The AI Cold War, Signalgate, CoreWeave IPO, Tariff Endgames, El Salvador Deportations

Summary

• Nvidia’s accounts-receivable expansion is best explained by a Blackwell installation bottleneck—unless it persists beyond the July quarter. Gavin Baker called the rise from roughly $1.5 billion to $5.5 billion “never good,” but argued that Hopper-to-Blackwell is an unprecedented physical transition: racks move from about 1,000 pounds, 60 kW and air cooling to 3,000 pounds, 120 kW and liquid cooling. He rejected the round-tripping thesis because Nvidia could have sold those GPUs to Meta, Amazon or Microsoft without investing in neoclouds.
• CoreWeave’s IPO is a referendum on whether GPU-cloud operations are a commodity or a scarce execution capability. The proposed deal was cut to $1.5 billion at a $23 billion valuation after $2 billion of revenue, alongside nearly $8 billion of debt and Microsoft contributing more than 60% of sales. Baker’s counter to the bearish consensus: synchronizing tens of thousands of failure-prone GPUs is hard, and CoreWeave “runs these big GPU clusters as well as anyone.”
• The AI infrastructure cycle did not look overbuilt to Baker after DeepSeek-R1. China was buying every available GPU, DRAM pricing was rising, and OpenAI had begun gating image generation for lack of compute; meanwhile, hyperscalers face a prisoner’s dilemma in which underspending on Blackwell might concede a durable advantage. His analogy: Blackwell is a Formula 1 car from ten years in the future that operators must learn to drive.
• Export controls buy the US time while intensifying China’s incentive to build around Nvidia. Baker described enforcement as a “game of cat and mouse”: friction can preserve an American advantage, but necessity encourages domestic semiconductors and algorithmic advances such as DeepSeek. He put China’s five-year chance of matching or beating Nvidia at “zero,” then added, “Over the 10 years, who knows?”
• Agents could make compute the binding constraint while potentially collapsing incumbent cost structures. Baker saw MCP becoming a standard that lets services such as Stripe connect once and work across models; if agents become real, Blackwell ROI rises even though it is “not good for human employment.” Friedberg envisioned three or four people tackling projects previously requiring 300 or 400 specialists, while Chamath argued that one- or two-order-of-magnitude Opex reductions could “blow a hole” in the $3.5 trillion–$4 trillion software complex.
• Trump’s tariff program is a high-risk policy stack requiring reciprocal rates, deregulation, tax relief and domestic ownership of IP to move together. Baker framed reciprocity as “if you charge 5%, we charge 5%,” while saying every mention of tariffs should be accompanied by deregulation “two or three times.” Friedberg described the proposed elimination of taxes below $150,000 as the consumer offset to higher import prices; Jason likened the full maneuver to “four freight trains” exchanging sandwich ingredients through their windows.
• The deficit arithmetic leaves DOGE carrying far more weight if tariff revenue disappoints. Against a stated $1.9 trillion deficit, Chamath contrasted Howard Lutnick’s trillion-dollar revenue ambition with Wall Street estimates of only $170 billion–$300 billion, concluding that Musk might need to find roughly $1.75 trillion rather than $1 trillion. Baker thought the administration had a coherent theory and would adapt within six to nine months if it failed, because “the one thing they can’t afford is a recession.”
• Signalgate and the CECOT deportations turned execution, accountability and due process into threats to the administration’s larger mandate. The Red Sea operation had material stakes—Chamath cited shipping-price increases of 30%–40% into the US and 300%–400% into Europe—but Chamath and Jason regarded sensitive planning on personal Signal devices as a preventable mistake. On deporting 238 alleged gang members, Friedberg rejected a 4%–5% innocent-error trade-off; Jason emphasized due process and the prison’s conditions, while Baker’s warning was simple: “Execution matters.”

Deep dive

1. Nvidia’s receivables have a July-quarter falsification point

• Baker rejected the round-tripping accusation around Nvidia’s neocloud investments: without those investments, he believed the company would simply have sold more GPUs to Meta, Amazon and Microsoft. Funding alternative buyers fragmented a stable three-customer cloud oligopoly and reduced the hyperscalers’ bargaining power.

• CoreWeave also adopted Nvidia’s standard reference architecture, getting GPUs into service faster and, in Baker’s view, more smoothly than hyperscaler-specific server designs. Chamath compared the ecosystem strategy to Intel Inside: support and brand the underlying platform until buyers ask what is inside.

• The rise in accounts receivable from roughly $1.5 billion to $5.5 billion was “never good,” Baker conceded, but the context mattered. Hopper-to-Blackwell was “the biggest product transition in history” for semiconductors, with reasoning models such as DeepSeek-R1 consuming enough compute to let Nvidia grow through the transition despite an Osborne-like prelaunch lull.

• The physical change explains why customers might delay payment until systems work: Hopper racks were described as roughly 1,000 pounds, 60 kW and air-cooled; Blackwell as 3,000 pounds, 120 kW and liquid-cooled. It is an iPhone connector change that also requires “a new generator,” boiler and humidification system. Persistence beyond the July quarter would concern Baker.

2. CoreWeave may be operationally scarce, not computationally generic

• At taping, CoreWeave planned to raise $1.5 billion—down from a reported $2 billion—at a $23 billion valuation. The host’s snapshot included $2 billion of prior-year revenue, eightfold growth, nearly $8 billion of debt, Microsoft above 60% of revenue, and Nvidia at 15% while owning 6%.

• Baker was categorical that infrastructure was “definitely not overbuilt yet.” After DeepSeek-R1, China was absorbing every GPU it could obtain, a large Chinese server manufacturer was warning of shortages, DRAM prices were rising daily, and OpenAI was gating its image-generation service because it lacked GPUs.

• Hyperscalers also face a prisoner’s dilemma: Meta cannot risk underspending while Google advances. Blackwell is much better but initially harder to tune—like replacing a familiar Formula 1 car with one from ten years in the future and relearning how to drive at the limit.

• Against complaints about leverage, capex and customer concentration, Baker offered a retail analogy: operating 1,000 clean, stocked stores across 50 states sounds simple yet creates businesses worth more than $10 billion. Likewise, synchronizing tens of thousands of melting or disconnected GPUs without losing training work is rare competence. Separately, Chamath said CoreWeave’s acquisition of Weights & Biases could further differentiate and decommoditize the company.

3. Export controls trade short-term friction for long-term Chinese innovation

• Baker doubted the US could fully prevent GPU diversion when it cannot keep illegal drugs out and China actively wants the hardware in. Until a broader bargain emerges, enforcement remains a “game of cat and mouse.”

• The controls still impose enough friction to give America a theoretical advantage, but they create an “immense incentive” for China to build its own semiconductor ecosystem and find algorithmic workarounds. Baker treated DeepSeek as evidence that pressure had already produced real algorithmic innovation.

• Asked whether China could build something competitive with or better than Nvidia, Baker answered “zero” over five years because the task is exceptionally hard. His ten-year answer was deliberately open: “Who knows?” Ten years is brief for the CCP and “an eternity” for America.

4. Agents make compute the rate limiter

• Baker saw Manus as a possible ChatGPT-like moment for agents, but the more structural development was Anthropic’s Model Context Protocol and OpenAI’s adoption of it. If Stripe integrates once with MCP, any compatible model can interact with Stripe, removing a major standardization problem.

• If agents materialize, Baker expects very high ROI on AI infrastructure and Blackwell, alongside an employment cost: “It’s not good for human employment.” Yet the immediate constraint becomes compute because agents operating all day would consume far more capacity than exists; proposed first-agent pricing of $10,000–$20,000 per month reflected that scarcity.

• His near-term answer to the AI-ROI debate therefore came down to agents. Useful autonomous work would validate infrastructure spending, but widespread availability must wait for enough compute to be installed.

5. Small teams could attack the software industrial complex

• Friedberg pushed beyond automating easy human tasks. To plan an underwater plant-breeding facility today, he would need experts in oceans, underwater engineering, materials, physics and construction; agents might let two to four people specify every step, workload and cost.

• He applied the same logic to California’s roughly $100 billion high-speed-rail program: agents could reduce organizational confusion and potential grift while making complex projects executable. His broader claim was that teams of three or four might perform work that once required 300 or 400 deeply technical people across biotech, manufacturing, urban design and transportation.

• Chamath translated that into incumbent risk. The “software industrial complex” is a $3.5 trillion–$4 trillion industry growing about 10% annually—roughly $300 billion of added enterprise value—but a three-person agentic company could attack a 30,000-person incumbent. If Opex falls one or even two orders of magnitude, existing cost structures cannot compete.

6. Tariffs are a reciprocal-reshoring bet, not a temporary impulse

• Baker began with the conventional comparative-advantage case against tariffs, then said the administration was convinced that the previous 20 years served knowledge workers better than ordinary Americans. Its objective is to restore high-quality manufacturing jobs, even as automation raises doubts about how many humans reshoring will employ.

• Baker framed tariffs as a level-setting device after asymmetric trade: “If you charge 5%, we charge 5%; if you charge 10%, we charge 10%.” He pointed to Trump’s 1987 Larry King appearance and New York Times advertisement as evidence that the position was longstanding.

• The host noted a 25% automobile tariff and a still-moving policy ahead of April 2. Baker’s practical caution was that every use of the word “tariff” should come with “deregulation” two or three times, because easier domestic business formation improves the odds that reshoring actually occurs.

• Baker’s governing instinct came from a former Navy SEAL who had deployed to 80 countries: Americans obsess over making the country better, but “our only goal should be to not screw it up.” Tariff policy therefore required the delicate balancing he emphasized.

7. The policy stack has to move money and IP together

• Chamath added intellectual property to manufacturing and deregulation. While helping run Facebook, he was a signatory when the company established operations abroad and exported its IP to Ireland, followed by a decade-plus IRS dispute; his conclusion was that policy should keep both production knowledge and economic value inside America.

• He also argued that foreign companies operating in the American market might import IP if given a workable mechanism. Getting those incentives right could produce “a renaissance for the United States,” rather than merely relocating assembly work whose underlying value remains abroad.

• Friedberg said administration officials had discussed eliminating taxes for people earning under $150,000, a proposal he believed could happen. Tariffs remove consumer purchasing power while adding government revenue; simultaneous tax cuts return money to households, while spending cuts make the lower tax intake affordable.

• Deregulation must then direct incremental capital toward automation and onshore supply so costs eventually fall. Friedberg called the whole package a “grand economic experiment.” Jason compared the maneuver to four high-speed freight trains passing one another while sandwich ingredients are transferred between their windows.

8. The deficit math puts DOGE on the critical path

• Chamath cited John Arnold’s $1.9 trillion deficit estimate for both the current year and 2026. Lutnick’s stated bridge was roughly $1 trillion of revenue plus $1 trillion of Musk-led savings, but Wall Street estimates for the revenue side ranged from $170 billion to $300 billion.

• Using roughly $200 billion–$250 billion as the market midpoint, Chamath’s stated conclusion was that DOGE might need to find $1.75 trillion rather than $1 trillion. That makes waste, fraud and abuse—not merely tariffs—the critical variable in the balanced-budget case.

• Chamath cited one multibillion-dollar consultancy deriving more than 95% of revenue from US time-and-materials contracts, $65 billion paid across consulting organizations, and what he described as a canceled $2 billion award associated with Stacey Abrams; he said the Justice Department was investigating how the money was given to her. He also suggested military spending could potentially be cut by 8% annually for a couple of years.

• Baker defended the search for savings with an immediate-family example: a stolen Social Security number had allegedly collected Alabama unemployment benefits for two years despite attempts to stop it. Jason nevertheless wanted a quarterly or semiannual DOGE report whose waste, overspending and fraud categories were “audited and triple-checked.”

9. The Red Sea mission explains the urgency, not the Signal channel

• Jason’s account of Signalgate was stark: Atlantic editor Jeffrey Goldberg was somehow added to a senior national-security Signal group where Houthi strike targets, actions and timing were discussed. Hegseth appeared to paste operational material from another system, creating both device and clipboard-security concerns.

• Chamath insisted the missing context was the Houthi disruption of Red Sea and Suez traffic. Rerouting around the Cape of Good Hope pushed shipping costs up an estimated 30%–40% toward the US and 300%–400% toward Europe, threatening inflation and productivity; by the third week of March, he said volumes had recovered to year-earlier levels.

• The chat also exposed an alliance asymmetry: roughly 3% of US trade but 40% of European trade used the Suez route, yet America had the capability to act. Gavin found the officials’ phrase “our side of the ledger” revealing—behind hostile public rhetoric, they still treated Europe as part of the home team.

10. Signalgate exposed incompatible views of secrecy and accountability

• Chamath opposed firings, called the incident a mistake and argued that inherited security procedures and device management should be rewritten. Jason separately called Signal’s desktop app “garbage,” warned that multiple personal endpoints could multiply attack vectors, and said the incident should be forgotten “in 2 Scaramuccis.”

• Jason argued that secure compartmented facilities and other secure channels existed, so officials should not put attack planning on personal phones. Copying classified material through a clipboard was another known exposure; the administration should own the failure, secure the devices and explain the correction.

• Friedberg grounded the records dispute in the 1950 Federal Records Act, its 2014 electronic-communications amendment and court rulings requiring preservation of official communications. Yet his personal position was narrower: decisions and actions should be retrievable, while making every deliberative comment FOIA-accessible would chill candid, productive government discussion.

• Chamath also questioned whether Goldberg behaved ethically by remaining in the group; Jason said the journalist verified what he was seeing, contacted officials and then left. Rubio and Trump acknowledged a mistake, but Jason argued that attacks on the journalist, paired with Republican outrage over Hillary Clinton’s email practices, made the response look evasive and hypocritical.

11. CECOT’s crime result came with the due-process cost

• The administration sent 238 alleged gang members to El Salvador’s severe CECOT prison under the Alien Enemies Act of 1798, which the host said permits the president to detain and deport people from an enemy nation without due process during an invasion or “predatory incursion.” Jason said most, but not all, had US criminal records; a federal judge paused the transfers, yet the administration apparently continued them.

• The possible errors were concrete: former soccer player Jerce Reyes Barrios allegedly had a Real Madrid tattoo and an “I love you” hand sign misread as gang evidence. Shoe salesman Norberto Rodríguez’s cards-and-dice tattoo reportedly covered a scar, while a gay stylist and makeup artist named Andrés Hernández sought asylum but was flagged through tattoos his lawyers said were unaffiliated.

• Friedberg opposed imprisonment or detention without due process, comparing the value problem to Guantánamo Bay. If people entered illegally or belonged to criminal enterprises, he preferred returning them to a port of departure for trial elsewhere—or trying and imprisoning them in the US.

• He nevertheless explained the El Salvador precedent: homicides fell from 103 per 100,000 in 2015 to 1.9 after mass detention without normal process. One innocent man reportedly spent more than a year imprisoned yet said the societal result was worth it—an account Jason treated cautiously because nobody knew what pressures surrounded it.

12. Error correction is the administration’s mandate test

• Asked whether they would accept a 4%–5% innocent-error rate for El Salvador’s crime reduction, Friedberg rejected compromising due process. Jason did not separately answer that hypothetical but agreed that due process was important and opposed sending people to CECOT without processing. Chamath first said he would not compromise the value, then noted that US prisons might already have an error rate above 5% and said the president had been explicitly elected to make difficult security judgments.

• Jason argued that the identification of three possible errors showed some review was occurring; Chamath countered that the information came from families and the press, not the administration, and asked what about the other 235 people. Jason answered that discovering errors after people reach CECOT is not prior due process and that the exceptionally punitive destination is central to the objection.

• Friedberg suggested the transfers may be intended as a deterrent: the first roughly 240 people might prompt gang-affiliated immigrants to leave and discourage others from coming, creating a claimed net benefit even if several innocents were swept up. Jason called the promotional prison imagery and “sadistic” messaging an Achilles’ heel and predicted it could cost the administration the midterms.

• Gavin’s synthesis was ends versus means. He acknowledged compassion for Americans harmed by known foreign criminals, but said innocent imprisonment would be a terrible mistake requiring immediate correction. With an ambitious economic and political program, the administration can absorb only so many Signalgates or wrongful deportations before losing its mandate: “Execution matters.”