DeepSeek, Open Source, Tariffs, DOGE, Market Impact | BG2 w/ Bill Gurley & Brad Gerstner

Episode Summary

Executive Summary: The conversation centers on how DeepSeek/R1, tariffs, and Elon Musk’s DOGE effort are reshaping the technology, policy, and market outlook. The speakers argue that open source AI is more innovative and safer than closed systems, that China’s progress and U.S. export controls may accelerate a global open-source ecosystem, and that tariffs and spending cuts inject major uncertainty into valuations, supply chains, and interest rates.

Main Topics: DeepSeek/R1 and the open-source AI shift (Priority: 5/5): They assess DeepSeek as a real algorithmic breakthrough, emphasizing its innovation, MIT license, open weights, and rapid global adoption. The discussion frames open source as a stronger model for safety, competition, and diffusion of innovation. China, U.S. competition, and export controls (Priority: 5/5): The speakers debate whether U.S. attempts to slow China are effective or counterproductive. They argue that constraints can spur innovation, but also that China’s progress could alter global AI power dynamics and heighten geopolitical risk. Tariffs, trade policy, and industrial strategy (Priority: 5/5): They explore whether Trump’s tariffs are primarily negotiation tools or a deeper policy shift toward higher tariffs as a permanent revenue and industrial-policy mechanism. They weigh benefits like reshoring against higher costs and reduced competitiveness. DOGE, budget cuts, and executive power (Priority: 4/5): The conversation explains DOGE as a two-track effort: normal reconciliation budgeting and aggressive executive action to stop wasteful spending. They highlight potential constitutional conflict over the president’s authority to withhold spending. Market uncertainty and valuation risk (Priority: 4/5): The speakers argue that rapid AI change, tariffs, and political/legal conflict all raise uncertainty, which should increase discount rates and compress multiples. They expect more volatility even if long-term fundamentals remain strong. Sovereign wealth fund proposal (Priority: 3/5): They debate a U.S. sovereign wealth fund as a way to manage national assets, but worry about debt, required returns, and crony capitalism. Supporters view it as a way to benefit citizens; skeptics worry about political capture.

Key Arguments: DeepSeek’s success appears to be the result of genuine algorithmic innovation, not simple copying, including better expert allocation and optimization around compute. Open source improves safety, security, free speech, and innovation because it distributes knowledge and allows rapid fork-and-improve competition. The biggest strategic risk is not just China catching up, but the U.S. responding with more restrictions that could backfire and strengthen Chinese incentives to innovate. Tariffs may work as tactical leverage, but a long-term high-tariff regime likely raises costs, reduces competitiveness, and may slow U.S. AI and industrial progress. If tariffs are used on chips or Taiwan-related imports, they function like a tax on AI infrastructure and could reduce GPU demand and research activity. DOGE’s spending cuts could lower deficits and interest rates if markets believe the government will truly reduce wasteful spending. The constitutional fight over DOGE may become a major test of executive impoundment power versus Congress’s appropriations authority. Public market resilience reflects optimism about pro-growth policy, but the speakers argue that uncertainty should still push risk premiums higher. A U.S. sovereign wealth fund could be useful, but only if managed independently and without political interference; otherwise it risks corruption and inefficiency.

Data Points: DeepSeek DAU share in India: ~15% relative to ChatGPT - Used to show DeepSeek’s global adoption DeepSeek DAU share in China: ~10% relative to ChatGPT - Illustrates strong usage in major geographies DeepSeek DAU share in Indonesia: ~8% relative to ChatGPT - Shows broader Rest-of-World traction Hugging Face variants of R1 after launch: 500 variants in 48 hours; 1,300 variants by the time of discussion - Evidence for rapid open-source fork-and-improve dynamics DeepSeek inference price vs OpenAI: About 1/20th of OpenAI - Referenced from API pricing comparisons DeepSeek training run cost: $6 million - Discussed as final training run cost, not total project cost OpenAI 01 training cost: $10–15 million - Used for rough apples-to-apples comparison DeepSeek efficiency vs OpenAI: 30%–50% more efficient - Based on training cost comparison and timing DeepSeek compute capex/TCO: Close to $1 billion when power is included - Estimate of total compute cost discussed in the episode OpenAI compute stockpile: 30,000–50,000 GPUs - Mentioned as the prior comparison point for DeepSeek’s training capability Tariffs announced: 25% on Mexico and Canada; 10%–15% on China - Summarized from the weekend policy shock USAID spending: $50 billion per year - Referenced during the DOGE discussion as an example of targeted spending cuts Budget baseline referenced: Return to 2019 baseline plus 2.5% growth - Used to argue the U.S. could balance the budget by reducing post-COVID excess spending U.S. debt: $40 trillion - Used in the sovereign wealth fund debate Unfunded liabilities: Another $50 trillion - Used to argue the U.S. is a debtor nation Interest rate on debt: About 5% - Used as the hurdle rate for sovereign wealth fund returns

Pivotal Quotes: "it's a better world if the most disruptive LLM model is foreign and open source versus domestic and proprietary" — Bill: Argument that open source is better for safety, security, and free speech "I personally think we have been on the wrong side of history here, and we need to figure out a different open source strategy" — Sam Altman: Referenced from an AMA discussing DeepSeek and open source "pulling up the wall and trying to get people to make a $40. microwave in America is gonna fail" — Bill: Critique of tariffs and reshoring as a blanket policy

Implications: Listeners should expect more AI competition, more open-source diffusion, and more policy-driven volatility. The episode suggests AI winners may face stronger geopolitical and regulatory headwinds, while tariffs and DOGE could materially move markets, rates, and valuations.