AI Sovereignty Wars, Palantir-Nvidia Deal, SCOTUS Birthright Ruling, Newsom’s CA Budget Lie
In a Nutshell
Palantir and Nvidia are pushing "sovereign AI" where enterprises and governments own their models, data, and hardware to prevent frontier labs like Anthropic from stealing proprietary knowledge—Anthropic's pattern of watching customer applications then launching competing vertical products (Claude Design, Claude Legal, etc.) proves the risk. The Supreme Court upheld birthright citizenship 6-3, but California's fiscal crisis shows $1.5-2T in unfunded liabilities, 65% budget growth since 2019, and 15 Fortune 500 companies fleeing, setting up inevitable debt restructuring and pension wipeouts. AI automation will displace specific roles like driving and warehouse work but increase overall employment—high AI spenders grew headcount 10-12% while token costs drop 90% annually, enabling distributed enterprise infrastructure (70% cloud, 20% local, 10% alternative) rather than centralized frontier lab dependency.
These notes were generated by AI and may contain inaccuracies.
Palantir and Nvidia announced a sovereign AI partnership where Palantir will use Nvidia's Neotron open models to build a custom frontier-quality model to serve the US government. Palantir is calling this new platform sovereign AI operating system. The US government agencies will own the hardware, the data, and the model weights. Palantir also shared a viral tweet manifesto laying out the concept: "Data retention is your treasure. Transfer it at your own peril. Transferring that data hands over access to your pre-existing winning plays and yields the means of production for new ones."
Palantir CEO Alex Karp appeared on CNBC to announce the partnership and criticized frontier model companies like Anthropic, stating that enterprise clients are unhappy because they get no value while frontier labs get their IP. Karp questioned whether the Department of Defense should control model weights or if they should outsource the battlefield to Silicon Valley.
The term intelligence sovereignty was coined to describe the need for organizations to maintain control over their proprietary knowledge and data when using AI systems. This concept is distinct from privacy: privacy concerns whether others can access personal information like photos or notes, while intelligence sovereignty concerns whether AI systems can analyze and interpret an organization's data to tell them how to think or compete against them.
Alex Karp's CNBC appearance was characterized as a "televised nervous breakdown" by legacy media, though it was actually addressing a critical point about AI safety in the enterprise context. Technical customers want control over their compute, their models, their data stack, and their alpha (proprietary knowledge). They want to own the means of production rather than transferring it to someone else.
Enterprises are at risk of transferring their knowledge, know-how, trade secrets, and customer data to model providers who might eventually decide to compete with them. The example of Figma demonstrates this risk: according to The Information, Anthropic blindsided Figma with the launch of Claude Design, a new vertical app competing in the design category. Anthropic's chief product officer had served on Figma's board and did not resign until three days before the launch. Figma's stock has fallen approximately 50% this year while Anthropic's valuation has surged.
Anthropic has also launched Claude Science, Claude Security, Claude Legal, Claude Financial, and Claude Code - each expanding into categories previously served by companies building on top of Anthropic's models. The pattern shows Anthropic watching where value is being created on top of their models, then moving to vertically integrate by creating competing applications.
Dario Amodei at Anthropic has argued that open-source models are dangerous and need to be restricted. However, this danger appears to apply to his business model, which requires customers to have limited choice at the model layer. If enterprises want true AI safety, they must retain the ability to choose at the model layer who gets to see and use their alpha.
At 8090, testing was conducted comparing Claude alone versus using a software factory harness with Claude, and the best frontier open-source model alone versus the same harness with the open-source model. The software factory harness with Claude was 1.4x to 4x cheaper and 1.5x faster than using Anthropic Opus 4.8 alone. When wrapping the open-source model with the software factory harness, it was 16.4x cheaper, though three times slower. All testing was conducted using OpenRouter on a traditional hardware stack.
A BCG study examined the return on capital employed (ROCE) of various businesses. With long-term rates moving back to the historical average of 8-11%, this represents the actual cost of borrowing money. Half of large US companies cannot deliver returns that exceed this cost of capital. Additionally, persistently low returns of 1-2-3-4-5% affect approximately one in seven companies worldwide.
An ex-Meta PM pointed out that companies can host open-source models on their own GPUs in US data centers without sharing any data back to anybody. This approach allows companies to avoid giving away data to frontier labs while potentially achieving 100x cost savings compared to proprietary solutions.
Anthropic has been approaching life sciences companies with large proprietary datasets, offering early access to a new life sciences-focused model in exchange for sharing their data under NDAs. Nearly all companies approached have declined, recognizing that tens of billions of dollars invested in experiments and product development represent core differentiating assets that would be commoditized if shared with model companies.
The AI infrastructure model is shifting from a large hub, large spoke model to a large hubs, medium hubs, and distributed spoke model. Enterprises are developing their own weights and models using open-source bases or intermediary business models. They are building their own clusters for training proprietary models using their own data, and running inference locally in their own data centers or enterprise IT closets, while still using NeoClouds and hyperscalers for some inference workloads.
Founders were warned against partnering with frontier model companies like OpenAI or Anthropic, as these companies will eventually compete with and replace their partners' applications. The Microsoft example was cited: Lotus 123 and WordPerfect were replaced by Excel and Microsoft Word. Sam Altman offered $2 million worth of free tokens to Y Combinator startups, which was interpreted as a strategy to gain access to startup innovations and incorporate them into the platform.
Apple was identified as the only company at scale that has respected the developer community by taking a strategic approach to building an app store business. Apple explicitly told developers that if they built something obvious that could be created in a week or two, it might eventually become a stock app. However, Apple deliberately slowed their own apps to make them simple for users and avoid competing directly with their ecosystem, because they wanted to take the 30% tax. There is no equivalent 30% tax structure with Anthropic or OpenAI - instead, they compete directly with companies building on their platforms.
The fundamental issue with frontier model companies is that organizations cannot rent intelligence and judgment from the same place that rents it to their competitors. This creates a lowest common denominator problem where companies and their competitors end up looking exactly the same.
Companies are struggling to generate returns on invested equity, making it economically irrational to pay premium prices for AI models that deliver identical results to competitors. The deflationary nature of technology is driving down costs across the board, with tokens becoming cheaper over time.
Nvidia has developed Neotron, an open-source LLM accessible through Perplexity's dropdown menu with deep thinking capabilities. This model is competitive with Claude for 95% of searches. Jensen Huang previously downplayed Nvidia's open-source model because top customers expressed concern about Nvidia making significant progress on their open-source project. Following OpenAI's announcement of jalapeno chips, Anthropic's chip development, AMD's successful projects with both companies, and Elon's plans to build his own fab, Nvidia has shifted strategy to embrace open-source discussions more openly.
Nvidia aims to own the entire stack by providing both hardware and competitive open-source models at no cost, requiring only use of their hosting companies like Colossus. The offering needed time to mature before becoming compelling enough to compete against established AI labs.
The AI stack consists of three layers: chips at the bottom, models in the middle, and applications at the top. At the model layer, Anthropic and OpenAI dominate with approximately $60+ billion and $40+ billion ARR respectively. No other companies are generating meaningful revenue at the model layer, creating an emerging duopoly situation.
Anthropic is pushing a regulatory capture agenda through safety requirements that would enshrine this duopoly at the regulatory level, with Dario Amodei explicitly stating other models are not safe.
Application companies like Palantir and chip companies like Nvidia have strong incentives to maintain a competitive model layer. Applications need choice to avoid dependency on single providers. Enterprises require options to protect proprietary knowledge. Chip companies need diverse buyers rather than monopsony situations where only one or two companies can purchase their chips.
Open models enable enterprises to roll their own solutions, creating a long tail of buyers ideal for chip manufacturers.
The ecosystem—including chip companies, developers, applications, and enterprises—benefits from competitive model layers. Only Anthropic and OpenAI benefit from maintaining duopoly dominance. While lawful monopolies achieved through performance are acceptable in the U.S. system, government should avoid actions that make monopoly or duopoly more likely and should prevent anti-competitive tactics.
Token costs are projected to decrease 90% annually for the next three years, enabling purchase of 1000x more intelligent tokens at free or near-free pricing. This competition will be disruptive to existing pricing models. Enterprise buying is expected to accelerate beyond just neoclouds and hyperscalers, with companies deploying their own hardware for local inference to avoid cloud dependencies and reduce costs.
Companies are expected to adopt a 70/20/10 distributed approach: 70% in major clouds, 20% local infrastructure, and 10% in alternative clouds. This represents a shift from historical centralized cloud models. Organizations are realizing the benefits of running workflows on open-source models on local machines rather than sending data to third-party cloud providers.
With local hardware, organizations can allow employees to experiment freely with token usage without cost concerns beyond electricity. This enables rapid prototyping of short-lived applications that may burn through billions of tokens.
Microsoft announced a $2.5 billion investment to deploy forward deployed engineers, while Amazon is spending $1 billion on similar initiatives. These engineers are sent to study businesses and integrate them into their models, creating concerns about data lock-in.
The vision includes every employee having a max studio or Dell workstation with substantial RAM, with organizations spending $10-20,000 per employee on local compute. Employees would receive laptops that sync with these local servers, creating a server-per-individual model that maintains data privacy while enabling continuous model customization.
The narrative of AI causing widespread job loss is not supported by current data. Companies adopting AI are growing revenue and headcount rather than simply cutting costs. The media is unlikely to acknowledge shifts in this narrative due to credibility concerns, potentially doubling down on job loss predictions to support regulatory control agendas.
Job displacement differs from job loss—certain jobs will be retired faster than new roles can be created, but overall employment is expected to increase. Customer support, data entry, business process outsourcing, and driving jobs are identified as vulnerable categories. However, data from RAMP and Rellio Labs shows that firms spending most on AI grew headcount by approximately 10% in the two years following adoption, with entry-level headcount growing 12%.
In markets where Waymo has achieved critical mass beyond a few hundred vehicles, human driver recruitment has stopped and driver numbers are either static or declining. This provides current evidence of displacement rather than future predictions.
A study of over 21,000 U.S. firms combining payroll data with AI spending found that high-intensity AI adopters grew fastest and increased headcount. Gains appeared across engineering, sales, administration, and customer service roles rather than concentrated in specific categories. Companies with low or no AI adoption showed flat headcount trends.
Startups are raising hundreds of millions to replace specific roles including translators, telephone operators, bill collectors, and Waymo drivers. Amazon warehouse sorting and factory jobs are identified as targets for automation, with Optimus explicitly designed for this purpose.
Human interaction is expected to command premium pricing as automation increases. Examples include preferring human bartenders despite vending machine alternatives, choosing human drivers in Las Vegas over Waymo vehicles, and paying for human massages despite automated alternatives. Customer service roles requiring escalation handling and higher-value interactions are more difficult to replace than level-one support.
Companies like Clorox reversed decisions to fully automate customer service after recognizing the brand importance of offering human alternatives. Human-in-the-loop approaches are expected to become more valuable and command premium pricing. CEOs attempting to capitalize on media hype cycles by announcing mass automation often reverse such decisions when implementation challenges arise.
Optimus robots deployed inside Amazon factories will eliminate all human package sorting and delivery within 10 years. This automation follows the pattern already seen in consumer electronics manufacturing where human assembly has largely disappeared. Other countries will experience the displacement first while America benefits from entrepreneurial activity and a Cambrian explosion in startups. The driver displacement of 5-10 million jobs mirrors what happened with cashiers.
Fast food ordering has already shifted to self-service kiosks, with McDonald's implementing monitor-based ordering systems. Amazon warehouses and FedEx routing depots already use extensive conveyor belt automation and robots for package movement, though these are not humanoid robots.
The final mile delivery represents the most significant change, with Optimus walking packages from self-driving vehicles to porches, ringing doorbells, and completing deliveries. Construction applications could reduce house building time from 2-3 years in California to one year using crews of 20 humans supervising 50 robots.
Howard Lutnick lifted export controls on Anthropic's Fable 5 on June 30th, while Mythos 5 was restored to US customers on June 26th. The original restrictions stemmed from export controls placed on Fable, Mythos, Iliad, and Odyssey two weeks prior, affecting 50 unauthorized entities including SK Telecom with alleged China ties.
Anthropic replaced Dario Amodei as lead negotiator with co-founder Tom Brown, who showed better relations with the Trump administration by proactively thanking Lutnick via tweet for lifting restrictions. Anthropic's postmortem blog post stated the jailbreak triggering the shutdown was not unique to Claude.
Three conditions created this situation: Dario Amodei had boasted about creating a cyber weapon with Mythos, Amazon reported that Fable's guardrails failed during testing, and Dario refused to roll back Fable until the jailbreak was fixed. Changing any of these three factors would have prevented the government action.
The discussion addressed whether the US should block Chinese open-source models like Kimi and DeepSeek to protect domestic open-source development. Once models are open-sourced, they cease being Chinese because they can be forked, adapted, and run on American hardware without data leakage to China.
Banning open-source would isolate the US since the rest of the world would continue using cheaper, more customizable models. American companies would face a token tax paying for closed models while the rest of the world used open alternatives. The preference is for American open-source models from Nvidia and Google to compete on merit rather than through import restrictions.
Certain Chinese products should face restrictions, including connected cars and potentially robots that could create backdoor access. However, trade relationships require careful consideration since the US still needs Chinese rare earth minerals, and retaliation could result from import limitations.
Trump already banned Chinese drone imports as part of signature legislation. The approach balances national security concerns with maintaining necessary trade relationships.
The All-In Summit returns September 13-15 for its fifth year. Applications are open at allin.com. The event has already attracted 12 trillion in market cap attendees with expectations to double. The summit includes the largest poker tournament yet, gaming at the welcome party, and a full casino night at an exclusive LA mansion.
The Supreme Court struck down Trump's executive order ending birthright citizenship for children of illegal immigrants or those on temporary visas in a 6-3 decision, with Chief Justice Roberts writing the majority opinion joined by all three liberal members. 255,000 children are born annually to non-citizen parents.
Roberts stated that citizenship was historically the right to have rights, and the court upheld this promise. Justice Kavanaugh carved out space for Congress to address birthright citizenship issues consistent with the 14th Amendment.
Birth tourism cases include Dong Wan Lee, a Chinese national who pleaded guilty in 2019 to conspiracy to commit immigration fraud and visa fraud for helping pregnant Chinese women enter the US to secure citizenship for their babies. She served 10 months in prison.
Trump responded by calling on Congress to work on ending birthright citizenship with his complete support. The decision represents an objective loss for Trump, though he plans to continue fighting.
The 14th Amendment was ratified to ensure citizenship for freed slaves following the Dred Scott decision, which ruled that black people had no rights the white man was bound to respect. Senator Jacob Howard's debate transcript specified that citizenship would include every person born within US limits and subject to jurisdiction, but explicitly excluded foreigners, aliens, and families of ambassadors or foreign ministers.
The original intent focused on freed slaves and their descendants, not addressing the massive immigration that occurred later. The textual reading of the amendment conflicts with the intentional reading of its original purpose.
Birthright citizenship should apply to children of legal residents of the United States, whether citizens or not. Temporary visitors on visas or vacation status should not receive citizenship for their children. Legal resident status through green cards creates permanent residency distinct from visitor or illegal status.
For the 20-30 million illegal immigrants already in the country, many were waved in over 30 years under both Republican and Democratic administrations, including during pro-free trade periods like NAFTA. These workers filled jobs at below-minimum wage, contributing to American wealth. A moral obligation exists to provide paths to citizenship for non-criminals and automatic citizenship for their children.
The Supreme Court's interpretation removes Congress's ability to legislate these nuances, creating a rigid constitutional requirement rather than allowing democratic deliberation over edge cases.
The conversation addresses the current state of immigration enforcement, noting that ICE aggressiveness has stopped and the border has been effectively shut, creating a holding pattern. The discussion explores what the path forward should be for legal immigration policy.
Jamaal argues that Western countries are losing their unique cultural representation and that immigration without assimilation harms this. He describes how his parents made an explicit decision to become Canadian first, and how he chose to become American first, with his Sri Lankan ancestry secondary. He warns against seeding immigration that dilutes culture, pointing to the UK, Germany, and France as examples where cultural deviation has been worse for those countries.
David Freeberg presents a framework based on whether immigrants are makers or takers. He argues that if someone's primary motivation is to receive benefits, payments, social services, or support, they should be denied immigration. However, if their motivation is to work, progress themselves, become a maker, create things, generate income, save capital, buy things, and advance their family's position, they should be granted immigration status.
Freeberg emphasizes that productive immigrants create more value than they take in, growing the economy and creating more jobs, making them net positive contributors.
The discussion advocates for a point-based immigration system similar to Canada, Australia, and New Zealand. Points could be awarded for speaking fluent English, understanding American political systems and culture, and committing to not committing crimes or accessing welfare systems. The proposal suggests provisional visas where immigrants must demonstrate they create jobs and economic activity.
The group expresses willingness to accept unlimited numbers of high performers like Elon Musk, David Freeberg, David Sacks, and Chamath Palihapitiya, emphasizing that recruiting the best and brightest represents a gain for America and a loss for competitors like China and Iran.
David Sacks presents polling data showing that 32% of US adults favor deporting all illegal immigrants, while 51% favor deporting some, with nearly everyone in that group supporting deportation of those who committed violent crimes. Only 16% (the DSA group) believe none should be deported. Sacks notes that deportations are popular when immigrants have committed violent crimes or other unacceptable acts.
He references the Ellis Island era when one-third of immigrants returned home because there was no social welfare, and 2% were denied entry due to health or criminal problems.
The group criticizes the current system where immigrants can access $5,000 monthly in welfare, luxury hotel rooms, buses, and flights across the country. They propose a five-year moratorium on social programs for new immigrants, requiring them to create economic activity before accessing benefits.
A proposed system would set economic targets where creating $10 million in economic activity moves someone to the front of the line for review, while $5 million moves them to the next tier.
The conversation shifts to Democratic Party politics, noting that DSA candidates are gaining popularity. Colorado's 15-term congresswoman lost to a DSA candidate, and the group predicts socialism will continue winning in cities. They argue this represents a "quick win" strategy using the appeal of communism and socialism that will ultimately lose moderate and swing voters.
David Freeberg provides a detailed analysis of California's fiscal situation, calling it "the state of California" rather than "the state of the state." He identifies five key problems: ballooning costs, accounting tricks, revenue dependency, exodus and spiraling, and looming unaccounted liabilities.
California's state budget increased from $215 billion in 2019 to $355 billion today, representing a 65% jump. The budget uses accounting tricks including $20-40 billion in debt to create the appearance of balance, when expenses actually exceed revenue.
Personal income tax generates $142 billion of California's $211 billion revenue, with the top 1% (150,000 people) paying $70 billion—half of all income tax revenue. The top 1,000 people pay $22 billion annually. California's top personal income tax rate is 14.4%, and corporate tax is 8.9%.
In comparison, Florida has 5.5% corporate tax, Tennessee 6.5%, and Texas has zero corporate tax, making it cheaper for businesses to relocate.
Since 2019, at least 15 Fortune 500 companies have moved their headquarters out of California, along with 2,100 mid-to-large size companies. At least 5% of California jobs have been lost. The state is experiencing an average annual exodus of 1-1.5% of personal income (AGI), meaning 15% of the state's income leaves over a decade.
New taxes include an 8% sales tax on software subscriptions (Microsoft Word, Gmail, ChatGPT) generating $1 billion annually, and a new tax on healthcare insurance creating $2 billion annually. The temporary 13.3% top income tax bracket from 2012 became permanent and increased to 14.4%.
California has $1.4 trillion in public debt ($500 billion state, $800 billion local), $664 billion in reported unfunded pension liabilities (estimated closer to $1.5 trillion), and $175 billion in retiree healthcare obligations. Total incremental liabilities range from $1.5-2 trillion.
Pension liabilities sit senior to state bonds due to the "California rule," creating a looming fiscal cliff.
Freeberg predicts that California's fiscal crisis will eventually require federal intervention since states cannot declare bankruptcy. This would create a crisis where red states refuse to pay federal taxes to bail out California, potentially threatening the union.
Chamath predicts that rather than federal absorption, California will experience pension wipeouts, negotiated settlements, wholesale constitutional replacement, complete redistricting, and a red wave that dismantles the systems causing poor governance.
David Sacks discusses the proposed "Billionaire Tax Act" that he believes will get on the ballot and pass. He notes Gavin Newsom endorsed more taxes on billionaires after DSA wins in New York. The tax would trigger 200-300 lawsuits taking a decade to resolve through courts, potentially reaching the Supreme Court by 2036.
Sacks argues California had surpluses from Silicon Valley tax revenue but incompetence, selfishness, and corruption caused the crisis. He predicts continued socialism intensification leading to a "giant final confiscation" rather than a red wave or magical fix.
California faces bankruptcy within the next 10 years, requiring complete restructuring of debt and obligations. This restructuring will likely result in many guaranteed pensions being wiped out. Per capita state spending reveals significant disparities: Texas and Florida spend approximately $5,000 per citizen on services, while New York City spends close to $13,000 and California approaches $9,000. Despite spending two to two and a half times more, residents do not receive proportional value in services. The problems in these high-spending states continue to worsen despite increased government spending.
The fundamental issue is competence combined with a culture of self-reliance versus dependency. Blue states attract incompetent people into government positions. Texas exemplifies a different approach where residents receive a ranch and must independently determine solutions for garbage collection, water sources, and energy needs. This rugged individualism and self-reliance prevents the need for excessive taxation rates of 54-60-65%. California residents face guaranteed tax increases to 60-65%, while New York currently sits at 54% and may break 60% under new leadership.
Leftism functions as a system where people construct justifications for taking money from others. The ideology operates like monkeys typing on keyboards, experimenting with words to manipulate people into surrendering their money. This includes virtue signaling, guilt-tripping, and moral accusations, but never actually solving problems. All problems get worse under this approach, with the sole purpose being wealth redistribution.
Mackenzie Bezos donated $26 billion to left-wing causes, yet world hunger remains unsolved despite claims that such donations would solve global problems. This demonstrates that previous arguments about Elon Musk needing to donate similar amounts to solve world hunger were false. The donated money has not produced measurable results beyond enriching leftist NGOs and funding their operations, including purchasing houses and paying higher salaries.
Gavin Newsom's presidential prospects have fallen from a peak of 40% to 20% on Polymarket. His declining support stems from the Democratic Party moving further left toward democratic socialist positions, making Newsom appear too moderate. New York mayoral candidate Zohran Mamdani has emerged as a more left-wing alternative. Newsom attempted to hedge his position by releasing a video endorsing billionaire taxes on the same day he was expected to kill the Billionaire Tax Act, while simultaneously stating opposition to the specific legislation.
California billionaires mistakenly believed Newsom would serve as their political savior and protect their interests, but this assumption proved incorrect. The group chat among billionaires reflects ongoing political discussions about their vulnerability to increased taxation.
The discussion concludes with expressions of gratitude for American citizenship and recognition that the United States represents the best country in the world. Participants acknowledge the need to actively fight to preserve American values and institutions. The conversation includes lighthearted banter about citizenship revocation and references to the Enemy Aliens Act, while maintaining the core message of appreciation for American opportunities and achievements.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
