Innovators
14 Curated Dialogues
Brad Gerstner
Frontier Insights
Frontier Thesis & Strategy: Scale and infrastructure dictate the endgame. SpaceX’s multi-trillion-dollar valuation, backed by cash-generative Starlink and cloud-scale compute, establishes the blueprint for OpenAI and Anthropic mega-IPOs. While aggressive token commoditization (down ~90% annually) threatens low-tier providers, durable alpha concentrates in frontier models executing high-stakes, long-horizon tasks and specialized workflows (e.g., Cursor, Fable 5). Orbital compute remains a high-upside call option tied to Starship reuse.
Risks & Warnings: Capital misallocation looms large: adjusted enterprise AI ROI sits at a meager 0%–2% against ~$300B in debt-heavy capex, software multiples are violently rerating, and sovereign AI fragments global dominance.
Key Views & Dialogues
CNBC Halftime Report with Brad Gerstner - October 15th, 2026
Gerstner is tactically cautious after AI stocks’ April–May surge, but more bullish on 2027–28 as model capability drives revenue and recursive development creates “Claude Code moments.” Anthropic’s Q3 revenue of just under $17 billion, up 45% sequentially, supports his expectation of an IPO in “weeks, not months” with demand above $500 billion. CoreWeave debt and AI-safety framework remain key tests.
View summary (7)
- Brad Gerstner is tactically cautious after AI stocks’ April–May surge, yet more bullish on 2027–28. He cut exposure to “medium-small,” expecting consolidation amid higher rates, inflation and political risk, but may begin buying over the next couple of weeks as prices become enticing. His distinction is crucial: hedge-fund capital must manage drawdowns, while growth and venture funds can underwrite five to ten years—“two simultaneous but different worlds.”
- For Gerstner, the AI trade still turns on whether lab revenue can justify the capex buildout. He recalls asking Sam Altman how companies could make $1 trillion of spending commitments against $13 billion of GAAP revenue; this year’s AI revenues then “exploded higher,” with Anthropic expectations moving from roughly $30 billion at the year’s start to reports of at least $100 billion. Consolidation reflects questions about sustainability, open-source commoditization and financing costs, but he welcomes the churn because “it avoids bubbles.”
- Gerstner left Anthropic’s investor day more bullish on the company, its prospective IPO and the road ahead. His “vibe check” among major investors was “a nine or a 10 out of 10,” despite a negative public-market reaction to reported Q3 revenue just under $17 billion, up 45% sequentially versus roughly 140% earlier in the year. He expects an IPO in “weeks, not months” and suspects demand could exceed $500 billion.
- His central thesis is that revenue follows model capability, and recursive AI development should accelerate the next wave of products. Opus 4.5 enabled Claude Code, which he credits with unlocking this year’s revenue; machines are now writing “80 or 90% of the code” for subsequent releases. Gerstner expects analogous “Claude Code moments” in cyber, knowledge work, biology and materials science, with completed but unreleased models providing potential catalysts.
- Gerstner’s non-consensus view is that frontier labs are widening, not losing, their lead over open models. He argues that recursive intelligence compounds, “intelligence is the log of compute,” and the leading labs will add the most compute next year; public comparisons therefore match Chinese or open models against older released systems, not internal generations under safety testing. He still wants vigorous open-source competition and sees price cuts as rational sharing of increased “intelligence density,” not proof that frontier economics are collapsing.
- CoreWeave illustrates Gerstner’s distinction between a violent valuation reset and a broken business. He owns no shares today after reducing AI exposure and reallocating 10% of his book into Meta and another 10% into Tesla, but calls much of the CoreWeave debt narrative misinformation: the 15% public mark concerns about 20% of borrowings, while 80% is GPU-secured and relatively cheap. He sees potential opportunity in both debt and equity, while accepting that a GPU-rental business may not deserve an Anthropic-style multiple.
- On AI safety, Gerstner supports engineered containment and a human-controlled emergency brake, but rejects an “FDA for AI” as a static answer to an evolving technology. He describes Commerce-led frontier-model review, audit and board oversight under the White House superintelligence accord, and an AI task force overseeing cyber and critical industries. He allows that further formalization may eventually be needed, while arguing that safeguards must evolve without letting China reach the frontier first; his approach to China is “trust but verify.”
Brad Gerstner: The $100 Test That Could Save 50,000 Lives a Year
Brad Gerstner is pushing a $100 coronary calcium CT scan as “the mammogram for the heart,” targeting 40 million annual scans and claiming 50,000 lives and $20 billion in government care costs could be saved. Half of heart-attack victims have normal cholesterol, while preventive-cardiology chiefs scan at 30 or 35 and target LDL below 40 after any non-zero score. He thinks Medicaid and Medicare coverage could arrive “by the end of this year,” but soft-plaque limits remain unresolved.
View summary (6)
- Brad Gerstner’s moonshot is turning the $100 calcium CT scan into “the mammogram for the heart,” which he claims would save 50,000 American lives a year and $20 billion in government care costs. Heart disease kills 800,000 Americans annually and is the world’s number-one killer — yet the US does 40 million mammograms a year against just one million CT scans; his target is 40 million. “That’s like ending a Ukraine war every year in the United States.”
- The catalyst to watch: after meeting with Dr. Oz several times and talking with leaders on Capitol Hill and in the White House, Gerstner says he thinks CAC scans for diagnostic purposes will be approved and covered by Medicaid and Medicare “by the end of this year.” The mechanism matters — “doctors ultimately practice according to the standard of care based on what’s covered by insurance” — so he presents reimbursement as the route to mass screening.
- Gerstner’s history lesson is his “regulatory capture” framing: he believes Eisenhower had a heart attack in the Oval Office, after which heart health became a national issue; the stent path beat the calcium-CT path in Washington’s lobby contest. The resulting doctrine — “cholesterol is good enough to track… if you have a heart condition, we’ll solve it with a stent. Wrong.” — is what Gerstner says won out, even though half of heart-attack victims have cholesterol in the normal zone.
- Rob Goldberg’s account of his brother Dave’s death at 47 is the episode’s proof case against conventional screening: Dave passed a running EKG five months before dying of ventricular arrhythmia amid undiagnosed heart disease. EKGs carry a 4% rate of either false positives or false negatives, and his heavy cardio had trained his heart to compensate — “it was pumping harder and harder and harder, and then… it just stopped.”
- Gerstner says the preventive-cardiology chiefs at Stanford, Harvard, and Columbia gave a sharply different personal protocol when pressed past their “corporate speak”: they CAC-scan family at 30 or 35 and, on any non-zero score, drive LDL below 40 — versus the “green zone” of roughly 80-120. Stanford’s head conceded universal scanning won’t eliminate heart attacks, “but you’ll eliminate 80 to 90% of them.”
- The distribution model is a buy-one-give-one model plus a radiology-AI overlay: Gerstner proposed a Salesforce/Marc Benioff example in which employees would receive scans free while the company paid $200 per scan to fund free scans in East Palo Alto; he says major Silicon Valley companies are doing the model, and his team did it with the NFL at the San Francisco Super Bowl. Radiology AI systems like Cleerly refine readings atop Siemens scanners, and Gerstner says he is responsible for 40-50 members of Congress having been scanned — “a target-rich population.”
Brad Gerstner: No AI Bubble, Semis Eat the Nasdaq & AI’s Take Off Problem
Gerstner argues the rally is earnings-driven, even as semiconductor concentration accounts for 70% of Nasdaq’s return. The AI buildout still depends on lab revenue scaling, while rates, power constraints, regulation and a delayed Anthropic IPO remain key risks.
View summary (6)
- Gerstner’s core claim: this is an earnings-driven market, not a bubble — multiples have actually contracted. The market is up 15% this year and 39% since January last year, with earnings up 26% while the Nasdaq and S&P multiples fell; NVIDIA trades at “14 times next year’s fully taxed GAAP earnings.” “This is no bubble like it was in 2000” — but semiconductors are 70% of the Nasdaq’s return, and consumer discretionary, software and financials have barely moved.
- The single most important data point in the market is monthly AI-lab revenue. After Claude Opus 4.5 and Claude Code in December, Anthropic’s monthly revenue went from $2B (January) to $4B (February) to $11B (March), lighting “the fuse” for the April–May rally. The top three — Anthropic, OpenAI and SpaceX — have roughly $100B of collective run-rate revenue, based on July rumors, and Gerstner says they must reach at least $180B by year-end, adding another $80B “just to keep the AI trade intact.”
- The offtake math is the whole game: ~$1.5T/year of capex needs someone to pay the rent. Microsoft, Google and Amazon are “building it to rent it,” so offtake revenue must go from roughly $200B exiting this year to $450B, then $800B–$1T over the next few years, or “we can’t build this much capex.” He’s not worried about demand: knowledge work is “the largest TAM in the history of the world,” and capturing about 4% — roughly $1.2T — would cover the capex.
- He calls Dylan Patel’s 43GW-of-new-compute-next-year forecast too aggressive — expect about 25GW. Permitting, grid interconnection, labor shortages and sold-out power equipment constrain the buildout, but half of that 25GW going to Anthropic and OpenAI is still “enough to generate the revenue”: Anthropic’s reported $100B–$110B this year is being generated with about 1.5GW, so adding 4–5GW could add another $100B.
- Three risks: regulation, power and rates — with nuclear as the cautionary tale. Activist-driven fear shut down 67 fission reactors and “we unilaterally disarmed against China”; he says the same must not happen to AI. He puts the odds of rate hikes tomorrow above 90% and warns a 5.5% 10-year yield would be “a big burden” on equities — “interest rates are to stocks what gravity is to matter.”
- Positioning: medium, mentally flexible and emphatically no leverage. In 2023–25, “you only had to get one thing right” — that AI would be the biggest technology supercycle — and shove chips into the AI trade. In 2026, “it’s all priced now. It’s about facts and circumstances.” If monthly lab revenue approaches $8B and oil retreats, he would add; otherwise, “we’ll reserve the right to go even smaller.” “Don’t YOLO” and do not go 4x levered in this market.
AI Bubble, Stablecoin Boom, and Runnin’ Down a Dream | BG2 w/ Bill Gurley and Brad Gerstner
Circular AI financing, including Microsoft–OpenAI credits and Nvidia’s CoreWeave capacity backstop, can turn investment support into reported revenue while obscuring demand weakness. Mag 5 CapEx reached 66% of operating cash flow in 2025, while OpenAI’s deals imply roughly $150B of 2030 CapEx and at least $150B revenue, intensifying hyperscaler concentration. The key monitoring points are neoclouds and startup chips further out the risk curve, state-level AI regulation, and whether incumbents allow stablecoin rails to scale.
View summary (8)
- Bill Gurley is stepping back from co-hosting BG2 after two years to promote his book “Running Down a Dream” (out late February) and work on U.S.–China relations, regulatory capture, U.S. healthcare, and nuclear — “life begins where your comfort zone ends.” Brad keeps the pod, name, and mission, with Gurley returning as an occasional guest.
- Gurley’s core AI-financing warning: he fed six “non-normal” transaction structures into ChatGPT and the model itself “would find its way toward company names like Enron and WorldCom.” The circularity started with the original Microsoft–OpenAI credits deal — a “cashless transaction” that becomes income-statement revenue — and is now competitive: “we’re fairly pregnant with it.” It raises the odds of over-provisioning while “hiding some of the signs that would tell you things are slowing down.”
- The single most peculiar deal per Gurley: Nvidia’s promise to buy any CoreWeave capacity it can’t sell — it helps CoreWeave’s debt financing but means investors “probably won’t be told” if real demand at a pure-play softens. Brad’s counter: these are disclosed public-company transactions and every analyst will now ask the question every quarter.
- Gurley’s continuum framework: judge each deal by “would this revenue have been purchased but for this investment?” Brad says Nvidia deploying a fraction of ~$450B in coming free cash flow into oversubscribed companies (OpenAI, xAI) doesn’t worry him; Gurley says he isn’t concerned as an Nvidia shareholder. A chip with one customer funded by $10B from its own manufacturer would raise red flags. Expect yellow flags “further out the risk curve” — neoclouds and startup chips desperate for capital.
- Neither calls it a bubble on multiples — Brad cites Howard Marks: “you can’t be on bubble watch if the multiples aren’t high enough” — but the scale is “remarkably unprecedented”: Mag 5 CapEx hits 66% of operating cash flow in 2025 ($379B vs $156B in 2023), consensus fading to 45–50%. Meta’s stock works this time, unlike Reality Labs, because investors see the earnings.
- Brad frames the Broadcom announcement as well over $1T of incremental CapEx; Gurley reads OpenAI’s broader deal blitz as manufacturing “escape velocity” — “daring people to follow them.” Brad’s math: OpenAI is on the hook for ~$150B of CapEx in 2030, needing at least $150B revenue — “more than plausible” for OpenAI, but it makes the game “very, very difficult for anybody else other than the hyperscalers.”
- Both want federal preemption of state AI laws — Colorado’s algorithmic-discrimination act and California’s SB 243 (private right of action for chatbot “emotional harm”) create “mud” with “zero chance” of not slowing U.S. players versus China, which is “run by engineers” while America is “run by lawyers.” Brad floats a moratorium and even AI companies blocking citizens of offending states.
- Stablecoin rails are ready now: Coinbase/Circle’s 4% “rewards” (interest by another name, post-GENIUS Act bank lobbying) plus instant, pennies-cost transfers; Gurley says, “The rails are there. They’re ready.” Brad’s call: hyperscalers — Amazon, Meta — will re-enter stables because “money is a network effects business”; Gurley: “I hope the incumbents aren’t able to strangle this thing in Washington.”
China, China, China. Breaking Down China’s Tech Surge | BG2 w/ Bill Gurley and Brad Gerstner
Xiaomi reportedly builds 1,000 cars daily with 2,000 employees, versus roughly six employees per US car per day, and has a 30–40-week backlog. That productivity gap challenges auto reshoring, while China’s open AI stack and America’s limited trade leverage make joint ventures, policy shifts, and the K visa important watch items.
View summary (8)
- Bill Gurley returned from his first post-COVID China trip with one framing quote that carries the episode: “every founder and every VC in China studies the West at a nauseating level… The West doesn’t do that of China.” His core argument — if China is the most consequential relationship for America, “I don’t understand how you’d want to know less” — and he notes that I think all but four of the 24 members of the Select Committee on the CCP have never been to China.
- The auto data is the tradeable shock: likely Xiaomi builds 1,000 cars a day with 2,000 employees (vs. ~6 employees per car/day in the US), is sold out with a 30-40 week backlog, and its ~$40K car made Ford CEO Farley say “if we lose this we do not have a future at Ford.” Implication Gurley draws: reshoring auto jobs may find “there may not be any jobs” — global car manufacturing could total ~400K in 5-10 years.
- The US is only ~14% of China’s exports and ~3% of its GDP — Brad’s point that America overestimates its leverage. BYD’s $10-20K EVs illustrate the comparative-advantage argument: why would Mexico buy the $50K one from America?
- Hidden-value call: likely Baidu’s Apollo robotaxi rival to Waymo could cost ~$30K per vehicle, if that figure is right, vs. reportedly $150K+ for Waymo (partly China’s solid-state LiDAR edge), while people value Waymo at $170B inside Google — yet “you can buy shares of Baidu for zero enterprise value,” a $30B market cap with $30B cash.
- On AI: China’s 14th five-year plan mentioned open source; likely Qwen matters because Gurley thinks Alibaba is roughly a 70% player in China cloud, ByteDance is the company to watch on consumer AI, and open models helping each other create hyper-competition “maybe more” intense than EVs. Entrepreneurs there aren’t particularly concerned about a model monopolist.
- Gurley’s debunk of the “they only win by stealing and subsidies” narrative: BYD’s Stella Li — “come show me the money I’m getting from the government” — plus the thought experiment that free Tesla patents + subsidies still wouldn’t make Ford/GM competitive. Brad’s warning: the cheating narrative “allows us to delude ourselves into believing we don’t need to get better ourselves.”
- Policy stance (both in the “tech pragmatist” camp with Jensen and Cook): compete, don’t decouple — invite Chinese JVs in EVs/solar, including some 49/51 ownership arrangements of the kind used in China, let Korea build US nuclear at 1/4 the cost, reserve tariffs for genuine national-security industries (rare earths, chips, pharma), and strip the regulatory “mud.” A pragmatic Trump-China deal “could easily get done” if the rhetoric isn’t derogatory.
- Watch items: China’s new K visa may allow people studying technology to come without a job requirement, just as Gurley heard stories of groups of 50-100 admitted PhD students being told they could not attend in the US; China’s VC market is in a lull (Sequoia/GGV split, provinces investing on “non-starter” terms, “don’t be the tallest tree”) — yet entrepreneurial energy in Shenzhen is undiminished.
China Open-Source, Compute Arms Race, Reordering Global Trade | BG2 w/ Bill Gurley and Brad Gerstner
China’s seven-plus open-source labs are compounding through distillation, offering 90% of the intelligence at a 90% price discount, while 200x Google token growth and Groq’s rack utilization show inference demand remains strong. OpenAI and Meta could restore US enterprise share by Q4 this year or Q1 next year, but sub-cost pricing, model commoditization and a possible year-end China deal leave margins and policy outcomes unresolved.
View summary (7)
- Groq COO Sunny Madra’s core call: China’s seven-plus open-source labs are compounding on each other via distillation — K2 is “sort of a well-known remix of what DeepSeek had done” — and a Qwen 30B-parameter release now performs “as good as GPT-4o.” The Chinese leaders offer “90% of the quality in terms of intelligence, but at a 90% price discount,” and enterprises worldwide are taking that trade over American-values alignment.
- Yet Sunny predicts the pendulum swings back: by Q4 this year or Q1 next year a top-three worldwide model will be US-based open source, driven by OpenAI’s imminent open-source release (enterprise demand is “like a Tesla Roadster… that’s the model that everybody wants to use”) plus Meta’s efforts. Enterprises want brand and accountability — the Red Hat-on-Linux pattern — so a US-domiciled equal on price/intelligence would “rise to the top in a few days” on OpenRouter.
- Compute demand shows no evidence of a glut. Google went from 5 trillion tokens/month to a quadrillion — “that’s a thousand trillion… 200x” in a year and a few months; Groq racks are “fully consumed within a few hours” of firing up; Elon is targeting 50M H100-equivalents (~11GW) and the Abilene deal is 4.5GW, “well above” the $500B pledge. Brad cites CNBC’s report that Iconic is expected to lead a $5B Anthropic round at $170B; xAI rumored at $150–200B — OpenAI + Anthropic + xAI would be roughly $1T combined.
- Bill Gurley’s caveat on the boom: “no one here is raising price… everyone’s pricing to share. That means they’re pricing under cost” — negative gross margins are rumored at top AI brands, and Anthropic throttles usage rather than raise prices. When the Oracle/Sun → Linux/MySQL cost-optimization phase arrives, “that will create a bump in the supply-demand curve.”
- The model layer is commoditizing; value accrues at the application/consumer layer. Brad: ChatGPT may cross 1B weeklies this year and owns the first consumer threat to Google in 20 years; Bill argues OpenAI’s durable moat is “switching cost and lock-in more than… staying on the edge of the model race.” Coding agents are already distilling Apache-licensed Chinese models to “eliminate the entire cost of Sonnet… for 70% of use cases.”
- On tariffs, Brad declares the Bessent ~$300B consensus won over “nuclear Navarro”: EU at 15%/0% plus ~$750B energy purchases, Japan at $550B directed investment, Brad recalled Bessent saying June had the first monthly Treasury surplus since 2015, import prices rising slower than domestic goods, Atlanta Fed GDPNow back at 3%. “All we’ve seen so far is deals deals deals deals… he’s in line for a bonus” — hedged with “yet” on inflation.
- Brad goes “out on a limb” against consensus: a very big China deal gets done before year-end — the president is “a deal junkie,” not a China hawk; Beijing postponed retaliatory tariffs and invited a September–November visit; the deal could span rare earths, chips, “maybe even military cooperation,” with Brad expecting China to pay at least 15%. On positioning: back in since May 2, up 30% off the NASDAQ bottom, but “I think we’ve captured a lot of the return for the year” — still bullish on AI opportunities, including Groq’s (accidentally disclosed) new round.
China, AI Immigration, Rare Earths & Chips, Tariffs, Markets | BG2 w/ Bill Gurley & Brad Gerstner
China’s 500–1,000-competitor model produces low-cost, globally competitive products, but Beijing may prioritize employment, affordability, and strategic capacity over $3T market-cap winners. That objective function, a possible 30% BYD price cut, and the proposed rare-earths-for-Blackwell 30 trade frame a market shaped by policy as much as economics. Meanwhile, reversing green-card and Chinese-student visa signals could weaken the US AI talent pipeline, while 10–15% tariffs and the reconciliation bill remain near-term market catalysts.
View summary (7)
- The dot-com lesson for AI: Gerstner, prepping his HBS 25th-reunion talk, ran the split-adjusted math — Amazon is up ~440X from its 2000 high and ~1,800X from the low, and the Nasdaq is still up 5X from the bubble peak. “We overestimated what it was going to do in the short term… but dramatically underestimated the long term” — and AI’s short term is already running faster than the internet’s ever did.
- Data walls are going up in hyper speed: Reddit sued Anthropic despite a deal, Anthropic cut Windsurf off its models, and Salesforce changed its ToS so you can query but not train on your own CRM and Slack data. Gerstner predicts enterprise apps will have to declare themselves “open data or closed data” — and challengers will declare open and steal customers. With OpenAI hitting 400B annual searches 8 years faster than Google, the TripAdvisor-in-Google’s-underbelly dynamic is replaying in fast-forward.
- China’s success is Darwinian, not Soviet: Beijing seeds 500–1,000 competitors per strategic industry at the provincial level and lets markets whittle them down — producing $130/car solid-state lidar vs Waymo’s $5,000, and 500+ EV startups vs a US handful. Corollary for investors: China deprioritizes market cap — BYD’s 30% price cut may have been state-encouraged, but Gurley says he has no proof — so betting on Chinese Mag-7-style $3T winners may misread “the objective function of what a win is.”
- The visa 180 is a self-inflicted wound: Rubio’s move to revoke visas of Chinese students “in critical fields” reverses Trump’s All-In promise to staple green cards to diplomas, at a moment when ~40–50% of top US AI researchers are of Chinese origin. Gurley warns of “a McCarthy-like perspective that could be very dangerous to our long-term competitiveness”; Gerstner: build the championship team — “we should not care where in the world the basketball player comes from.”
- Gerstner’s trade proposal ahead of the Trump–Xi call: rare earths for the deprecated Blackwell 30 — no HBM, no CoWoS, degraded for training but keeps China’s researchers and developers in CUDA, avoids accelerating Huawei, generates billions for the US and Nvidia, and buys years to stand up domestic rare-earth supply. The tail risk of the ban: if AGI is existential and China is locked out, “we just have to go take Taiwan.”
- Market call: land the plane or give back 10–15%. Nasdaq +20% off the lows makes sense if blended tariffs settle ~10–15% and the reconciliation bill passes (failure = a $4T tax increase); core PCE supports two Fed cuts in the back half, and 10-year hand-wringing ignores that yields have sat in a 4–5% range for two years. “Don’t take yourself out of the game” — but the proof is in the pudding.
- Governance lightning round: Gurley’s Delaware warning is landing — high-profile and private companies are reconsidering incorporation after evidence of 66x lawyer fee multipliers — and the ISS/Glass Lewis proxy duopoly (97% share, >80% foreign-owned) is an open incubation target: “reach out to us. We could help fund it.”
AI, Middle East, China, Tariffs, Recon Bill, Invest America | BG2 w/ Bill Gurley & Brad Gerstner
The repeal of the Biden diffusion rule has redirected US AI policy toward partnership, accompanying $1T commitments each from Saudi Arabia, Qatar and the UAE and a 5GW UAE-US AI campus with 2.5M GPUs. Tariffs now appear closer to the Bessent route at roughly $300B, while the reconciliation bill adds stimulus without solving the deficit; H20 approval and China’s open-source momentum remain key watchpoints.
View summary (8)
- After Gerstner’s 10 days in the Gulf, Gurley frames the Trump administration’s May ‘25 repeal of the Biden diffusion rule as flipping US AI policy from “Washington control” to “Silicon Valley partnership” — ~$1T investment commitments each from Saudi, Qatar and the UAE, plus a 5GW UAE-US AI campus (~2.5M GPUs, “almost like a global Stargate”) with Nvidia, OpenAI and Oracle. “We’ve ripped the chains off… now we’re going to allow our companies to go compete and win.”
- The counterfactual both hosts feared: a year more of the old path and the Gulf commits to Huawei full-stack AI — “we were 100% on a path towards building a wall around America, not building a wall around China,” the exact opposite of the internet era, where the US won because the world ran on American tech.
- Tariffs appear to have leaned toward the Bessent door, away from the “nuclear Navarro” door: Gerstner’s team re-ran the math — China at 30% strategic / 15% non-strategic, rest of world ~10% yields ~$300B in tariffs vs. the $800B implied on Liberation Day and $77B last year. Still a 4x revenue jump but digestible on a $28T economy — which is why the market that fell 20% bounced 20% in 20 trading days, “just back to where we started the year.”
- On the remaining H20 export ban, Gerstner would reverse it: China already has frontier AI (“their open-source models might even be in front of the United States”), Nvidia was tracking to a $50B China business, and keeping ByteDance/Tencent in the CUDA ecosystem beats gifting data and profits to Huawei. He “would not be terribly surprised” to see a deprecated Nvidia chip approved for China within the 90-day window.
- The reconciliation bill is stimulus, not austerity: extending the 2017 cuts (a $3-4T tax hike avoided), no tax on tips/overtime/social security, immediate expensing — roughly $400-500B annual stimulus — but tariff revenue isn’t scored, DOGE cuts need rescission, and defense rises ~$150B to ~$1T. “People should not get their hopes up” the $2.2T deficit gets halved; Gerstner calls saddling kids with $50T+ of debt “morally unacceptable.”
- Invest America made the House bill (renamed the “MAGA account”): $1,000 in an S&P 500 account for each of 3.7M kids born yearly, a “401k from birth” compounding to ~$50K at 18 and ~$150K at 30 — Gerstner’s redistribution alternative to a 57% all-in rate for high earners that he says is nearing the Laffer tipping point.
- Gurley’s governance alarm: Joe Grundfest’s May 14 study shows Delaware fee awards of 7x+ are 23x more likely than federal courts (10x+ awards, 57x more likely), with 55% of the extreme cases from just two judges — the Tesla nine-share plaintiff wasn’t a one-off but structural. “Any company that I’m involved with, I’m going to encourage to leave”; Texas SB 29 (codified business judgment rule, 3% derivative-action threshold) is the destination.
- Two crypto/China kickers: the Genius Act cleared its Senate hurdle — stablecoin issuers could become “the biggest holders of US debt on the planet” within 5-10 years — and China now appears to have four deep-pocketed open-source models (a likely Baidu release is scheduled for June 30), while the US open-source response still waits on OpenAI, Meta post-Llama 4 blowback, and Elon.
Tariffs, Free Trade, Export Controls, H20 & Rare Earth Ban | BG2 w/ Bill Gurley & Brad Gerstner
The H20 export ban was self-defeating: Nvidia lost $12-15B in China sales after designing the chip to comply with prior rules. Bernstein says it handed Huawei monopoly profits and removed CUDA competition, while raising the trade deficit by $15B. If uncertainty lasts 8-16 weeks, S&P earnings could crack as CEOs retreat from capex.
View summary (7)
- Gurley’s core call: “win the AI war” is finite-game thinking applied to an infinite game — no one he talks to believes the US can permanently block China’s AI development, and “there’s no clock that ends… no one that declares a winner.” Policy built on that premise — H20 bans, DeepSeek bans, pressuring ASML and TSMC — will produce “spectacularly problematic policy.”
- Both agree the H20 export ban was self-defeating: Nvidia designed the chip to comply with the prior rule, then got ~$12-15B of China sales banned anyway — a $5B inventory write-off and, per Bernstein, “we basically just handed the China AI market to Huawei,” clearing the decks for monopoly profits and removing CUDA competition while Huawei’s CloudMatrix 384 remains China’s GB200 competitor and Huawei is near the chip frontier.
- Gerstner’s line on where he’d draw it: no GB200s to China is defensible, H20s were a close call gone wrong — Huawei Ascend 920s already surpass the H20, China has 100 nuclear plants under construction at 1/4 US cost, so “if it comes down to power, I guess they’re going to win” (Gurley). The ban also raises the trade deficit by $15B and forfeits $2-3B of US tax receipts — “we just unilaterally disarmed.”
- The tariff regime remains two irreconcilable policies conflated: tactical reshoring of chips/pharma/rare earths (broadly supported) vs. the “nuclear Navarro” external-revenue-service vision. Gerstner still believes Trump “is fundamentally a free trader” and will walk China to
25-30% and rest-of-world to 5-10% ($200B) — but that’s still a 3-4x tariff increase vs. last year. - Markets haven’t priced the earnings risk: S&P consensus has only slipped from $273 to $265 (15%→12% growth) while CEOs say they’re “on hold” and managers are at “max caution” — United Airlines literally gave dual guides ($7-8 recession / $12 not). Altimeter is in the “bottom third” of normal risk units, leaning back in on big risk-off days; the next leg down, if it comes, “is going to be around growth.”
- Escalation is the underpriced tail: China’s April 5 rare-earth ban (60% mined, 90% refined in China) was called “a kill shot” inside the White House orbit, and if AI is framed as existential, export bans are “firing a shot at somebody’s national security” — Gurley warns any single decision “could go as far as leading to a hot war,” and Apple or Tesla getting kicked out of China is a plausible scenario that must be discounted.
- Timing matters more than terms: Gerstner says if this same uncertainty persists 8-16 weeks, S&P earnings crack; Gurley thinks “months is too long” — reflexivity is already self-reinforcing as CEOs decommit capex and capacity, which is itself inflationary. Gold outperforming the S&P by ~40% YTD tells you where money is hiding.
Liberation Day, Tariffs, US v China Open Source, OpenAI, $CRWV, TikTok | BG2
Trump’s Liberation Day tariffs, the uncertain path to a deal with Xi, and the resulting shock to business investment frame a broader discussion of how geopolitics is reshaping AI. The conversation weighs OpenAI’s strategy and valuation, agent-driven threats to Google, CoreWeave’s durability, and the policy choices that could determine which technology platforms endure.
View summary (8)
- Liberation Day landed at the big end: headline tariffs jump from $77B last year to ~$750B — above even hawk Peter Navarro’s $600B — with China at a combined 54%, a 10% floor on every country, and 25% on autos, though pharma and semiconductors were exempted (Brad’s hunch: effective number “right around 600 billion”). S&P/NASDAQ futures swung 600bps lower during the presentation, on top of indices already down 8-15% YTD.
- Brad’s tradeable call: the plane lands at $300-400B, “not 6 or 700 and certainly not a trillion” — Trump “can’t afford to lose a single Republican vote” on reconciliation and senators are hearing it from constituent CEOs. No clearing event for 30-60 days, and the key one is Xi-Trump: there’s “no way that lands at 54%.” Meanwhile, “the best opportunities to buy something are when people are a little fearful.”
- Gurley’s core objection isn’t the rate, it’s the ambiguity: reshoring a factory might take ~3 years at higher cost — “I don’t think our labor is globally competitive, nor do I think it wants to be” — and not knowing policy 3-12 months out “makes it very hard to allocate capex in any meaningful way whatsoever.” The Montana CEOs, “almost to a one,” said February and March slowed; one had four contracts cancelled by angry foreign counterparties.
- China’s open-source push is longstanding, not a Xi directive — the perfect answer to 40-50 years of IP-theft accusations — and DeepSeek (1,500+ Hugging Face forks) is prolific. DC may move to limit it, opening a short window for a US player to “go left of DeepSeek in terms of openness”; Gurley: “the next 3 to 6 months could dictate who’s standing on top of the hill 5, 10 years from now.”
- OpenAI’s open-weight model with reasoning — one of its first since GPT-2, explicitly without Meta’s 700M-user license cap — reads to Gurley as a Kubernetes-style deft move that “wipes out the business opportunity for other models on the API side.” Llama 4 is rumored this month: 400B-parameter MoE, ~10M context window. Washington’s motive per Brad: no “Huawei-DeepSeek Belt and Road.”
- OpenAI’s $40B SoftBank-led round at ~$260B pre ≈ 20x forward run-rate revenue vs Anthropic ~50x and X/xAI ~80x — the leader priced at a discount to peers. Gurley: headline “smells of being promotional,” it’s a “sport of kings” where unit economics get postponed ($5-7B losses, ~$15-20/yr per free user); Brad’s rebuttal: Uber was “never going to be profitable” in 2020 and does $6B FCF this year — and demand is throttled at 1M new users/hour, 500M WAU, 20M paid.
- Agent economics break Google’s LTV math: Booking.com pays Google $50 on a $20 hotel take using lifetime-value logic; a white-label service under an agent can share at most $10 — all upside for OpenAI, a major replacement risk for Google (already $200→$150). The antitrust irony: “finally in 2025, the first time we’ve actually had competition for Google… now we get around to breaking up their search monopoly.”
- $CRWV broke issue at ~$37, then rallied to ~$60 on a Google/Nvidia Grace Blackwell deal that answered the Microsoft-concentration critique; Brad argues GPUs retain value beyond two years (A100s still working in year five, V100s ran seven), while Gurley questions whether depreciation should be accelerated — “the answer may lie somewhere in between.” And the proposed TikTok fix: a NewCo licensing the algorithm with ByteDance under 20%, unlocking ~$600B of trapped US venture value — unless China, freshly hit with 54% tariffs, blocks it.
NVDA GTC, M&A Wiz / Goog $32 B Deal, April 2 Tariff Uncertainty; Huawei Belt & Road; ChatGPT | BG2
Jensen raised NVIDIA’s AI data-center TAM to $1 trillion annually by 2028, while Bill warned that slower pre-training and customer depreciation cycles challenge the bullish narrative. April 2 tariffs and the Biden diffusion rule remain the major catalysts, with export controls already strengthening Huawei and China’s vertically integrated chip supply chain; AI unit economics could bring broader resets.
View summary (8)
- Brad’s February “golden age of uncertainty” call has played out — NASDAQ off ~10%, many components down 20-30% — and the market may now be rolling into leg two: recession fear, which “can become a self-fulfilling prophecy.” The data is genuinely mixed: consumer and business confidence worsening, airlines (Delta, United, Frontier) flagging demand, GDPNow turning down, and the Fed just cut its GDP median to 1.7 from 2.1 while raising inflation to 2.7 — yet BofA’s real-time card data shows consumer spend up 6% y/y and hedge funds sit in bottom-quartile exposure wanting back in.
- On April 2 tariffs, Brad is emphatic this is not a negotiating tactic — the administration has “a principled view about restructuring globalism” (JD Vance’s “two conceits of globalism,” Bessent’s “American detox period”). His tradeable frame: tariff revenue going from $65B to ~$150B is easily absorbed; $65B to $1 trillion “pushes us into a recession.” Expect no big flows into the market for 60 days while participants wait for the clearing event.
- Google’s $32B all-cash Wiz deal (~30x+ forward revenue on ~$1B ARR, vs Google at 5x) is the litmus test for whether M&A animal spirits get unleashed. Bill’s read: EU pressure blocks Google from buying anything in search, AI, or YouTube, so enterprise security was a plausible lane left — and Ferguson’s FTC (“I’m not a regulator… I’m a cop on the beat,” “go to court or get out of the way”) is a genuine break from Khan-era purgatory. With public software at 6-7x revenue and this deal at 30x, “there’s ample room for buyers and sellers to meet in the middle.”
- At GTC Jensen doubled down: DeepSeek was “profoundly misunderstood,” compute needs are “a hundred times greater than what we believed a year ago,” and the AI data-center TAM reaches $1 trillion per year by 2028 — versus Wall Street consensus that Nvidia somehow tops out at $250B revenue. At 20x next-year earnings (below Costco’s 50x), Brad sees no re-rating needed — just earnings growth in a market where Nvidia holds or gains share.
- Bill’s counterweight on Nvidia: pre-training scaling has “at least slowed materially,” the token-count framing is “an ultra promotional way to talk about things,” and the “chief revenue destroyer” riff creates a real economic problem for customers who just stretched server depreciation from four to six years. His net: of the MAG7, Nvidia is both “executing the fastest” and “most exposed to global trends” — and “the number one risk on the stock is government action from DC.”
- Both agree the Biden AI diffusion rule plus high structural tariffs would be “unilaterally disarming America in the race to AI” — export controls already backfired, with DeepSeek running inference on Huawei’s Ascend 910 and China building a vertically integrated chip supply chain. Bill’s warning: repeat the Nortel telecom mistake and you get a “Huawei Belt and Road” across Saudi, UAE, India, and Southeast Asia.
- Consumer AI may already be decided: ChatGPT holds #1 in the App Store while DeepSeek fell out of the top 100 and Grok slid to ~65 — “you’re going to have to be 10x better than ChatGPT to slow down that inertia.” OpenAI is “massively supply constrained” at 400M+ weekly users, building Abilene and Denton campuses just to launch “four or five products literally sitting on the shelf.”
- Bill’s “Gurley negative gross margin AI theory”: stacked negative-margin layers (hyperscaler → model → app) subsidize end prices and triple-count revenue, VC money plus winner-take-all pricing psychology makes it systemic, and “you could have substantial resets as people get in touch with unit economics.” Brad’s corollary — Anthropic’s AWS-mediated revenue carries “very little or no contribution margin,” while mostly-direct OpenAI is the best-positioned independent; the AI internet analogy should be e-commerce, not Google.
Grok 3, AI Memory & Voice, China, DOGE, Public Market Pull Back | BG2 w/ Bill Gurley & Brad Gerstner
Grok 3’s frontier performance makes it a new player, but Gurley sees a ceiling as benchmarks converge, shifting competition toward consumer aggregation, where OpenAI has 400M weekly users and $11–12B expected revenue. Memory and voice could disrupt lock-in, yet $20B annual losses, gigawatt campuses, China policy, and DOGE-driven austerity leave model economics and public-market downside in focus.
View summary (7)
- Grok 3 reached the frontier in record time — but Gurley reads it as ceiling, not headroom. Investors billed the Memphis cluster as proof “pre-training still has headroom”; instead Gurley “had the opposite reaction — I felt like they just slammed up against this ceiling that’s holding everyone in,” echoing Ilya and another unnamed commentator. Gerstner’s counter: it also ships inference-time reasoning, and X’s productization took it to #1 on the App Store — “there’s a new player in the model market.”
- Benchmarks are converging; the tradeable question is consumer aggregation. Gerstner’s search-wars analogy: AltaVista and Lycos also scored fine on benchmarks, but the value went to Google. He expects 70–80% share to the winner (not a 99% monopoly), and OpenAI — 400M weekly users, ~$11–12B expected revenue, MAU inferred at 700–800M against the “magic number” of a billion — is “nearer at escape velocity” and accelerating: “everybody else caught up on the benchmarks… nobody caught up on the consumer velocity.”
- Google’s cannibalization is now measurable: public companies report organic clicks down 20–40% YTD, “SEO is dead,” and Gerstner’s own Google usage is “80% cannibalized by ChatGPT.” He calls the AI-answer takeover the right call — innovator’s dilemma head-on — but the paid-click-growth vs OpenAI-user-growth chart is “not going in the right direction.”
- The windows to break OpenAI’s lock-in, per Gurley: memory, voice, a breakout feature, or a network effect. Memory is the big one — “if you get memory the switching costs explode” and free-to-paid conversion rises. Gerstner’s demo: advanced voice mode interviewing his 89-year-old mother for her life story — the capability already exists; the product problem is that “you don’t know that it can do those things.”
- The buy-in to this “sport of Kings” is roughly $20B/year of losses and multi-gigawatt campuses (Meta rumored shopping a $200B, 6–8GW site; Microsoft at $80B capex). Satya’s “I’m happy that some of these are leases” reads as a hedge/brake-tap; only Sam and Elon can still raise into the game; and the microeconomics are treacherous — a 20x price gap between today’s model and yesterday’s makes a model “a fast depreciating asset the second you’re off the frontier.”
- On China, both agree Washington’s frame is broken: “I can’t imagine an end state where we control all the AI and they don’t have any — it’s already too late… that would be remarkably naive.” The Biden diffusion rule forces US chipmakers to “compete globally with Huawei with one hand tied behind our back” and “almost guarantees a Huawei belt-and-road” — “if I owned Nvidia, my number one concern would be excessive regulation coming out of Washington.”
- Gerstner is running half his normal risk. Tariffs going from $56B to ~$500B plus DOGE pulling $500B–$1T of federal spend is austerity — liquidity in reverse through C+I+G — and could produce a “random run-of-the-mill 10–15% drawdown.” He calls it necessary shock therapy (“get fit to avoid bankruptcy”), notes Buffett’s $400B cash pile and Druckenmiller/Marks/Cohen turning cautious, and warns government revenue lines go outright negative: one airline’s government tickets are already down 50% YTD.
DeepSeek, Open Source, Tariffs, DOGE, Market Impact | BG2 w/ Bill Gurley & Brad Gerstner
DeepSeek’s R1 was genuinely 30 to 50% more efficient, but reaching o3-class performance may require 10x compute as export controls and inference demand constrain its cluster. Open-source adoption is accelerating while tariffs and DOGE raise policy risk, leaving a super cycle intact but discount rates, multiples, and Nvidia’s sanctions exposure under pressure.
View summary (7)
- DeepSeek’s moment of maximum parity is already behind it. Gerstner’s team, speaking directly with DeepSeek, confirmed the ~$1B total TCO alongside the real $6M final training run (vs ~$10–15M for o1 — “30 to 50% more efficient”), but the log-linear scaling math now demands ~10x the compute to reach o3-class — with no Blackwell access, export controls binding, and most of their limited cluster consumed by inference demand. “This was the moment in time where their compute comparison to OpenAI was the closest it’s ever going to be.”
- Open source may have hit its tipping point. Hugging Face went from 500 R1 variants 48 hours post-release to 1,300 today under a no-strings MIT license, and Sam Altman conceded in his AMA: “I personally think we have been on the wrong side of history here.” Gerstner predicts closed labs open-source one-to-two-generation-old models this year while true frontier models — from OpenAI and DeepSeek — stay unreleased, with “only the agent” shipped.
- Gurley’s contrarian core: a foreign, open-source disruptor beats a domestic proprietary one — for safety, free speech, and innovation — and China is a 30-year open-source veteran (Linux donors, RISC-V) precisely because “if you believe you have the fastest, cheapest, most capable engineers… you’d rather live in a world where there’s no IP protection.” His warning: “Linux doesn’t have a geography” — the winning model may be a fork nobody’s country owns, and walling off the US could hand the rest of the world (DeepSeek at ~15% of ChatGPT DAUs in India) to China.
- Don’t be lazy in believing the “Bessent consensus” that tariffs are just a loaded gun rarely discharged. Gerstner’s variant view: Trump holds a decade-old, principled McKinley-era belief that pre-1910 tariff revenue built America and that swapping tariffs for income tax “gutted the middle class” — meaning tariffs as permanent tax-reform architecture, not negotiating theater. Gurley’s rebuttal: he doesn’t know of a long-term successful tariff program, the Fed’s 2018 washing-machine study showed prices up and few jobs created, and “you’re just going to make yourself Europe.”
- DOGE runs on two tracks, and track two could reach the Supreme Court. Track one is normal reconciliation (a signable bill by April/May); track two is Elon cutting by executive authority — asking “who sends out the wires,” shutting USAID (“no apple to be saved, it’s a total ball of worms,” $50B/yr). Gerstner expects a Schumer lawsuit on Congress’s power of the purse but thinks it’s “on pretty weak footing” against the doctrine he called “empowerment” — and if a credible $1T cut back to the 2019 baseline lands, “interest rates are going to come down.”
- The trade: fundamentals on fire, multiples at risk. Stacked technological, economic, and political uncertainty forces the discount rate up and multiples down even as Jevons-style demand explodes — “I think the super cycle is on fire… but at the moment it’s the Golden Age of uncertainty.” Gerstner tells his team to downsize risk (“you don’t do the tenth best idea”), expects heavy volatility for six months, and warns mid/late-stage VC marks lag but follow public multiples downhill.
- Second-order trap for NVDA bulls: loudly claiming DeepSeek secretly had more GPUs invites Washington to tighten sanctions that hit Nvidia’s own revenue — “you think you’re protecting Nvidia… and you may end up with sanctions that end up hurting Nvidia’s revenues.” A threatened Taiwan chip tariff is just a tax on Nvidia, AMD, Meta and Amazon that buys less AI research — unless structured as a 2.5-year-delayed lever tied to US fab buildout.
Market Predictions, Rates & Inflation, DOGE, CES, AI Compute | BG2 w/ Bill Gurley & Brad Gerstner
Higher rates, not AI, are the immediate valuation risk as the 10-year approaches 4.7% and could anchor stocks at 5.25–5.5%. Large-tech capex has risen from $160B to $260B while compute remains scarce, supporting Nvidia demand but intensifying power and business-model constraints. The market’s upside depends on non-tech earnings accelerating from 2% to 11%, with DOGE cuts and a reconciliation package by April or May as the key catalyst.
View summary (8)
- The boogeyman is rates, not AI: Gerstner’s team-wide year-end exercise landed on inflation plus higher rates colliding with elevated valuations. The 10-year has gone up almost 100bps to ~4.7% while the market repriced from many cuts to one in the first half of next year; if the 10-year goes to 5.25–5.5%, “it’s going to be an anchor on the stock market.” Altimeter started 2023 at 95% net long, last year ~80% — fewer chips on the table because “the price of entry to play is higher.”
- Not a bubble, but no cushion: S&P at 23x vs. the recent 22x peak and 16x trough; Meta 23x, Google 21x, Nvidia down from 66x to 36x consensus (28–30x on Altimeter’s numbers) while growing at unprecedented rates. The catch: Mag-5 earnings decelerate from 44% to 21% while consensus needs the other 495 to accelerate from 2% to 11% — healthcare +4→+20%, industrials -4→+16%, materials -10→+17% — and tax cuts explain only ~30% of that bump. If non-tech earnings miss, “it’s hard for me to see how you can have a big year.”
- Capex is real and pre-committed: combined large-tech capex stepped from ~$160B to ~$260B, Microsoft alone guiding $80B, consuming roughly 100% of incremental free cash flow at Meta/Microsoft. The debate over whether the 2025 spend materializes “we can put to bed” — it shows up, partly prisoner’s dilemma, but mostly because if Elon is right that AI does every cognitive task within 3–4 years, the labor being replaced “is measured in trillions.”
- Compute stays scarce all year: Microsoft was compute-constrained all 2024; OpenAI couldn’t ship Sora or voice widely for lack of compute; Jensen says inference — already ~half his revenue — goes up “a million x, maybe even a billion times” as reasoning models and agents devour tokens. Brad: “I think we’re going to end this year compute constrained.”
- The tradeable recipients: consensus Nvidia data-center revenue of $61B (2023) → ~$200B (2025) implies share loss Brad doesn’t believe, since the $60B step-up equals the hyperscalers’ incremental spend almost wall-to-wall. Memory is the contrarian leg — SK Hynix at ~6x forward because the market prices commodity boom-bust while Altimeter sees “secular not cyclical” plus a re-rating. Power is the binding constraint: likely Gavin Newsom still hasn’t signed Diablo Canyon past 2029 (10% of California’s clean power) while “China is building 100x as much as we are.”
- DOGE is the swing factor for the 10-year: 2024 federal spending of $6.7T vs. a 2019 baseline grown at 2.5%/yr of $5T — a $1.7T differential that explains the $2T DOGE target. Trump tax cuts add ~$400B/yr of stimulus; the market wants $300–400B of offsetting cuts and “I’ll believe it when I see it.” Brad’s out-of-consensus forecast: this administration actually cuts, the 10-year comes in, and “it could be off to the races” — resolved in a single reconciliation package by April/May.
- State AI regulation is the stupidity risk: the post-1047 regulation push “moved underground” into as many as 25 state initiatives, with the Texas bill (risk assessments, backward-looking liability) the most ironic given Texas’ pro-business brand — Bill calls a state-by-state gauntlet “mind-numbingly stupid” versus federal preemption under the commerce clause. Brad expects Trump to rescind the Biden EO, and 2025 to bring much more open-sourcing across major labs.
- Researchers think they’re “looking into AGI” and are surprised by outside skepticism; reasoning-model scaling is only at a “ChatGPT 2.0 level” on the curve, with dramatic 2025 breakthroughs expected in coding agents from Google and OpenAI. Google faces “the largest innovator’s dilemma in the history of Silicon Valley” but the breadcrumbs (Gemini, deep reasoning, NotebookLM) have Apple worried for the first time; Elon’s xAI is effectively the hyperscaler for a robotics-autonomy-Grok “keiretsu” with a very low cost of capital.