DOGE vs USAID, Crypto Framework, Google's $75B AI Spend, US Sovereign Wealth Fund, GLP-1s
Speakers
Antonio Gracias
— Founder and CEO of Valor Equity Partners, SpaceX Board Member
David Sacks
— General Partner, Craft Ventures
Chamath Palihapitiya
— CEO, Social Capital
David Friedberg
— CEO, The Production Board
Jason Calacanis
— Angel Investor / Founder, LAUNCH
Summary
The All-In hosts are joined by Valor Equity Partners CEO Antonio Gracias for an episode dominated by DOGE's audit of USAID and the broader question of whether federal spending can be reconciled at all. David Sacks calls in from the White House to lay out the administration's crypto legislative plan, stablecoins first and market structure second, and to defend rescinding the Biden AI executive order in the wake of DeepSeek. The panel then debates Trump's proposed US sovereign wealth fund and dissects Google's 75 billion dollar 2025 capex guide, which Chamath, Gracias and Friedberg all read as a return-on-capital positive rather than a red flag. Friedberg closes with a large VA-database study showing GLP-1 drugs cut risk across many conditions beyond weight loss.
- DOGE's read-only audit of USAID surfaces alleged waste, funding of foreign political opposition and payments to media outlets.
- Gracias argues the federal payment system has no controller and no reconciliation step, which is why agencies cannot pass audits.
- He also says long-dated Treasury yields have started falling as markets take the spending cuts seriously.
- David Sacks says the chairs of the four crypto-relevant committees are committed to legislation within about six months, stablecoins first.
- Sacks defends rescinding the Biden AI executive order, arguing DeepSeek shows China has effectively caught up.
- Trump's executive order tasks Treasury and Commerce with designing a US sovereign wealth fund, possibly anchored by a TikTok stake.
- Google fell 7 percent on earnings after guiding to 75 billion of 2025 capex; the panel reads the spend as manageable and ROIC-positive.
- A VA study of roughly 215,000 GLP-1 patients shows broad risk reduction, including about 30 percent lower cardiac arrest, with mainly gastrointestinal side effects.
Ideas
Long bond yields fall as DOGE bites
Gracias sizes the federal problem concretely: about 6.5 trillion of spending against 4.5 trillion of revenue, with roughly a trillion dollars a year of interest cost. He argues the long end of the Treasury curve had been selling off precisely because bond buyers did not believe Washington could ever stop spending, which they read as a future inflation problem. In the first weeks of DOGE he says that perception is already changing: as the spending cuts prove real and credible, he points to long-dated yields trading back down, which implies the long bond rallies as fiscal credibility is restored.
Avoid government-payer healthcare services
From direct investing experience, Gracias says Valor tried to build businesses that ran on Medicaid and Medicare reimbursement and walked away because they kept finding fraud in essentially every services company they diligenced. He now applies a hard rule at the firm: if a government payer is more than roughly a third of the business, they will not invest. His edge is that government-payer revenue in the services space carries structural fraud and payment-integrity risk, and that risk becomes acute exactly when the government finally starts auditing its own payment flows, which is what DOGE is now doing.
US AI deregulation keeps pace with China
Sacks says the president rescinded the Biden AI executive order, a hundred pages of burdensome rules on American AI companies, and that DeepSeek has proven the decision even more right. The Biden order was written as if the US were the only player in AI, whereas China has basically caught up or is very close. Loading compliance costs onto US labs now simply hands the lead to China. A replacement AI action plan is being drafted in the White House, so the American regulatory regime for AI is deliberately turning permissive, which is a live input for anyone positioned in the AI build-out.
Google selloff is disclosure problem, not fundamentals
Chamath stipulates that Google's base models are probably the best across a broad range of capabilities, so whatever they are spending on is working. The gap is productization rather than model quality: OpenAI's Deep Research is judged faster and better on the margins, which he attributes to post-training and packaging, not to the underlying model. He adds that Google owns an advertising money machine that compounds directly from AI-driven ad targeting, an advantage only Meta comes close to matching. The 7 percent drop after the 75 billion dollar capex guide is therefore a disclosure failure rather than a fundamental one: had Google segregated the number into ad-optimization spend versus speculative pre-training and post-training, he believes the market would have eaten it up.
Google selloff is disclosure problem, not fundamentals
Chamath stipulates that Google's base models are probably the best across a broad range of capabilities, so whatever they are spending on is working. The gap is productization rather than model quality: OpenAI's Deep Research is judged faster and better on the margins, which he attributes to post-training and packaging, not to the underlying model. He adds that Google owns an advertising money machine that compounds directly from AI-driven ad targeting, an advantage only Meta comes close to matching. The 7 percent drop after the 75 billion dollar capex guide is therefore a disclosure failure rather than a fundamental one: had Google segregated the number into ad-optimization spend versus speculative pre-training and post-training, he believes the market would have eaten it up.
Google wins on chips and ROIC
Gracias argues the model layer is commoditizing, in the best case a land war in Asia and in the worst case a pure commodity, so the durable variable is return on invested capital in the data center itself, which depends on how densely, cheaply and quickly it is built. On that test he expects Google to win: it designs and makes some of its own chips, it explicitly manages to return on capital, and it has a monopoly cash machine to fund the build-out. He therefore does not think the AI capex cycle is overblown for Google, and he endorses Chamath's framing that the market simply needs the return-on-capital split disclosed.
Google's $75B capex is a positive signal
Friedberg calls Google probably the most frugal, thoughtful and well-managed computing infrastructure investor of all time, tracing cheap throwaway racks from 1998 to 2005, then energy efficiency, then cloud repurposing, each step lengthening server depreciation from two or three years to six years by 2021. He then runs the math on the 75 billion dollar capex: at a 20 percent return on invested capital Google needs roughly 15 billion of incremental annual profit plus about 12 billion of amortization, so about 27 billion of incremental operating profit a year, which is just under 20 percent of current annual operating profit and not a crazy hurdle. He reads the size of the spend as evidence of a clear line of sight to full utilization and as a proof point that Google is confident it can carry search into chat. Underinvesting would worry him far more than this number.
GLP-1 benefits extend far beyond weight loss
Friedberg walks through a new study mining the VA's anonymized medical records: roughly 215,000 patients on GLP-1 receptor agonists compared with 1.2 million untreated diabetics and 600,000 on other diabetes drugs, a segmentation that isolates the drug effect. The only elevated hazards are gastrointestinal and related, nausea and vomiting, reflux, musculoskeletal complications and sleep disturbance, while risk falls across shock, hepatic failure, respiratory failure, schizophrenia and roughly 30 percent for cardiac arrest. He ties it to Eli Lilly's ongoing clinical programs for additional indications, which he discussed with the company's CEO, and to a gene-expression cascade that turns off inflammatory markers and turns on cellular repair, arguing the benefits are not merely a consequence of losing weight. Phase two data is published and phase three should follow, which would widen the addressable market well beyond obesity.
GLP-1 pills will keep growing dramatically
Chamath makes a demand-side argument for GLP-1 adoption. The competing longevity protocols from Gary Brecka, Andrew Huberman and Brian Johnson are similar enough to be confusing and different enough to impose a real cognitive load on a busy person; he describes paying for doctors in two cities and a third party to reconcile them, and getting worse care for the money. GLP-1 drugs win because they collapse all of that into popping one pill that solves the underlying problem without requiring the patient to assemble a protocol, and an oral formulation is almost here. On that basis he expects these products to keep growing pretty dramatically.
Up Next
This All-In Podcast video, published February 07, 2025, features Antonio Gracias, David Sacks, Chamath Palihapitiya, David Friedberg discussing TLT, Government-payer healthcare services, AI-SECTOR, GOOG, META, LLY, GLP-1 drugs. 9 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Antonio Gracias, David Sacks, Chamath Palihapitiya, David Friedberg · Tickers: TLT, Government-payer healthcare services, AI-SECTOR, GOOG, META, LLY, GLP-1 drugs