OpenAI's GPT-5 Flop, AI's Unlimited Market, China's Big Advantage, Rise in Socialism, Housing Crisis
Speakers
GavinSBaker
— Portfolio Manager, Atreides Management
Phil Deutch
— Energy investor, NGP Energy Technology Partners
Jason Calacanis
— Angel Investor / Founder, LAUNCH
David Friedberg
— CEO, The Production Board
Ben Shapiro
— Co-founder, The Daily Wire
Summary
Jason Calacanis and David Friedberg host a summer All-In episode with investors Gavin Baker and Phil Deutch and commentator Ben Shapiro, recorded the day after OpenAI released GPT-5. The panel argues GPT-5 was the first OpenAI model that failed to decisively beat rivals, but that AI capex is already generating measurable returns, unlike the dark-fiber telecom bubble. A long energy segment covers AI's 50-gigawatt power demand, the solar-versus-nuclear subsidy fight and the One Big Beautiful Bill's rollback of clean-energy incentives. The show closes on China's competitive model, tariff revenue, Nvidia chips smuggled into China and Apple's $700 billion of buybacks alongside its weak AI execution.
- GPT-5 landed without a decisive benchmark lead, with Grok 4 ahead on Humanity's Last Exam and ARC-AGI-2; its main advance was automatic model routing.
- Gavin Baker argues AI capex is lifting return on invested capital, citing Meta's revenue beat and Microsoft Copilot uptake, and that GPUs are fully utilized unlike dark fiber.
- Phil Deutch says AI power demand, including Anthropic's 50-gigawatt request, is a bigger and longer-lasting energy driver than the Inflation Reduction Act and has revived small modular reactors.
- Friedberg blames solar and wind subsidies for starving nuclear R&D, leaving the US decades behind China on Gen 4 reactors; Deutch defends the subsidies and criticizes the OBBB rollback.
- China's advantages and limits are debated: manufacturing scale, BYD and Luckin Coffee brands and drone supply chains against entrepreneurship suppression since 2018 and weak rule of law.
- Tariffs have raised about $127 billion in 2025; Ben Shapiro calls the trade war a drag that offsets tax cuts and deregulation, with US indices flat since January.
- Nvidia GPU smuggling into China is judged immaterial versus domestic Blackwell clusters, and resuming H20 sales into China is seen as smart policy.
- Apple is criticized for $700 billion of buybacks and poor AI execution, with a counterview that it may still lead in AR glasses within a few years.
Ideas
AI capex is earning real returns.
AI data-center spending is already earning its cost of capital. Most of the companies making these investments are public and report quarterly, and their return on invested capital has actually gone up since the AI buildout began - first through opex savings as silicon replaced human opex, and more recently through real revenue acceleration. Meta just reported a quarter with revenue significantly ahead, much of it from AI keeping users engaged longer and improving ad targeting so advertisers get better ROAS and spend more, and even Microsoft shared astonishing Copilot uptake stats. The contrast with the telecom bubble matters: that capital went into dark fiber that sat unused, whereas GPUs are used so intensively that they melt. Nobody is building ahead of demand; supply is still behind it.
AI capex is earning real returns.
AI data-center spending is already earning its cost of capital. Most of the companies making these investments are public and report quarterly, and their return on invested capital has actually gone up since the AI buildout began - first through opex savings as silicon replaced human opex, and more recently through real revenue acceleration. Meta just reported a quarter with revenue significantly ahead, much of it from AI keeping users engaged longer and improving ad targeting so advertisers get better ROAS and spend more, and even Microsoft shared astonishing Copilot uptake stats. The contrast with the telecom bubble matters: that capital went into dark fiber that sat unused, whereas GPUs are used so intensively that they melt. Nobody is building ahead of demand; supply is still behind it.
AI power demand revives small modular reactors.
The energy implications of AI are probably bigger than the Inflation Reduction Act and will certainly be longer lasting. Anthropic just asked for 50 gigawatts of US power over three years - about 5% of total US electricity production and roughly as much as the country adds in a year - after decades in which US electricity demand growth was essentially zero and is now 2 to 3%. Data centers need continuous power in specific locations, are relatively price-insensitive, and want carbon-free supply and are willing to pay for it without waiting for government programs. That concrete, immediate private demand from hyperscalers, not policy, is what has given small modular reactors their renaissance.
Chinese brands out-compete US incumbents globally.
The real Chinese threat is that they beat America at capitalism rather than at communism, using consumer brands the way the US used Apple and Google. BYD is producing cars at roughly half the price of American cars and shipping them on its own vessels, Luckin Coffee is expanding into the US with $2 cups aimed squarely at Starbucks and Dunkin, and TikTok already owns US attention. State-sponsored corporatism lets them pour resources into one industry at a time and undercut the rest of the market, so the competition shows up as branded consumer companies taking share, not as ideology.
Entrepreneurship suppression caps China's long-run competitiveness.
The CCP can throttle entrepreneurship up and down, and it throttled it down. Financial Times data on companies founded in China per year peaked in 2018, aligning with the peak of venture investment into China, and then declined rapidly after the policy changes around the Jack Ma episode. What took Chinese GDP per capita from roughly 3,000 to 30,000 was ground-up entrepreneurship and self-organization by ordinary citizens, not central planning, so suppressing that is a rate limit on China's competitiveness even where it retains manufacturing and central-command advantages.
Weak rule of law limits China.
Innovation alone is not enough - durable growth requires rule of law, and China has not shown it can deliver it. From first-hand experience operating factories there, including a wind-blade plant, every single day was a renegotiation of terms no matter what had been agreed, and if you did not control the chop that made the stamp you did not actually control the factory. Absent enforceable property rights, Chinese success still depends on the government propping it up, which is not a recipe for decades of compounding innovation.
Solar keeps scaling despite subsidy rollback.
The incentives worked: subsidies to solar and wind cut costs by roughly 80 to 90% and produced the hyperbolic growth visible in global generation, with storage the remaining piece needed to match gas. The problem is policy whiplash - what is needed is rational subsidies that persist across Republican and Democratic administrations, as happened when Texas wind requirements carried through. Removing the incentives in the One Big Beautiful Bill takes generation supply backwards at exactly the moment the US needs all it can eat on energy for artificial intelligence, so the right answer is as much solar and wind plus as many SMRs as possible rather than pitting them against each other.
Nuclear is the most scalable energy source.
Solar and wind subsidies distorted energy capital allocation and created a financial disincentive to invest in nuclear. Because the government paid half the cost, private capital chased the subsidized return instead of forcing solar and wind costs down natively, and the R&D dollars that should have carried the US to Gen 4 systems never appeared. China has now deployed those clean, meltdown-proof Gen 4 reactors at scale while the US has none and at best one starting construction in 18 months. Nuclear is ultimately the most scalable energy production system available, and the US is decades behind purely because capital was redirected into subsidized markets.
Trade war offsets tax cuts and deregulation.
Trump is making three systemic economic changes: low taxes and deregulation, which give investors a sense of solidity, and the tariff war, which is creating a pretty significant dampening effect on economic growth right now. The evidence is that the major US indices are almost precisely where they were when he took office - essentially flat since January - despite the tax and regulatory tailwinds. Without a gigantic trade war on top of deregulation and tax cuts, the economy would be ripping with more jobs, and what investors report today is confusion and choppy water rather than the stability that lets capital commit.
Chip smuggling immaterial; China sales are smart.
GPU smuggling into China will be impossible to stop - the US cannot keep illegal drugs out while trying hard, and China is doing everything it can to bring chips in - but it is not material. Roughly $1 billion of smuggled GPUs, even Blackwells, is nothing against the $10 to $20 billion Blackwell clusters hyperscalers are standing up across America and is not comparable to Colossus, so smuggling will not be decisive in the AI race. Resuming sales of H20s and a de-specced Blackwell into China was smart for the same reason legalizing a weaker product shrinks a black market: the US supplies the lower-potency product and keeps the frontier chips at home.
Apple's AI inaction wastes its buyback capacity.
Buybacks versus R&D is a false dichotomy - Apple could easily have afforded both - so the problem is not the roughly $700 billion of repurchases but that it did not redirect, say, $200 billion of it into data centers and become a real competitor in AI. Apple's inaction and ineptitude in AI is shocking: the products are terrible from a user perspective, Siri has gotten worse rather than better, and the company is losing talent. Being comfortable arriving late worked with the iPhone and iPad, but in AI Apple is not even in the running.
Apple wins AR glasses within four years.
In fairness to Apple, augmented-reality glasses play directly to its strengths in hardware, silicon and design, and it would be shocking if Apple does not have the best AR glasses on the market within three or four years. That may be exactly what the company is quietly focused on, which would explain the silence around Vision Pro - a VR device whose eventual purpose is AR - in keynotes and developer updates.
Aggregating distributed EV storage will be valuable.
Households and businesses are accumulating backup generators, rooftop solar, power walls, EVs and grid connections, and eventually someone has to aggregate and control all of it, which will require a whole new infrastructure layer in the US. By 2030 an estimated 30 million US electric vehicles carrying 80 to 200 kWh each represent something like 2,400 to 3,300 gigawatts of storage that is moving, and adding autonomy makes that an enormous pool of flexible distributed power. Whoever controls the charge, discharge and management of that fleet - starting with something as mundane as a corporate parking lot - owns a very valuable operating system for the power system.
Up Next
This All-In Podcast video, published August 09, 2025, features GavinSBaker, Phil Deutch, Jason Calacanis, David Friedberg, Ben Shapiro discussing META, MSFT, Small modular reactors, 1211.HK, LKNCY, MCHI, TAN, Wind Power, URA, SPY, NVDA, AAPL, Distributed energy storage. 13 trade ideas extracted by AI with direction and confidence scoring.
Speakers: GavinSBaker, Phil Deutch, Jason Calacanis, David Friedberg, Ben Shapiro · Tickers: META, MSFT, Small modular reactors, 1211.HK, LKNCY, MCHI, TAN, Wind Power, URA, SPY, NVDA, AAPL, Distributed energy storage