Featuring
Travis Kalanick joins for CloudKitchens, China's copy-to-innovation machine, and an autonomous-future tour; David Sacks drops in from Washington for DeepSeek. The sharpest clash asks whether open models made frontier labs a terrible business overnight. Later, DOGE's first ten days become a harder question: what can a president cut without Congress? Sacks has the best AI-economics read, and Friedberg wins the budget-law round. The revolution, it turns out, still has a cap table and an appropriations clause.
Spice rack
Would open-source models erase the economics of frontier-model companies?
Original point: Large language models are rapidly depreciating assets that will be open-sourced and commoditized, so most durable value will move to applications and robotics.
What everyone argued
Chamath Palihapitiya
Rapid model turnover makes any single model a fragile dependency. Builders should create a model-agnostic shim so they can replace Claude, OpenAI, Llama, or a later model without rebuilding the product.
Jason Calacanis
The model layer would become like storage or GPS: useful infrastructure, but the best equity value would sit in applications and physical products. He made the strongest version of the commodity claim: the models would be open-sourced and 'not worth anything.'
David Sacks
Cheap AI could expand total usage, and that expanded market could keep rewarding frontier innovation. R1 was a fast follower comparable to an older OpenAI reasoning model, so it was premature to infer that frontier labs would be commoditized or earn no return on capital.
David Friedberg
If model performance improves while costs fall, value may migrate to wrappers, users, or the broader economy, much as cheap electricity created more value outside electric utilities than inside them.
Winner circle
David Sacks wins. DeepSeek proved that frontier leads can narrow quickly and cheaply, but it did not prove that frontier-model companies had become worthless utilities. Later funding, adoption, and continued product differentiation support Sacks's narrower claim that the obituary was premature, while leaving the commodity pressure very much alive.
Commentary
Chamath Palihapitiya
Assumptions and fact checks
Frequent changes in the best model will make model portability a durable source of application value.
Why it mattersThe premise is technically sound: routing and portability reduce supplier lock-in. It is a product advantage, though not proof that the underlying suppliers lack pricing power.
DeepSeek introduced meaningful technical innovations, including GRPO-based reinforcement learning and lower-level GPU optimization.
CheckDeepSeek's R1 paper and later Nature publication document GRPO-based training and efficiency-oriented engineering. Those innovations are real even though they do not validate every public claim about total development cost.
Jason Calacanis
Jason's analogy makes the debate legible, but 'not worth anything' carries a much heavier burden than 'depreciates quickly.' He does not show why Jevons-style demand growth or frontier differentiation could not sustain a concentrated supplier market.
Assumptions and fact checks
Because model performance diffuses quickly, frontier models will have no meaningful standalone economic value.
Why it mattersRapid diffusion weakens moats, but later funding, usage, and revenue evidence shows that customers and investors continued to assign substantial value to frontier capability, distribution, reliability, and product integration.
Most long-run value will accrue above the base-model layer.
Why it mattersApplications can capture large value without eliminating value below them. The allocation remains unsettled and differs by consumer, enterprise, and infrastructure market.
David Sacks
Sacks wins by resisting a categorical conclusion while granting the real open-source pressure. He could have been more precise about which frontier moats—research cadence, distribution, enterprise trust, data, or compute—might persist.
Assumptions and fact checks
Lower AI prices will expand consumption enough to preserve incentives for frontier investment.
Why it mattersLater adoption and fundraising support the direction of the claim, though private valuations do not by themselves prove eventual profits or attractive returns on invested capital.
DeepSeek R1 reported performance comparable to OpenAI o1-1217 on major reasoning benchmarks.
CheckDeepSeek's paper explicitly reports results on par with OpenAI o1-1217 across several reasoning tasks. Benchmark comparability is narrower than overall product equivalence, but the claim as framed is supported.
Frontier-model companies still had access to very large pools of capital after R1.
CheckOpenAI announced a $40 billion round at a $300 billion post-money valuation in March 2025. Anthropic announced a $3.5 billion Series E at $61.5 billion and later a $13 billion Series F at $183 billion.
David Friedberg
Friedberg frames value migration better than the absolute claim that models become worthless. His case would be stronger with an explicit distinction between social surplus and investor returns at each layer.
Assumptions and fact checks
Falling model costs will shift a larger share of surplus to applications and end users.
Why it mattersLower input costs normally broaden feasible applications and consumer surplus. The remaining uncertainty is the share retained by frontier providers with differentiated capability and distribution.
Could DOGE deliver trillion-dollar spending cuts mainly through executive action?
Original point: The central test would be how far courts allowed the executive branch to stop spending that Congress had required by statute.
What everyone argued
Chamath Palihapitiya
DOGE's early pace showed little visible service impact, and its engineer-led review of government payment systems could uncover more than $2 trillion in waste after workforce and real-estate savings.
Jason Calacanis
Return-to-office rules, deferred resignations, lease exits, competency reviews, and payment scrutiny could deliver fast savings while popular transparency campaigns built political support.
David Friedberg
Fast administrative cuts could help, but courts would decide whether agencies could stop congressionally mandated spending. Social Security, Medicare, and other mandatory programs were too large for executive friction alone to close a roughly trillion-dollar fiscal gap.
Travis Kalanick
Even where Congress had directed spending, the executive could slow hiring, procurement, and payment through administrative friction. He also argued that holding spending to its 2019 nominal level against 2024 revenue would have produced roughly a $500 billion surplus.
Winner circle
David Friedberg wins. DOGE demonstrated that executive management can materially shrink headcount and change agency operations, but it did not dissolve Congress's power of the purse or the statutory structure of mandatory spending. The later deficit and GAO decisions show why large administrative savings and trillion-dollar structural cuts are different propositions.
Commentary
Chamath Palihapitiya
Chamath spots a real modernization opportunity, but he treats the payment ledger as if every surprising line item were discretionary waste. The missing step is a legally and operationally defensible bridge from identified transactions to cash outlays that can actually be canceled.
Assumptions and fact checks
Forensic access to federal payment systems would reveal more than $2 trillion of waste that could be removed without material service loss.
Why it mattersPayment visibility can expose errors and duplication, but $2 trillion approaches the scale of the annual deficit and exceeds the plausible administrative-waste base documented by nonpartisan watchdog work. Much federal spending is mandatory or reflects explicit policy choices rather than operational leakage.
Little immediate disruption in the first ten days was evidence that very large cuts would be painless.
Why it mattersImplementation lags make ten days a poor test of service impact, legal durability, or realized cash savings.
Jason Calacanis
Jason is strongest on specific management tools and weakest when he annualizes gross claims before they become Treasury outlay reductions. A credible scorecard needs realized net savings, service metrics, and a baseline that excludes spending merely shifted in time.
Assumptions and fact checks
Headcount and office reductions would translate nearly one-for-one into large, durable deficit reduction.
Why it mattersThey can reduce administrative costs, but federal compensation and real estate are much smaller than mandatory benefits and interest; one-time separation costs and later hiring also matter.
The 2025 deferred resignation offer kept eligible employees on pay and benefits through September 30, roughly eight months after the January offer.
CheckOPM's official offer stated that participating employees would retain compensation and benefits through September 30, 2025, subject to the program terms and exclusions.
The workforce initiatives could induce attrition on the order of 5 to 10 percent.
CheckOPM's later dashboard reported 139,917 deferred-resignation participants and a net federal workforce reduction of 272,283 by May 2026. The exact percentage depends on workforce scope and timing, but the forecasted order of magnitude was plausible.
David Friedberg
Friedberg does the best job matching the mechanism to the size of the target. He also acknowledges that executive actions can matter; his advantage is refusing to confuse meaningful administrative savings with authority over the whole budget.
Assumptions and fact checks
Major durable deficit reduction would eventually require Congress to change appropriations, mandatory-program law, revenues, or some combination.
Why it mattersThe scale and statutory composition of federal outlays make legislation unavoidable for trillion-dollar structural changes.
The executive branch cannot simply cancel spending that federal law requires.
CheckGAO explains that Congress holds the power of the purse and that withholding enacted funding must follow the Impoundment Control Act. GAO later concluded that FEMA violated the Act when it withheld certain FY2025 funds required by law.
Mandatory programs, especially Social Security and Medicare, make up most federal mandatory spending and cannot be reduced through ordinary administrative discretion.
CheckCBO reported $4.2 trillion of mandatory outlays in FY2025, with more than half going to Social Security and Medicare. Those programs are governed primarily by statutory eligibility and benefit formulas.
Travis Kalanick
Travis is right that implementation details create real executive leverage, but he slides from 'can slow some spending' to a scale claim the mechanism cannot carry. His nominal-budget comparison needs a category-by-category feasibility test.
Assumptions and fact checks
Administrative friction can substitute for legislation at trillion-dollar scale.
Why it mattersFriction can slow discretionary operations, but it cannot lawfully rewrite entitlement formulas, permanently rescind most appropriations, or erase interest obligations.
A nominal FY2019 spending ceiling is a meaningful measure of removable 2024 waste.
Why it mattersIt is a provocative counterfactual, not a waste estimate. It omits inflation, demographic change, enacted benefit expansions, defense priorities, and higher debt-service costs.
Using FY2019 nominal outlays with FY2024 receipts would yield a surplus of roughly $500 billion.
CheckCBO reports roughly $4.4 trillion of FY2019 outlays and $4.9 trillion of FY2024 receipts, a difference close to $0.5 trillion. This validates the arithmetic, not the feasibility of freezing five years of nominal spending.
Chamath separates application architecture from model selection well. He would have strengthened the investment claim by distinguishing falling inference prices from the economics of training and distributing the next frontier model.