Summary

The tariff campaign is simultaneously a border lever, an industrial-policy experiment, and a bid to rebase America’s reliance on the U.S. dollar and its ability to fund deficits. Joe Lonsdale reads the Canada/Mexico threats primarily as fentanyl negotiations; Chamath Palihapitiya sees a broader attempt to rebase dollar dependence, deficit funding, and corporate competitiveness. David Friedberg supplies the connective tissue: tariffs, lower income taxes, and lower government spending form “three legs on a stool,” but erratic on-off implementation can make the investment case impossible to underwrite.
The investable version of tariffs favors commodities and sectors with domestic substitutes, while unique technologies remain inflation traps. Chamath calls autos a workable case because many global manufacturers can absorb reciprocal tariffs, but an irreplaceable ASML machine, innovative drug, or single-vendor technology can simply become 25%-30% more expensive. Predictability matters as much as the rate: no executive will commit $10 million to a U.S. factory if the tariff may disappear the following week.
The administration may tolerate equity weakness if it suppresses consumption, inflation, and long-term Treasury yields. Chamath’s mechanism is explicit: lower stock prices reduce the wealth effect, money rotates toward the 10-year, and cheaper borrowing helps refinance roughly $1 trillion due over the next 9-12 months. His larger call is that MAGA’s durable coalition includes working- and middle-class people who do not necessarily own many stocks or homes, pro-innovation and pro-tech people, and patriotic business owners, so policy may be viewed through Main Street rather than Wall Street.
DOGE’s easy waste targets quickly collide with the harder constraint that AI requires vastly more electricity. The panel cited 35,855 ServiceNow licenses across three products with only 84 users and 11,000 Acrobat licenses with zero users, yet Chamath warned that renewables supplied almost 91% of December’s incremental generation while 35,000 projects awaited FERC approval. With a new gas turbine unavailable before 2030 and nuclear before 2035, indiscriminately repealing IRA incentives could leave America without “enough electrons” for GPU clusters scaling from 100,000 toward one million units.
CoreWeave offers spectacular growth wrapped around a difficult asset-liability bet. The neocloud operates 32 data centers and 250,000 Nvidia GPUs, reported $1.9 billion of 2024 revenue, and was expected to raise at least $3.5 billion above a $30 billion valuation—but it also carried almost $8 billion of debt, lost nearly $1 billion, and derived about 60% of revenue from Microsoft. The decisive underwriting question is whether GPU useful life resembles ten years or five: “This business is deeply underwater” if management amortized the hardware too slowly.
The AI model layer is becoming abundant enough that distribution and real-world workflow integration matter more than benchmark launches. Chamath preferred Claude 3.7 for code and Grok 3 for consumer use, while calling GPT-4.5 merely “good” and warning that model makers are overfitting SWE-bench, IMO, and AIME. His metaphor captures the market: “100 Michelin-star restaurants” have opened at once, leaving users overwhelmed while X, Google, YouTube, and Meta turn existing distribution into the durable advantage.
Sacks’s Bitcoin policy creates a long-term reserve, a managed stockpile, and a disclosure-first regulatory framework. Bitcoin forfeitures enter a “digital Fort Knox” that Treasury may not sell; other adjudicated digital assets enter a stockpile Treasury may rebalance, while budget-neutral strategies can add Bitcoin. Sacks estimated the federal government once held 400,000 BTC and sold roughly half for $360 million—coins later worth more than $17 billion—and said he divested roughly $85 million personally attributable to him before day one of the administration to remove the appearance of self-dealing.

Deep dive

1. Tariffs are being asked to solve three different problems

Jason Calacanis framed the week as policy whiplash: Canada and Mexico tariffs moved markets sharply, exemptions followed, and TSMC simultaneously announced a $100 billion U.S. investment. The central puzzle was whether this was negotiation, industrial policy, or a broader bond-market and financial strategy.
Lonsdale’s narrowest interpretation was fentanyl leverage. He said tens of thousands of young Americans had died, Canada had done “nothing about their border,” and Trump was using trade pressure to force action. Lonsdale noted that Canada was still on while Mexico might be off, and that importers still needed to know what the rules were.
Chamath saw a genuine schism after the Trump-Elon honeymoon: one camp viewed DOGE and tariffs as deranged chaos, while another believed the administration was “sticking to the plan.” His definitive formulation was that tariffs could rebase dollar reliance, deficit financing, and the long-term economic vitality of American companies.

2. Friedberg’s master plan is a three-legged fiscal stool

Friedberg’s optimistic synthesis joined tariffs with lower income taxes and lower government spending. His greenhouse LEDs had just become 25% more expensive, creating a crossover point where manufacturing them domestically might finally beat sourcing them from Asia—one example of tariffs redirecting industrial capital.
Lower corporate and personal income taxes would then leave capital available to fund the newly viable factories. Friedberg also entertained moving from income taxation toward consumption taxation, with tariffs functioning as a tax on spending, but stressed that America had not run that experiment for roughly 150 years and its outcome remained contested.
Spending cuts complete the stool by moving labor from government-supported work into private production while offsetting tariff inflation. Friedberg cited government’s roughly 30% share of GDP; Lonsdale translated the proposition more bluntly as redeploying “a million workers” from Washington’s consulting and paper-pushing economy into productive businesses.

3. Tariffs work cleanly on commodities and dangerously on scarcity

Chamath’s “other side of the tariff knife” separates competitive markets from irreplaceable products. Autos have many manufacturers and can support reciprocal treatment; an innovative drug or sole-source ASML machine cannot, because the supplier may simply pass a 25%-30% increase downstream. That is inflationary and problematic if the product is something people must have.
Chamath added an environmental argument: American producers bear cleaner-air and cleaner-water rules while overseas producers may externalize those costs. A tariff can compensate for that asymmetry rather than rewarding production that “shits all over the environment.”
The panel converged on predictability and reciprocity. Friedberg cited the EU’s 10% tariff on American cars against America’s 2.5% levy on European cars; matching rates seemed defensible, but turning tariffs on and off leaves a manufacturer unable to justify a multiyear, multimillion-dollar U.S. plant.

4. DOGE’s headline waste is real, but payroll understates government labor

Jason noted that direct government employment had risen by about 1.3 million across the prior two administrations. Friedberg and Lonsdale pushed back that payroll misses contractors, subcontractors, nonprofits, and state workers supported by federal checks—roughly four million contractor jobs around Washington and another 1.6 million at states, by Lonsdale’s figures.
Lonsdale’s criticism was aimed at Congress as much as DOGE: reconciliation proposals contemplated only $1 trillion-$2 trillion of cuts over ten years, while he wanted at least $5 trillion. Even if Elon Musk found $500 billion or $1 trillion, legislators still lacked “the balls” to remove comparable spending from their own bill.
The sharpest procurement specimen was DOGE’s report of 35,855 ServiceNow licenses across three products with 84 users, plus 11,000 Acrobat licenses with zero users. Jason demanded refunds and an investigation into possible fraud or kickbacks; others cautioned that vendor misconduct was unproven and simple buyer incompetence might explain it.

5. Cutting the IRA wholesale could cost America the AI power race

Chamath said a FERC report attributed almost 91% of December’s incremental electricity generation to renewables. That matters because AI clusters are moving from 100,000 GPUs toward one million, while 35,000 generation applications sit in the FERC queue and existing forecasts do not capture the new demand.
The physical lead times are unforgiving: order a gas turbine now and, in Chamath’s telling, it cannot operate before 2030; new nuclear cannot arrive before 2035. Renewable projects are therefore the only near-term incremental source, supported partly by an approximately $200 billion tax-equity market reinforced by the IRA.
His objection was not that the IRA lacked waste, but that repeal requires surgery. “You don’t have enough electrons” if Congress removes the incentives behind roughly 90% of incremental energy-generation incentives, so a budget cut intended to discipline spending might surrender the AI race through an avoidable power shortage.

6. Friedberg’s battery grant illustrates why the chainsaw must become surgical

Friedberg contrasted his battery-materials company with a grant profiled in a Barry Weiss investigation. Beginning in 2019, his team invested tens of millions, secured an OEM deal, spent millions on an application, was rejected, survived, reapplied, and finally received a $100 million Energy Department grant after roughly two years.
The reported comparison enraged Friedberg: an organization connected to Democratic infrastructure allegedly formed and received $7 billion within 30 days—“70 times bigger than us.” The episode did not adjudicate that claim, but Friedberg’s point was institutional: functioning companies endured technical validation while politically connected applicants appeared to move at extraordinary speed.
Jason’s sequencing metaphor became “start with the chainsaw,” then get surgical. Cancel unused software immediately; preserve the narrow tax and permitting mechanisms that produce power, materials, or other measurable outputs after examining the details.

7. Acquisition reform is the answer to both old and new kleptocracy

Asked whether replacing legacy contractors with Silicon Valley companies merely creates a new kleptocracy, Lonsdale argued for contests rather than favors. He said Epirus defeated entrenched defense vendors by stopping hardened drones 9.5 times farther away, at much lower cost, yet faced specifications apparently written around incumbents’ older cathode-ray-tube approach.
His historical contrast was deliberately stark: a one-page requirement and six competing manufacturers produced the Colt 1911 within about three months; a modern pistol effort generated a 700-plus-page document over years and still failed to deliver. Procurement should specify outcomes, not incumbent inputs.
Friedberg’s safeguard was open standards: publish every RFP, specification, evaluation criterion, and test result, then provide escalation when the data contradict the award. Chamath’s related recommendation was to make the process open source and use open standards. That does not eliminate perceived conflicts for government advisers, but it makes preference visible and shifts the incentive for service toward actual patriotism.

8. Campaign-finance limits split the panel on first principles

Lonsdale defended super PAC spending as one form of speech among many sources of influence—celebrities, unions, doctors, health systems, and media already shape politics. Removing wealthy donors’ channel could simply increase the relative power of those other institutions.
Friedberg’s animal-welfare example sharpened the boundary problem: he wants the freedom to fund websites, books, community engagement, advertisements, and other efforts supporting humane treatment. A book about communism distributed near an election may influence votes without naming a candidate, making “political activity” difficult to define without censorship.
Jason favored hard caps, potentially $5 million-$25 million, plus public support for the final candidates. Chamath favored returning to the prior system by getting rid of Citizens United, arguing that money’s downstream effect on state and local elections, redistricting, and gerrymandering had made communities “sclerotic.”
A Grok query cited on-air estimated that 56% of 2024-25 political spending went to advertising and 44% elsewhere. That moved the discussion toward narrower caps on explicit candidate advertising and canvassing, but no consensus emerged.

9. CoreWeave’s technical edge turned rented GPUs into a giant business

CoreWeave entered its IPO process with 32 data centers, 250,000 Nvidia GPUs, $1.9 billion of 2024 revenue, and a November 2024 secondary valuation of $23 billion. Analysts expected at least $3.5 billion of new capital at more than $30 billion.
Chamath credited a simple architectural choice: CoreWeave avoided conventional hypervisors and let customers write closer to bare metal. That native approach improved performance and proved a startup could still maneuver around AWS, Azure, and Google Cloud through one consequential technical decision.
Friedberg highlighted CoreWeave’s SUNK orchestration framework for scheduling workflows and batches, along with its acquisition of Weights & Biases, which is widely used. Lonsdale instead traced the edge to the founders’ commodity-trading background: they initially acquired hardware for crypto mining, then locked down scarce chips, power, and data-center capacity as a coordinated supply trade.

10. CoreWeave’s IPO is ultimately a duration and concentration trade

The balance sheet is the counterweight: almost $8 billion of debt, nearly $1 billion of 2024 losses, and substantial interest expense incurred to buy GPUs. Chamath’s decisive variable was useful life—if management modeled ten years and economic obsolescence arrives after five, “this business is deeply underwater.”
Customer concentration compounds that risk, with roughly 60% of revenue attributed to Microsoft. The panel could not determine whether that demand represented durable Azure usage, temporary capacity for OpenAI, or another arrangement, leaving the largest revenue line difficult to forecast.
Friedberg compared CoreWeave with 2003 “speed doublers”: highly profitable caching services that accelerated dial-up websites until broadband erased the need. With four or five hyperscalers each spending roughly $80 billion annually, neocloud scarcity may be a transitory arbitrage—though mapping future capacity, power, GPU economics, and demand requires an economist rather than a simple software multiple.

11. A Main Street coalition could make lower equities politically tolerable

Chamath described MAGA’s potential durable coalition as working- and middle-class voters with limited stock or home ownership, pro-innovation technologists, and patriotic business owners. Scott Bessent’s message—“Wall Street’s done great”—and Trump’s statement that he was not watching stocks suggested policy was being reframed around Main Street.
Jason’s proposed market mechanism starts by accepting weaker equities. Lower asset values reduce margin borrowing, second-home purchases, car spending, and other wealth-effect consumption; that could depress inflation without requiring the same degree of labor-market pain.
Jason then argued that volatility drives a flight to quality: investors sell stocks and buy the 10-year Treasury, pushing its yield down. With approximately $1 trillion to refinance over 9-12 months, Chamath said borrowing near 3%-4% rather than 4.5%-5.5% could save the government trillions over time.
Europe supplies the mirror image. Curtailing U.S. aid and intelligence for Ukraine pushed European governments toward debt-financed defense plans; their bond yields rose as markets demanded a higher price for continuing the conflict, increasing fiscal pressure while U.S. yields could benefit.

12. Refinancing helps Main Street, but it does not repair the fiscal deficit

Lonsdale agreed that lower rates would revive real-estate agents, title companies, brokers, and transactions frozen by expensive mortgages. He resisted calling that merely “trickle down”: housing turnover directly activates a wide Main Street employment network.
His preferred disinflationary alternatives were higher productivity through AI and genuine spending cuts. Both are less destructive than manufacturing 8%-10% unemployment, though government can mainly avoid obstructing AI rather than command its productivity gains.
Friedberg assigned a 60/40 probability that Trump 2.0 would tolerate more short-term market pain and listen to Bessent. His caution separated financing from fiscal policy: cheaper refinancing improves the liability structure, but America must still address how it spends money and its fiscal position.

13. Peace in Ukraine does not settle what NATO is for

Jason put the financial ledger at roughly $175 billion of U.S. support and floated recovering $500 billion. He said that because the support was structured on a lend-lease basis, Trump had negotiating leverage. Jason calculated that receiving $500 billion would represent a 42% three-year IRR, while acknowledging that sovereignty and lives cannot be reduced to dollars.
Lonsdale was categorical that “Putin’s a bad guy” who should not have invaded, while arguing the prior administration mishandled deterrence. His desired endpoint was peace through strength: pressure Putin, bring both sides to the table, and require Volodymyr Zelensky to accept that some Ukrainian territory might be surrendered.
On NATO, Lonsdale would not casually abandon the UK, Germany, or long-standing allies. He nevertheless backed fierce conditions, including more European responsibility and pressure over what he sees as deteriorating free-speech norms.
Chamath widened the question to whether NATO, the UN, WHO, and similar transnational institutions have outlived their utility. Europe’s individual nations remain powerful, he argued, but the EU accumulated laws without resolving whether political identity and authority are fundamentally national or European.

14. Technological abundance weakens the resource case for empire

Friedberg contrasted a scarcity worldview, where powers fight for access to finite resources, with a techno-optimistic one in which AI, automation, better mining, abundant energy, housing, food, and water reduce the rationale for global policing and territorial conflict.
His concrete example was a reported Inner Mongolia thorium reserve said to contain 60,000 years of energy at current consumption if used in molten-salt reactors. He also cited an AI-and-sensing startup finding that rare-earth deposits may exist at orders of magnitude beyond current estimates.
Lonsdale accepted the direction but not the timing. Until technology produces something like 5%-6% growth, global trade, security, and nuclear nonproliferation still matter; a post-scarcity world may reduce NATO’s relevance, but policy cannot assume it has already arrived.

15. AI benchmarks are losing signal as model abundance explodes

Chamath called Claude 3.7 “exceptional” for automated code generation and said his consumer use had shifted largely to Grok 3. GPT-4.5 looked merely “good,” while Alibaba’s Qwen was among the best open-source models and DeepSeek remained strong despite receiving less attention.
Grok’s advantage was distribution inside X: click beside a post and receive sources and context without copying text or formulating a prompt. Chamath’s desired next step was an explicit veracity analysis, turning the same interface into a scalable fact-checking layer.
His “dirty little secret” was benchmark overfitting. Model makers optimize against SWE-bench, IMO, AIME, and other known evaluations like students drilling the SAT, so scores increasingly measure test preparation rather than broad capability; the industry needs difficult, changing, independent, verifiable tests.
Distribution therefore becomes decisive. Google can insert AI into search and YouTube, Meta can push a standalone assistant toward a billion users, and X already owns the information stream. Users are “drowning in abundance”—the equivalent of 100 Michelin-star restaurants opening in one city.

16. Cheap intelligence is moving from models into workflows and labor

Jason’s Superhuman example showed the cost curve in practice: AI now drafts every email reply and creates summaries continuously, work that would have been economically prohibitive six to 12 months earlier. Paid software can absorb that inference cost and make the assistant ambient.
Lonsdale described email systems that route work and create team reports automatically, relieving CEOs of being “traffic cops.” His portfolio companies use multiple models and tools such as Cognition’s Devin, with capabilities improving an estimated 10%-15% each month rather than asymptoting as he expected.
Chamath cited journalists and domain experts being hired by data annotators for about $40 an hour to evaluate answers and improve models. He envisioned AI-training, fact-checking, and refinement roles paying $50,000-$150,000 annually, perhaps as a new reinforcement-learning job class.
Friedberg no longer tries to parse each weekly release. He pays $200 monthly for ChatGPT Deep Research, uses multiple models on Google Cloud at Ohalo, and sees the model race like the early web: important, fast-moving, and currently impossible to predict winner by winner.

17. Bitcoin gets Fort Knox treatment while the rest of crypto gets rules

Sacks said Bitcoin merits special treatment because it was the first cryptocurrency, has no issuer, is highly decentralized, carries roughly a $2 trillion market capitalization, has resisted compromise, and is the most widely accepted crypto store of value. The reserve is a “digital Fort Knox” that Treasury is prohibited from selling.
He estimated the federal government once held about 400,000 BTC and sold roughly half for approximately $360 million; those coins would later have exceeded $17 billion. Roughly 200,000 may remain, but the executive order requires a government-wide audit, and assets enter the reserve only after final adjudication and victim-restitution claims are resolved.
Other finally forfeited assets enter a digital-asset stockpile under Treasury’s “responsible stewardship.” Treasury may sell or rebalance that portfolio, while Treasury and Commerce may devise budget-neutral ways to accumulate additional Bitcoin; Sacks rejected Jason’s proposed 0.01% crypto transaction tax, warning that “modest” taxes rarely remain modest.
The appearance issue remained live: Lonsdale said Trump naming specific coins looked bad and surrendered moral high ground. Sacks said the policy was not picking winners beyond Bitcoin and disclosed that Craft sold about $200 million of crypto—roughly $85 million attributable to him—while he exited Bitwise, Multicoin Capital, Blockchain Capital, and other crypto exposure before day one.
Sacks’s regulatory principle was freedom to trade backed by accurate disclosure. FIT21-style market structure would distinguish securities, commodities, collectibles, and other property; issuers should disclose insider ownership, sales, lockups, token creation, and scarcity, with fraud punished “like a ton of bricks.”
Collectibles may have no intrinsic value and still trade if that fact is clear, while functionality claims create heavier obligations. Jason pushed for prominent presentation rules and broader investor education; Sacks placed detailed implementation with Congress, the SEC, and CFTC, citing French Hill, nominee Paul Atkins, and Commissioner Hester Peirce as the relevant actors.

Vol.160 Macro Conversations 77 | A Different Lens on These Nine Things in the Government Work Report (Recorded March 7)Is Google Search Cooked? + We’re Getting a U.S. Crypto Reserve? + What You’re Vibecoding