Big Beautiful Bill, Elon/Trump, Dollar Down Big, Harvard's Money Problems, Figma IPO
Summary
The bill’s biggest strategic miss was losing federal AI preemption just as states accelerate rulemaking. Friedberg estimated that roughly 70 state laws already exist and more than 1,000 bills have been proposed, creating a compliance moat for incumbents while slowing startups. Chamath called AI central to national security and argued that state-by-state model regulation makes as little sense as allowing states to maintain “competitive armies.”
America needs every viable electron, but supply chains and permitting matter more now than generation technology. The U.S. plans to move from 1 TW to 2 TW by 2040 while China goes from 3 TW to 8 TW and adds roughly one America of capacity every 18 months. Chamath’s call: remove subsidies without hindering production, because “the marginal cost of energy has to go to zero,” yet gas turbines ordered today may not operate until 2030 and new nuclear may arrive only in 2032-33.
Elon Musk’s attack on the bill reflects a real fiscal emergency, even if his alliance with Trump probably survives. Musk called America a one-party “Porky Pig Party” after the bill raised the debt ceiling by a record $5 trillion. Friedberg said DOGE actions require congressional appropriations to become permanent, while the White House pointed to later appropriations, possible impoundment and tariff revenue not included by the CBO as potential offsets. Sacks argued that MAGA and tech are codependent, and neither can afford a durable split.
The fiscal bet is now overwhelmingly a growth bet, with Bessent’s “3-3-3” target far from current conditions. The goal is a 3% deficit-to-GDP ratio, 3% GDP growth and 3% inflation; the cited starting points were 6%, 1.4% and 2.4%, respectively. Congress is unlikely to cut deeply enough once constituents depend on spending, leaving AI-driven productivity as the remaining escape hatch—while skeptics say “the train’s left the station.”
Dollar weakness is painful for earners but does not yet break the bid for U.S. assets. The dollar fell roughly 11% in the first half, its worst start in more than 50 years, potentially raising the dollar cost of $4 trillion-$5 trillion in annual imports. Chamath treated depreciation as a manageable “slow bleed” so long as American equities, real estate and hard assets appreciate faster; his true boundary condition is deteriorating human capital and innovation, not debt or currency levels alone.
Harvard’s confrontation with Washington could force a distressed unwind of its private-equity portfolio while AI could erode higher education’s underlying franchise. Sacks said “Harvard’s cooked”: its private-equity allocation nearly doubled from 20% to 40% since 2019, and future forced sales could demand 20%-40% discounts rather than the 7% discount on a recent $1 billion transaction. Friedberg’s broader thesis is that free internet knowledge, AI tutoring and independent research institutions can break both traditional university functions.
Figma offers excellent current economics but exposes the unresolved question hanging over every non-model software company. It reported $228 million of Q1 revenue, 43% operating cash-flow margin, roughly $95 million of free cash flow and about 130% net revenue retention. Chamath thinks such companies may be businesses investors “love to own for a year or two,” yet struggle to underwrite in years three through five if foundational models absorb their workflows—or if specialized departments that buy the software disappear.
Deep dive
1. Congress advanced the bill but abandoned federal AI preemption
The Senate passed the Big Beautiful Bill after a 50-50 deadlock, with J.D. Vance casting the deciding vote and three Republicans voting no. Because the Senate changed the House version, it still required another House vote; Polymarket priced passage by Thursday at 96% and by the July 4 deadline at 97%.
The most consequential deletion for Friedberg was the proposed 10-year moratorium on state AI regulation. Ted Cruz’s attempt to reduce it to five years also failed, leaving an environment in which Friedberg estimated that roughly 70 state laws had already passed and more than 1,000 AI-related bills had reportedly been filed during 2025.
Friedberg defended federalism generally but argued that AI crosses state boundaries through internet services, interstate commerce and national labor markets. Fifty regulatory regimes could make nationwide service “practically impossible,” particularly for smaller developers lacking the legal and financial resources of Google or OpenAI.
California supplied his warning specimen: proposed regulation used model-parameter counts as a review threshold, reflecting an early assumption that AI meant LLMs while ignoring vision-action models, robotics and physical-control systems. Chamath’s sharper framing was that AI sits at “the tip of the spear” of American technological and economic supremacy.
2. State experimentation collides with AI’s national-security role
Chamath rejected Jason’s abortion analogy: abortion concerns an individual woman, her doctor, partner and God, while AI is a national-security capability. Letting states legislate AI, he said, makes as much sense as allowing states to operate “competitive armies.”
He acknowledged legitimate state concerns—Tennessee’s child-safety debate and musicians’ copyright protections among them—but believed they belonged inside a federal framework. Otherwise, fragmented rules would advantage incumbents before the industry’s winners and technical boundaries are even mature.
Jason’s pushback was that AI affects granular physical systems, especially self-driving vehicles using both federal highways and local roads. Friedberg said self-driving touches both states and the federal government, while Sacks maintained that “a city has no idea how to regulate a model.”
3. The energy debate shifted from decarbonization to national capacity
The bill’s energy provisions would remove major Inflation Reduction Act incentives for wind and solar, end the $7,500 EV-purchase credit beginning in September and possibly impose an annual $250 charge on EV owners. Friedberg disliked subsidy dependence but stressed that eliminating incentives could also remove capacity expected within the next several years.
His new benchmark is not merely clean power but total power: America plans to grow from 1 TW to 2 TW by 2040, while China moves from 3 TW to 8 TW. China, he said, is adding electricity capacity equivalent to the entire United States every 18 months.
Energy Secretary Wright’s case, as Friedberg relayed it, is that gas, oil, coal and nuclear can meet demand without subsidizing private-equity returns in solar and wind. Reduced nuclear regulation could let unmet demand create a natural market for reactors, although Friedberg repeatedly hedged: “The jury is absolutely out.”
The legislative process itself drew criticism. The filibuster encourages “massive, massive mega bills” passed with 50 or 51 votes, producing last-minute changes with little rigorous debate or transparency rather than discrete, digestible energy decisions.
4. The binding power constraints are supply, transmission and time
Chamath illustrated the scale through his planned 1 GW data center outside Phoenix: a roughly $25 billion investment cycle requiring an estimated $2 billion-$3 billion of equity. The project is underwritable because it sits downstream from a nuclear reactor, yet it still needs gas and other sources.
A gas turbine ordered today cannot be made viable and turned on until 2030—not because the technology is missing, but because the supply chain is constrained. Even after federal nuclear deregulation, state and local approvals mean new reactors begun now may not operate until 2032 or 2033.
His rueful example was a coal company he nearly bought for roughly $1.2 billion-$1.3 billion before the pandemic. His team objected that it was too dirty; the company later quintupled. “I should have just bought the bloody thing,” he said, explaining why he now favors every form of electricity production.
Solar’s great underwriting advantage is speed: money invested can begin producing revenue in roughly 17 months, versus a decade or more for nuclear. Chamath favored removing all subsidies but no barriers to generation or storage. Jason said PG&E’s grid averages 45%-50% utilization, while Chamath said California has only about five deficit days annually, reinforcing the view that downstream distribution is often the real constraint.
5. Musk’s fiscal rebellion need not end the Trump-tech alliance
Musk attacked the bill’s spending and record $5 trillion debt-ceiling increase, saying America has a one-party “Porky Pig Party” and needs a party that cares about citizens. Trump responded by raising deportation rhetorically and suggesting DOGE could examine Musk’s subsidies.
Friedberg agreed with Musk’s diagnosis: “What the f are we doing? We are in a fiscal emergency.” Yet he also reconstructed the White House case—mandatory-program savings now, a later appropriations bill to formalize DOGE reductions, possible year-end impoundment and tariff revenue the CBO had not included.
Vietnam was the concrete tariff example: approximately $130 billion of annual U.S. imports at a 20% tariff could generate about $26 billion if volumes held. Scott Bessent’s stated destination was a federal deficit below 3% of GDP, though Friedberg emphasized that the outcome remains unproven.
Sacks called Musk the “de facto king of tech” and warned that persistent conflict would make MAGA appear hostile to the sector. Chamath expected the alliance to hold because the two men agree on more than they dispute; Jason suggested Musk institutionalize his influence through pledges covering fiscal efficiency, sustainable energy and U.S. manufacturing, pronatalism and technical excellence.
6. America’s debt problem has become a race between growth and arithmetic
Ray Dalio’s reported takeaway from meetings with both parties was that Washington is unlikely to alter the debt trajectory or avoid painful consequences. Chamath’s prior warning about DOGE was similar: operational efficiencies do not change statutory spending; only Congress can do that.
Bessent’s “3-3-3” target means a deficit equal to 3% of GDP, 3% GDP growth and 3% inflation. The cited current estimates were roughly 6% deficit-to-GDP, 1.4% growth and 2.4% inflation, leaving open whether tax cuts, AI productivity and tariffs can close the gaps without pushing inflation above target.
The discussion also identified burdens beyond federal debt: consumer and corporate debt, state and local obligations, and trillions in unaccounted-for public-pension liabilities. Someone must service them, or the country must “inflate away” the liabilities through money creation.
Friedberg’s political mechanism was more important than the personalities: legislators represent districts that continuously demand more funding, while administrations fulfill campaign promises. Once households and businesses depend on spending, “everyone just votes themselves the dollars”; he therefore sees GDP growth, especially from AI, as the remaining plausible exit.
7. Dollar depreciation hurts consumption while rewarding scarce assets
The dollar fell more than 10% in the first half of 2025—roughly 11% against major currencies—its worst opening half in over 50 years. It had jumped 7% after Trump’s election, so measured from the pre-election dollar-index level near 103, the decline was closer to 6%.
Friedberg connected the move to America’s $4 trillion-$5 trillion of annual imports: consumers and businesses need more dollars to purchase the same foreign output, even before possible tariff price increases. If earnings and assets lag, he argued, demands for government-provided goods intensify—despite “free stuff” requiring still more spending.
Chamath zoomed out: the dollar has devalued roughly 50% over 35-40 years, making depreciation a long-running “carry” rather than a new crisis. If U.S. hard assets rise faster, holders remain ahead; equities, property and other American assets continue to act as a flight to quality.
The boundary condition for that thesis is not the dollar or even the debt, but a collapse in U.S. human capital and innovation. If the Magnificent Seven somehow left America for France’s CAC, the equation would change; absent that, Chamath argued that American ingenuity sustains the asset bid, potentially for another 50-100 years.
8. Harvard’s illiquidity gives Washington negotiating leverage
Harvard had borrowed roughly $1.2 billion amid uncertainty after the administration canceled more than $2 billion of research grants. A $750 million bond offering dated April 9 was followed six days later by supplemental disclosure concerning the White House civil-rights investigation.
The administration demanded an end to DEI initiatives, third-party admissions oversight and mandatory measures against antisemitism. Harvard initially declined a deal but was reportedly negotiating; the estimated consequence of funding cuts and tax changes was an annual budget shortfall approaching $1 billion.
Sacks’s financial call was categorical: “Harvard’s cooked.” Its private-equity allocation climbed from about 20% in 2019 to 40%, concentrating the endowment in the least liquid asset class near the market peak. A recent $1 billion secondary sale cleared at a 7% discount, but buyers facing a forced seller could rationally demand 20%-40%.
Jason noted that proposed foundation excise taxes had reached as high as 8% in some bill language, while explicitly acknowledging uncertainty about the final provision. Even his hypothetical 4% rate implied roughly $2.5 billion of additional annual cost by his estimate, further strengthening Trump’s leverage.
9. AI challenges both functions and the brand moat of universities
Friedberg reduced prestigious universities to two core functions: educating students and enabling researchers. The internet already democratized course content; AI could deliver the equivalent of a Harvard graduate-school education to students in Africa or South Asia through personalized tutoring at “a cost of zero,” while independent institutes compete directly for research grants.
Chamath’s pushback was brand and employer behavior. Goldman Sachs and other firms recruit from Harvard because its logo filters applicants; with “500 million kids graduating a year globally,” replacing that signal creates an enormous differentiation problem.
Jason proposed testing work directly through projects, internships, coding challenges and AI interviews capable of screening 10,000 candidates rather than 50 prefiltered graduates. His own venture firm hires three associates into a training program, with typically two—and sometimes only one—meeting its standard.
Friedberg cautioned that narrow challenges miss late bloomers: he was a marginal university performer whose Waterloo co-op placements gave him time to develop. Chamath said the Thiel Fellowship instead selected people with a mission and demonstrated progress; Jason emphasized self-reliance and warned that taking on $200,000 of degree debt may be unrecoverable.
10. Figma’s IPO tests whether application software can outrun models
Grammarly acquired Superhuman after the email company had raised $114 million, reached an $825 million zero-rate-era valuation and generated roughly $35 million of annual revenue. Combined with Grammarly’s earlier Coda purchase, Jason saw an emerging suite of AI workplace products.
Figma filed with $228 million of Q1 revenue, 13 million monthly active users, $1.5 billion in cash, no debt and CEO Dylan Field controlling 75% of voting power. It sought to raise $1.5 billion amid an IPO reopening that included Circle, which Jason said was up sevenfold from its IPO peak, Chime, eToro, Hinge Health and Wealthfront’s filing. Jason also cited Polymarket at 52% for a September rate cut versus 46% for no change, and thought Powell had indicated tariffs prevented a cut.
Friedberg highlighted Figma’s land-and-expand economics: about 40% revenue growth, 43% operating cash-flow margin, $95 million of Q1 free cash flow and roughly 130% net revenue retention. He preferred a hedged expression—long $50 million-$100 million of Figma while shorting the same amount of Adobe—to “book the spread.”
Meta’s reported $300 million-$500 million recruiting packages, OpenAI’s forecast from $13 billion in 2025 to $125 billion in 2029, and Anthropic’s projected $35 billion in 2027 all signal foundational models absorbing more economic value. Chamath’s deeper risk is organizational: if AI replaces specialized departments or turns workers into more generalized operators, the software buyer may disappear along with the department.