AI Sovereignty Wars, Palantir-Nvidia Deal, SCOTUS Birthright Ruling, Newsom's CA Budget Lie

Summary

Theses (episode spine)

• Palantir’s Nvidia-powered “sovereign AI” deal reflects a broader enterprise revolt against handing proprietary data to frontier labs; Karp argues enterprises want to own their compute, models, data stack and “alpha” rather than transfer it to Anthropic/OpenAI, and Sax, Chamath and Friedberg all say they agree with this framing.
• Sax argues Anthropic is running a Microsoft/Google-style playbook — dominate the model layer, then vertically integrate into the lucrative verticals built on top of it — citing Anthropic’s launch of Claude Design blindsiding its partner Figma (whose stock fell roughly 50% this year) plus Claude Science, Security, Legal, Financial and Code as evidence; he also says OpenAI’s equity looks relatively more reasonably priced than Anthropic’s because OpenAI can fall back on a large consumer business.
• Chamath says 8090’s own testing shows that wrapping an open-source model in a proprietary “software factory” harness was 16.4x cheaper than using Anthropic’s Opus alone (though about 3x slower), which he says proves it is now “derelict and irresponsible” for enterprises to keep feeding proprietary data to frontier labs.
• All four hosts converge on the idea that companies must build “AI sovereignty” — running open or self-trained models on their own hardware/on-prem — rather than trust frontier labs; Friedberg frames this as a shift from a large-hub/large-spoke model to large-hub/medium-hub/distributed-spoke inference architecture.
• Sax argues the US should not ban Chinese open-source models (DeepSeek/Kimi) because once an American company forks and runs the model on its own hardware it “stops being Chinese” and carries no data leakage back to China; Jason instead argues for import-style restrictions to force more investment into competitive US open models like Nvidia’s Nemotron.
• Sax and Jason sharply disagree on AI job displacement: Jason insists customer support, driving, data-entry and warehouse jobs are already disappearing (citing Waymo’s driver counts going flat once it hits critical mass in a market), while Sax counters there is no present data showing job loss, citing a Ramp/Rho Labs study of 21,000 firms showing AI-adopting companies grew headcount roughly 10% (12% at entry level) versus flat headcount at non-adopters.
• Sax says the brief export-control letter that temporarily blocked Anthropic’s Claude/“Mythos” (Fable 5) model was a narrow, fact-specific episode — driven by Daario’s own “cyber weapon” framing, Amazon’s report that its jailbreak guardrails failed, and Anthropic’s initial refusal to patch — that was resolved within two weeks once co-founder Tom Brown took over negotiations from Daario, not a signal of a broader shift in export policy.
• On the SCOTUS birthright-citizenship ruling (Trump v. Barbara), Sax says the 14th Amendment was originally meant to protect freed slaves and personally favors granting birthright citizenship to children of legal residents but not to illegal or temporary visitors, arguing the ruling improperly strips Congress of the ability to legislate the edge cases; Friedberg argues the US has a moral obligation to grant a path to citizenship for long-resident, non-criminal undocumented immigrants who were “waved in” over past decades.
• Friedberg argues Newsom’s claimed “balanced” $351B California budget is disguised by $20-40B of borrowing/accounting tricks, sits atop $1.4T in existing public debt plus $664B (potentially ~$1.5T) in unfunded pension liabilities and $175B in retiree healthcare obligations, and masks an accelerating exodus of high earners and corporations (1-1.5% of the state’s AGI leaving annually), projecting recurring ~$40B annual deficits by 2028-29.
• Friedberg predicts a scenario in which a future Democratic administration (he floats AOC as a likely 2028+ president) could federalize California’s liabilities and trigger a broader “crisis of the union” as red states balk at bailing out blue states; Chamath instead predicts California’s pensions get wiped out via a negotiated restructuring and a subsequent political “red wave,” not a federal bailout.

$PLTR (Palantir Technologies)

Convergence / divergence: Only Sax explicitly discussed Palantir’s strategic positioning; no other host offered a competing or corroborating price call, so there is no convergence/divergence to synthesize beyond Sax’s single thesis.

Speaker calls:

• Sax (bullish, unspecified timeframe): Sax argues Palantir is strategically well-positioned by partnering with Nvidia because application-layer companies like Palantir want a competitive model layer rather than dependency on the Anthropic/OpenAI duopoly, echoing Karp’s argument that enterprises must retain control over their compute, models, data and proprietary knowledge.

Cross-check:

• Price: $129.78 (P/E ~141.7x, mkt cap $309.7B). Next earnings: 2026-08-10.
• Recent headlines worth knowing: Q1 2026 revenue up 85% YoY to $1.633B, beat estimates; EPS $0.33 beat $0.28 forecast; guided Q3 EPS $0.37 and Q4 EPS $0.39 on rising revenue projections ($1.8B, $1.986B).
• ⚠️ Inconsistencies: Valuation remains extremely elevated (P/E >140x) relative to most software peers; bullish thesis on the podcast rests on strategic positioning versus AI model providers rather than on the stock’s current multiple.

$NVDA (Nvidia)

Convergence / divergence: Jason and Sax both see Nvidia benefiting from promoting a more open, competitive AI model layer, though for slightly different reasons — Jason emphasizes Nvidia’s own model (Nemotron) becoming directly competitive, while Sax frames it as a chip-vendor incentive to avoid a monopsony of one or two buyers building their own chips.

Speaker calls:

• Jason (bullish, unspecified timeframe): Jason says Nvidia’s open-source Nemotron model is now competitive with Claude for most searches and argues Nvidia is going to “take the gloves off” and own the full AI stack (chips, open model, and hosting) now that OpenAI, Anthropic and AMD have all announced their own chip efforts.
• Sax (bullish, unspecified timeframe): Sax says chip companies like Nvidia want a competitive, diverse model layer rather than a monopsony of one or two buyers who make their own chips, so Nvidia has an incentive to promote open models that widen the pool of buyers for its hardware.

Cross-check:

• Price: $194.83 (P/E ~29.8x, mkt cap $4.73T). Next earnings: 2026-08-26.
• Recent headlines worth knowing: P/E has compressed roughly 33% versus its 12-month average (44.35x) even as stock trades near record highs; OpenAI, Anthropic and AMD have all announced their own custom chip efforts, a competitive dynamic directly referenced on the podcast.
• ⚠️ Inconsistencies: none flagged

Topics discussed

Palantir-Nvidia “sovereign AI” deal and enterprise AI sovereignty

Summary: Palantir and Nvidia announced a partnership using Nvidia’s open Nemotron models to build a custom frontier-quality model for the US government, with agencies owning the hardware, data and model weights. Karp used a CNBC interview to argue enterprises are unhappy handing their proprietary data (“alpha”) to frontier labs like Anthropic and OpenAI, framing this as an AI-safety and sovereignty issue.

Speaker views:

• Jason: Jason says he has been calling this “intelligence sovereignty” since February — that companies should not train other people’s AI with their own knowledge, and open-source/local hardware are key to consumers and companies rolling their own models.
• Sax: Sax says Karp is right that enterprises want control over compute, models, data and alpha, and warns that Anthropic and OpenAI both have an incentive to vertically integrate into applications built on top of their models, citing Anthropic’s Claude Design launch blindsiding its Figma partnership.
• Chamath: Chamath cites 8090’s own testing showing an open-source model wrapped in their proprietary harness was 16.4x cheaper (3x slower) than using Anthropic’s Opus alone, and argues it is now “derelict and irresponsible” for enterprises not to pursue an independent path to AI.
• Friedberg: Friedberg says life-sciences companies are resisting Anthropic’s attempts to get them to contribute proprietary data to a shared model, and predicts the industry is shifting from a large-hub/large-spoke model to a large-hub, medium-hub, distributed-spoke architecture where enterprises train and run more of their own models on their own hardware.

Potential impact: Sax argues this dynamic could push a growing pool of enterprise buyers toward open models and their own hardware, which he says benefits chip makers like Nvidia and application companies like Palantir by keeping the model layer competitive rather than letting a duopoly (Anthropic/OpenAI) capture enterprise data and application value.

Anthropic’s export-control letter and “palace intrigue”

Summary: The US Commerce Department briefly restricted exports of Anthropic’s Claude/“Mythos” (Fable 5) model after Amazon reported that a jailbreak defeated its guardrails, then lifted the restriction after roughly two weeks once co-founder Tom Brown replaced Daario as Anthropic’s lead negotiator with the administration.

Speaker views:

• Sax: Sax says the reversal required three specific conditions — Daario publicly calling Mythos a “cyber weapon,” Amazon (a trusted partner) reporting the jailbreak, and Daario initially refusing to roll back the model — and that people should not overextrapolate this into a broader shift in US AI export policy, since the administration remains pro-export and pro-infrastructure.

Chinese open-source AI models and import/export policy

Summary: Jason asked why the US allows import of Chinese open-source models like DeepSeek and Kimi while restricting other Chinese technology (Huawei, self-driving tech), suggesting an import ban to favor US open models like Nvidia’s.

Speaker views:

• Jason: Jason argues the US should consider blocking Chinese open-source models the same way it blocks Huawei or Chinese self-driving technology, to strengthen domestic open-source alternatives.
• Sax: Sax argues that once a Chinese open-source model is forked and run on US hardware in a US data center, it “stops being Chinese” with no data leakage back to China, and that banning it would isolate the US since the rest of the world will keep using cheaper, more customizable open models; he says the US should instead focus on building better American open models and not force anyone to use models they don’t want, while remaining open to restricting other categories like Chinese connected cars or robots.

Potential impact: Sax notes that restricting Chinese technology imports risks retaliation given the broader US-China trade relationship, including US dependence on Chinese rare earths.

AI and job displacement

Summary: The hosts debated whether AI is currently destroying jobs. Jason argued certain job categories (customer support, driving, data entry, warehouse sorting) are already disappearing or will soon, while Sax argued there is no current data supporting a job-loss narrative, citing a study of AI-adopting firms’ hiring trends.

Speaker views:

• Jason: Jason says customer support, entry-level data entry/BPO jobs, and driving jobs will be retired quickly, pointing to Waymo’s driver counts flattening once it reaches critical mass in a market and to Amazon warehouse sorting eventually being automated (comparing Figure’s and Optimus’s robot progress).
• Sax: Sax cites a Ramp/Rho Labs study of over 21,000 US firms showing companies that spent the most on AI grew headcount roughly 10% (12% at entry level) in the two years after adoption, while low/no-AI-adoption firms saw flat headcount, concluding there is no present data showing AI-driven job loss, only “job displacement” as some roles shift.
• Friedberg: Friedberg agrees there is job displacement rather than net job loss, and predicts a premium will emerge for human-in-the-loop service, citing Klarna’s reversal of its plan to replace its customer service department entirely with AI.

Potential impact: Jason argues other countries doing lower-skill entry-level work will feel job displacement most acutely first, while the hosts expect the US, with its entrepreneurial culture, to see a “Cambrian explosion” of new startups and job categories instead.

SCOTUS birthright citizenship ruling (Trump v. Barbara)

Summary: The Supreme Court struck down Trump’s executive order ending automatic citizenship for children born to illegal immigrants or temporary visa holders, with Chief Justice Roberts writing for the majority (joined by the three liberal justices) while Kavanaugh left room for Congress to legislate around the edges.

Speaker views:

• Sax: Sax says the original intent of the 14th Amendment was to secure citizenship for freed slaves, and personally believes birthright citizenship should apply to children of legal residents but not to illegal or temporary visitors; he says the ruling wrongly removes Congress’s ability to legislate the harder edge cases like long-resident undocumented immigrants or birth tourism.
• Friedberg: Friedberg says the US has a moral and ethical obligation to give a path to citizenship to non-criminal undocumented immigrants who have been in the country for decades (and to their children) because the country effectively “waved them in” to work for below minimum wage over past administrations.

California’s state budget and fiscal exodus

Summary: Friedberg laid out Newsom’s final budget as $351B and claimed to be balanced, arguing it actually relies on $20-40B of borrowing/accounting tricks while the state faces a shrinking high-earner tax base, new sales taxes on software and health insurance, and roughly $1.4T in existing public debt plus large unfunded pension and retiree healthcare liabilities.

Speaker views:

• Friedberg: Friedberg details that California’s budget grew 65% since 2019, that the top 1% of earners pay roughly a third of the state’s personal income tax revenue, that at least 15 Fortune 500 companies and 2,100 mid/large firms have left the state since 2019, and that recurring ~$40B annual deficits are projected for 2028-29 on top of ~$1.5-2T in unfunded pension and retiree healthcare liabilities, which he says could eventually force a federal bailout and provoke red-state backlash.
• Chamath: Chamath agrees the fiscal picture is dire but predicts the resolution will instead be a wipeout of California pension obligations via a negotiated settlement, a rewritten state constitution, redistricting, and a political “red wave,” rather than a federal absorption of the debt.
• Sax: Sax says he expects the state to double down on tax increases (citing the proposed billionaire tax act) rather than reform, and predicts a decade of litigation if the billionaire tax passes, while noting his own move to Texas reflects the trend of high earners leaving.

Potential impact: Friedberg argues that if the federal government were to bail out California’s debts, red states would question why they should keep funding the union, which he frames as a potential trigger for a broader crisis of the union within the next decade.