OpenAI's GPT-5 Flop, AI's Unlimited Market, China's Big Advantage, Rise in Socialism, Housing Crisis
Summary
GPT-5 ended OpenAI’s run of decisively superior launches, even if calling it a “flop” overstates a mixed result. Gavin Baker cited Grok 4 at 44.4% on Humanity’s Last Exam versus GPT-5 at 42%, plus a meaningful Grok lead on ARC-AGI-2; GPT-5 was only “very narrowly ahead” on Artificial Analysis. His signal for investors: this was “the first time” OpenAI failed to clear every rival decisively, while Grok 5 and a new Gemini remain pending.
OpenAI’s more defensible GPT-5 advance may be product simplification, not frontier intelligence. Friedberg found that GPT-5 routed a simple retrieval request to a fast response and a complex research task into thinking mode, eliminating the model-picker alphabet soup that ordinary users cannot interpret. The router was reportedly inconsistent for the first 12–16 hours. Jason relayed Benchmark investor Eric Vishria’s view that even if OpenAI “never releases another model,” its users, distribution and product quality could still make it highly valuable.
The AI capex cycle already shows economic returns and full utilization, weakening comparisons with the telecom bubble. Gavin said return on invested capital at the hyperscalers has risen during the buildout, first through “silicon opex to replace human opex” and now through revenue acceleration: Meta’s engagement and ad targeting improved, while Microsoft reported significant Copilot uptake. Dot-com capital funded dark fiber that sat unused; today’s GPUs run so intensively “that they melt,” meaning supply remains behind observable demand.
Power, not model access alone, is becoming the binding asset in the AI race. Phil Deutch cited NVIDIA’s Climate in a Bottle and said the sector remains in the first half of the first inning for solving energy’s hardest problems. The discussion cited Anthropic’s request for 50 GW of US power over three years—roughly 5% of current US electricity production and comparable to an entire year’s recent capacity additions—while data centers need continuous, location-specific power and are relatively price-insensitive. The discussion also cited China adding one terawatt every 18 months, with solar additions at extraordinary scale and more than 150 nuclear reactors in some stage of deployment.
China’s advantage is coordinated manufacturing and infrastructure, but its own suppression of entrepreneurship and weak rule of law may cap it. The existential case presented in the discussion spans AI, military power, TikTok, BYD and state-backed brands; Friedberg instead asked, “Who’s going to win the internet?” and argued that advantage will vary across models, drones, supply chains and markets. China can “throttle up and throttle down entrepreneurship,” but company formation peaked around 2018, DeepSeek personnel are tightly controlled, and Phil’s factory experience was that “every day was a renegotiation on terms.”
The socialism debate converged on housing costs and economic insecurity as the combustible inputs, but not on whether failure will cure the politics. Friedberg warned that rising rent, groceries and government dependence create a “slow-burning” ratchet in which every failed program becomes an argument for more intervention; David Sacks warned that a downturn could produce a backlash “much further to the left than Obama ever was.” Jason’s counter-program was abundant housing, broader ownership and cheaper education: build 10 million homes, loosen regulation, and reduce the anxiety of households spending 60% of their budget on shelter.
Trump’s tariff experiment is producing real revenue but still lacks a stable test of its inflation, demand and growth effects. July receipts reached $30 billion and 2025 receipts about $127 billion, but Ben noted that weak inflation could reflect falling demand—and that maximizing tariff revenue ultimately conflicts with reshoring, which eliminates imports. Phil Libin put success at “50/50.” David Sacks and Chamath Palihapitiya argued that retaliation and early damage had been lighter than expected, while Gavin’s more cautious hedge was that Trump appears responsive to market feedback but needs to restore policy stability.
Apple’s $700 billion of buybacks crystallized the episode’s capital-allocation argument: the company could fund returns and still compete far more aggressively in AI. Gavin rejected the false choice between repurchases and R&D but suggested redirecting even $200 billion toward data centers, calling Apple’s AI products “terrible” and its inaction astonishing. Gavin also offered the upside case for augmented-reality glasses, while Friedberg suggested Apple could eventually become the operating system for homes, EV batteries and distributed power.
Deep dive
1. GPT-5 broke OpenAI’s benchmark-winning streak
Jason framed the release as underwhelming relative to GPT-3.5 and GPT-4: the presentation contained flat or incorrect charts, calculation errors and typos, while Sam Altman’s Death Star teaser had encouraged expectations of something “otherworldly.” Alongside GPT-5, OpenAI released two open-weight models—explicitly not open source.
Gavin’s central observation was historical: “This is the first time that OpenAI has released a new model that was not decisively the best.” Previous challengers from Google or xAI had been overtaken quickly; this time, a frontier competitor released roughly one month earlier remained ahead on important tests.
On Humanity’s Last Exam, a deliberately punishing multidisciplinary benchmark, Gavin put Grok 4 at 44.4% and GPT-5 at 42%. Grok 4 was also “significantly better” on ARC-AGI-2, which targets the jagged frontier of tasks that humans find comparatively easy but AI systems find difficult; GPT-5 led Artificial Analysis only very narrowly.
Gavin hedged his xAI optimism carefully: his firm owns xAI shares, and “to be human is to be biased.” Still, xAI has the largest Blackwell cluster, Grok 5 is coming, OpenAI’s talent losses “may be beginning to weigh,” and the next Gemini will provide another clean test of whether OpenAI has surrendered the frontier lead.
2. GPT-5’s router may matter more than its raw intelligence
Friedberg’s practical test exposed the product advance: GPT-5 answered a simple information-retrieval question immediately, then independently entered thinking mode for a difficult research request. The system selected the underlying route without making the user decipher model names, letters and version numbers.
His UX thesis was categorical: “No one understands what those letters and numbers mean.” Mass-market AI needs a clean interface in which the system determines the appropriate model and reasoning depth, which could make unified GPT-5 more consequential than a small benchmark delta.
Gavin added an important launch caveat: the router was reportedly broken or inconsistent for many users during the first 12–16 hours. Benchmark scores and user impressions might therefore improve once the intended routing system operates reliably.
OpenAI’s moat may outlive its technical lead. Jason relayed Benchmark investor Eric Vishria’s argument that even if OpenAI “never releases another model,” its users, distribution and product execution could still support a highly valuable company.
3. AI is entering workflows before it reaches its economic ceiling
Phil said his investment team uses language models constantly for research efficiency but called the industry “the first half of the first inning.” His best specimen was NVIDIA’s “climate in a bottle”: a model of Earth that hints at AI eventually addressing energy trade-offs and other thorny physical problems.
At Daily Wire, AI already accelerates research and editing. Ben’s concrete example was a Jeremy’s Razors commercial made entirely with AI two months earlier: it retained “the kind of AI slop feel,” but it accomplished the job at far lower cost than a conventional shoot.
Ben’s investor question was whether productivity will arrive fast enough to justify hundreds of billions—and potentially $1 trillion–$2 trillion over five years—of data-center spending. He recalled that the internet functioned by 1999–2000, yet much of its economy-wide productivity took until roughly 2006–2007 to appear.
4. Utilization separates today’s AI boom from dark fiber
Gavin said the returns can be measured because the largest builders report quarterly. Return on invested capital has risen since AI spending began, initially through operating savings—“silicon opex to replace human opex”—and increasingly through revenue growth.
Meta’s recent upside was his strongest revenue example: AI keeps users engaged longer and improves ad targeting, producing better return on ad spend and encouraging advertisers to spend more. Even Microsoft, whose Copilot Gavin does not consider the best product, disclosed “pretty astonishing” adoption statistics.
The dot-com bubble was substantially a telecom bubble built on unused capacity. Tens of billions funded dark fiber in anticipation of future demand; AI infrastructure is the reverse, because GPUs are operated so intensively “that they melt.”
Gavin’s conclusion was that builders are not placing capacity far ahead of demand: “We’re clearly still behind demand.” Economic returns this early in the cycle do not eliminate investment risk, but they are materially better evidence than investors had at the comparable stage of the internet buildout.
5. AI demand has changed the power market before policy could
David Sacks argued that AI’s energy consequences could exceed those of the Inflation Reduction Act and last longer. Hyperscalers have supplied immediate, concrete demand for nuclear and other generation, helping revive small modular reactors without requiring investors to depend entirely on either political party.
Ben Shapiro cited Anthropic’s stated need for 50 GW of US power over three years as the scale marker. He described that as about 5% of US electricity production and comparable to all capacity recently added in a year, after decades in which national electricity demand was effectively flat and is now growing 2%–3%.
Data centers require continuous power in particular locations and are relatively price-insensitive. Hyperscalers also want carbon-free electricity and will pay for it—not merely because government mandates it, Ben argued, but because their organizations and workforces demand it.
Friedberg contrasted this with China adding one terawatt—1,000 GW—every 18 months. The discussion also noted that China imports much of its coal and gas, while its solar buildout is extraordinary, hydro and nuclear reduce fuel dependence, and more than 150 reactors are reportedly in some stage of deployment.
6. Solar’s success reopened the fight over nuclear crowd-out
Friedberg’s chart showed “hyperbolic growth” in solar after costs fell roughly 80%–90%. His reading of the One Big Beautiful Bill was that withdrawing incentives removes generation precisely when the US needs “all we can eat on energy” for AI and national competitiveness.
Friedberg’s counterargument was capital allocation: subsidized solar and wind raise private returns, diverting money from nuclear R&D and helping leave the US decades behind China’s Gen 4 systems—cleaner, meltdown-resistant reactors that China is already deploying at scale.
Jason called the idea that renewable subsidies caused nuclear’s problems “nutty.” Nuclear’s binding constraint was bureaucracy, not another five or ten cents per kilowatt-hour; a subsidy cannot solve a broken regulatory pathway, and centralized nuclear and distributed renewables need not be treated as interchangeable children fighting for one allowance.
Friedberg’s investor evidence was that capital entered SMRs when hyperscalers committed to buying their output, then accelerated when the administration issued four or five orders aimed at removing regulatory barriers. He remained unconvinced that solar and wind subsidies were responsible: government-directed returns can create bubbles and suppress investment in the path with the best long-term economics.
7. America and China disagree on whether AI can have many winners
Ben placed himself firmly in the “existential race” camp. TikTok demonstrated China’s ability to gather data and distribute propaganda at scale; an AI lead could extend that influence into business tools, military capacity, freedom of the seas, allied security and America’s economic position.
Friedberg rejected a single-winner frame: asking who wins AI resembles asking in 1996, “Who’s going to win the internet?” Models will keep evolving while advantage fragments across drone warfare, autonomous devices, manufacturing productivity, materials, supply chains and conventional businesses.
His warning was that superior models alone do not manufacture fleets. China already possesses production and materials advantages in autonomous warfare, while the US may dominate other vectors that AI enables, accelerates or magnifies.
Gavin said China itself clearly treats the contest as existential. On his information, DeepSeek employees cannot share a room with an American while in China and often have a CCP minder—an “astonishing” degree of control that reveals how strategically Beijing treats frontier-model knowledge.
8. China can mobilize capital faster than it can protect entrepreneurship
Jason’s Luckin Coffee, BYD and TikTok examples suggested China could use state-backed capitalism and globally attractive brands against the US. Ben called the system “state-sponsored corporatism”: concentrate extraordinary resources in one chosen industry, accept losses, and undercut market competitors.
Ben favors heavy tariffs on China, but only after building free-trade networks with everyone else and excluding China. Without that preliminary boxing-in, Beijing can “climb out the window” through alternative chip routes, TikTok structures and trade relationships.
Friedberg’s chart showed Chinese company formation peaking in 2018 alongside global venture investment, then falling after policy tightened and Jack Ma disappeared from public business life. The CCP can throttle entrepreneurship up, but wealth and independent power then accumulate; when it throttles down, the economy loses the bottom-up force that lifted living standards.
Gavin’s Soviet analogy made the allocation argument stark: near the end of the USSR, entrepreneurial production on roughly 2% of agricultural acreage reportedly generated a majority of the food. Phil supplied the institutional constraint from his Chinese wind-blade factory: without control of the company chop, ownership meant little, and “every day was a renegotiation on terms.”
9. Competing with corporatism tempts America to copy it
Ben said corporatist systems look spectacular early because they can mobilize everything toward one target, while a messy capitalist democracy spreads resources across many targets. If superintelligence is genuinely existential, AI might require something closer to a Manhattan Project—but that would not justify extending industrial planning across steel, copper and the whole economy.
On federal AI rules, Ben characterized preemption as a defense against state authoritarianism, not simply Washington seizing state power. The intended target is legislation from states such as California that could fill the field with restrictions and destroy competition.
Jason argued—and Ben Shapiro agreed—that an elected, accountable president may be preferable to an imperial judiciary when Congress has failed to legislate. Ben traced the deeper problem to the administrative state: Congress passes vague packages, delegates decisions and political heat to regulators, then returns home to claim the spending.
Ben Shapiro defended career civil servants doing essential safety, defense and intelligence work for limited recognition and pay. Jason placed the blame on Congress for abdicating authority, while Ben proposed that a third party prove itself through governorships and state legislatures before attempting to capture Washington.
10. Socialism advances through ratchets, not a single election
Friedberg predicted socialism would “sweep over this nation,” driven by stagnant wages, visibly higher groceries and rent, and a government that directly or indirectly employs, by his calculation, nearly half of Americans. Chamath countered that US capitalism, freedom and rule of law remain resilient and that political arithmetic eventually swings back.
Chamath viewed Zohran Mamdani as the near-term test: he disagreed with every policy but called him “an incredibly charismatic, effective politician.” If Mamdani wins and implements the agenda, Chamath expects failures in policing and government-run retail to discredit socialism quickly.
David Sacks’ pushback was that destruction takes time and can always be blamed on wealthy financiers who fled. More dangerously, a market crash, unemployment or renewed inflation could turn “the rich got richer and we got poorer” into a movement much further left than Obama.
Friedberg described the policy ratchet: if food stamps prove insufficient, expand them; if rent control fails, have government build housing. “The solution is never, let’s reduce the government’s role”—failure becomes evidence that intervention was too small, increasing taxes and regulation until the stated endpoint is seizing the means of production.
11. Housing costs are the most actionable source of political anger
Jason argued that households spending 60% of their “nut” on housing experience radically different anxiety from households that can afford ownership. His prescription combined Australian-style super funds, individual market ownership, housing deregulation, 10 million additional homes and even two new cities.
Education is already adjusting: young people are considering schools such as UT where an entire degree might cost $50,000, or entering trades where plumbers and electricians can earn $150,000 a year. Housing, however, remains the expense neither party has made its defining national project.
The discussion widened the diagnosis to falling geographic mobility. The traditional American dream was not necessarily to die where one’s grandparents lived; it was to cross an ocean or mountain, build a “ramshackle hut,” and move again if opportunity failed. Aggregate job openings cannot help workers unwilling or unable to relocate or change industries.
12. Tariff revenue is real, but its source may be shrinking demand
July tariff receipts reached about $30 billion and the 2025 total roughly $127 billion; $30 billion monthly annualizes to $360 billion. Jason and David Sacks discussed $250 billion as a plausible 2026 level, given that the broad regime began around midyear.
The June trade deficit fell to roughly $60 billion, its lowest since June 2023, without the immediate inflation surge many expected. Ben Horowitz’s caution was that muted prices might mean demand has already dropped, making current tariff receipts partly a one-time spike before import volumes and revenue decline.
Ben identified an internal contradiction: the administration cannot permanently maximize tariff revenue while celebrating reshoring. If imports are replaced by domestic production, the tariff base disappears; revenue and industrial substitution can coexist during transition, but “those two things are eventually mutually exclusive.”
Friedberg used Chinese solar panels to state the unavoidable trade-off: tariffs may level the playing field or protect supply chains, but they also increase US energy costs. Globalization was a net positive during his lifetime, and he could not confidently judge whether security benefits now outweigh its retreat.
13. Trump’s flexibility limits the downside but prolongs uncertainty
Ben Horowitz described Trump as a zero-sum thinker who treats trade deficits as evidence of losing, explaining why “reciprocal” rates were based on bilateral deficits rather than foreign tariff schedules. Switzerland’s 39% rate illustrated the ad hoc approach: individual battles and headlines, not a stable systems design.
Phil Libin separated three policies: low taxes and deregulation are constructive, while the tariff war significantly dampens growth. The counterfactual matters—markets surviving the disruption is not proof of success if tax cuts and deregulation could have produced a much stronger economy without it.
India’s proposed combined 50% tariff captured the mixed objectives: 25% for alleged unfair trade practices and 25% for buying Russian oil. Japanese and EU investment agreements were still described as handshakes rather than fully papered commitments, while litigation could return tariff authority to Congress.
Gavin declined false precision but emphasized that Trump “looked into the abyss” in April and changed course; the critical question is whether he remains responsive early enough to brake before the cliff.
14. The tariff experiment is milder than feared, not yet proven
David Sacks’ constructive case was that the early damage and foreign retaliation have both been smaller than expected. That supports the administration’s claim that access to the American consumer gives the US leverage, especially where other countries have long imposed higher barriers.
He still preserved the economic objection: reciprocity sounds fair, but comparative advantage can make asymmetric tariffs desirable. The value of the current policy may simply be empirical—after tariffs animated both parties for 30 or 40 years, the country will finally know more in 12–18 months.
Sacks said he likes “running experiments” and would use tariff receipts to reduce direct taxation, but only alongside a smaller government. His largest fiscal concern is that a new federal revenue stream becomes “another crutch to keep spending up,” converting a trade tool into a permanent expansion of the state.
15. Chip leakage is manageable; Apple’s capital allocation is the bigger indictment
US authorities alleged that ALX Solutions routed advanced GPUs through Singapore and Malaysia while receiving payments from Hong Kong and China; the felony charge carries up to 20 years. The Financial Times estimate discussed was at least $1 billion of NVIDIA chips entering China over three months at roughly a 50% black-market premium.
Gavin expects smuggling to persist because chips are too valuable to exclude completely, just as the US cannot eliminate illegal drugs. Yet $1 billion is not comparable with the $10 billion–$20 billion Blackwell clusters being built by US hyperscalers or with Colossus; selling China restricted H20s and a “despecced Blackwell” could also reduce demand for illicit frontier parts.
Apple, by contrast, spent $700 billion on buybacks over a decade—more than the market capitalization of all but 12 S&P 500 companies—while Jason counted AirPods and perhaps M4 chips as its leading innovations. Gavin said buybacks are valuable but asked why even $200 billion was not redirected into AI data centers: “It almost feels like they’re trying to lose.”
Gavin refused to count Apple out: critics ridiculed the iPad before sales rose from 10 million to 100 million, and he would be surprised if Apple did not deliver the best AR glasses within three or four years. Friedberg extended the upside to energy: 30 million EVs by 2030, each with 80–200 kWh, could represent 2,400–3,300 gigawatts of moving storage whose charging, discharge and aggregation need an operating system.