us to actually parent versus just policing device usage.”

14. Moderna’s “cancer vaccine”: a technique, not a drug — and a $500K price Friedberg rejects

• Jason reported that Moderna roughly tripled from a $20B to a $60B market cap on positive readouts. Friedberg rejects the label — “a vaccine means you prevent yourself from getting something,” while this is immunotherapy for cancer that already exists.
• The mechanism as told: sequence the tumor’s unique DNA, especially in melanoma where UV-driven mutations make each cancer “very unique”; make the neoantigen — or, in Moderna’s route, mRNA that makes the patient’s cells produce it in vivo — and let the activated immune system hunt down and destroy the cancer. The idea dates to the late 1990s, with hundreds of trials tuning delivery, dosing and combinations.
• Friedberg says it is “much more of a technique than a drug,” developed over decades with substantial NIH and other public funding, yet Moderna is discussing a $500K price. He contrasts that with roughly $50K custom-peptide treatments he says are already available through clinics, including in Montana. Moderna’s particular mRNA tools may have strong patent coverage, but an E. coli bioreactor can make the protein without mRNA.
• The wider modality map: CAR-T at $1M shows “incredible” efficacy in blood cancers such as multiple myeloma; Friedberg mentioned a mutual friend developing a cheaper operation, and expects the approach to proliferate overseas. Jason’s practical addendum is that early detection is critical — Grail’s Galleri blood test — so people should “get tested early and often.”