China's open-source GLM 5.2 just matched GPT-5.5 on coding benchmarks, trained entirely on indigenous Huawei chips — and the founder says they'll hit Anthropic Fable-level capability before Q1 2027.
All-In with Chamath, Jason, Sacks & Friedberg
China's open-source GLM 5.2 just matched GPT-5.5 on coding benchmarks, trained entirely on indigenous Huawei chips — and the founder says they'll hit Anthropic Fable-level capability before Q1 2027.
TL;DR
Socialists sweep NYC congressional primaries as Mamdani-endorsed DSA candidates unseat established incumbents, sparking debate about the Democratic Party's future [1] — Jason Calacanis "Mamdani-endorsed DSA candidates went 3 for 3 in NYC congressional primaries, unseating major incumbents in a 26% probability upset. The DSA…" 01:05 . Gavin Baker argues AI is the greatest economic equalizer of our lifetime but Silicon Valley has failed to communicate that message [2] — Chamath Palihapitiya "I think AI is the greatest economic leveler thing that we'll ever find in our lifetime. I think it's the thing that can create the greatest…" 03:13 . China's GLM 5.2 becomes the best open-weight model ever, raising urgent questions about US export strategy and model regulation [3] — Gavin Baker "China runs farms of tens of thousands of devices querying Claude and GPT APIs through masked accounts, harvesting every reasoning trace and…" 47:16 . Micron smashes earnings with revenue up 4x to $42B as the AI memory crunch drives Apple price increases of 14–25% [4] — Jason Calacanis "Micron revenue up 4x YoY: $9B to $42B: Micron reported revenue of $42 billion, up 4x year over year from $9 billion, beating expectations b…" 1:01:40 . The episode closes with a deep dive into orbital compute economics: Starship could put a gigawatt of compute in space for $40B versus $60B terrestrially [5] — Gavin Baker "A terrestrial 1-gigawatt data center costs $60 billion ($35B chips + $25B power/cooling). A reusable Starship puts the same compute in spac…" 1:11:50 .
Gavin Baker and Travis Kalanick join the All-In Pod as Mamdani-endorsed socialists sweep NYC congressional primaries. Topics include the DSA's takeover of the Democratic Party, China's GLM 5.2 catching up to US frontier AI models, Micron's blowout earnings amid an AI memory crunch, the math behind orbital compute and space-based data centers, and an IPO update covering Anthropic's $3T valuation, SpaceX's float dynamics, and Cerebras breaking deal price.
Jason welcomes guests Gavin Baker of Atreides Management and Travis Kalanick of Adams, noting Friedberg took a mental health day after the socialist sweep in NYC.
Three DSA candidates backed by Mamdani unseat major incumbents in a 26% probability upset. The DSA co-chair openly admits using the Democratic Party as a ballot access vehicle [1] — David Sacks "We're using the Democratic Party as a ballot access vehicle. Not because we share its goals. We build our own organization, get elected und…" 10:30 .
Z.AI releases GLM 5.2 — the highest-scoring open-weight model ever — trained on Huawei chips, 85% cheaper than GPT-5.5, with founder claiming fable-level capability before Q1 2027 [1] — Jason Calacanis "GLM 5.2: 744B params, 1M token context, MIT license: China's Z.AI released GLM 5.2, a 744 billion parameter open-source model with a 1 mill…" 45:20 .
Debate over whether Dario Amodei engineered Fable's rollback to gain a regulatory moat [1] — Gavin Baker "Anthropic has spent years advocating for an FAA-style approval process for AI. When a jailbreak report from a trusted partner triggered Fab…" 55:37 . David Sacks warns the US is on a shot clock — slowing down won't slow China, which operates outside US jurisdiction.
Micron revenue up 4x to $42B [1] — Jason Calacanis "Micron revenue up 4x YoY: $9B to $42B: Micron reported revenue of $42 billion, up 4x year over year from $9 billion, beating expectations b…" 1:01:40 , entire 2026 supply sold out. DRAM is projected at 30–40% of hyperscaler CapEx. Apple forced to raise Mac prices 14–25% as AI data centers hoover up memory supply [2] — Jason Calacanis "Apple MacBook Neo price up 14–15%: Apple raised the price of the MacBook Neo from $699 to $799, a 14–15% increase, driven by the AI memory …" 1:03:00 .
Tesla's Megapod trademark filing suggests modular AI data centers at Supercharger stations. Travis describes putting compute in Adams restaurant kitchens. Debate over distributed training vs. inference efficiency constraints [1] — Travis Kalanick "Distributed training across physically separate locations is orders of magnitude less efficient — even 2 kilometers of fiber kills performa…" 1:12:14 .
Gavin Baker values Anthropic at $3T [1] — Gavin Baker "Gavin Baker believes Anthropic would trade at $3 trillion as a public company, with revenue on track to hit well over $100 billion by year …" 1:27:48 . SpaceX lockup fear may be overstated as employees bought on IPO. Cerebras breaks deal price triggering systematic sell-off [2] — Gavin Baker "When Cerebras broke IPO deal price, giant fund managers with an ironclad rule to sell regardless of fundamentals triggered a doom loop. Sho…" 1:29:25 — Gavin advises always price IPOs to never break deal price.
Chapter 2 · 01:05
Three DSA candidates backed by Mamdani unseat major incumbents in a 26% probability upset. The DSA co-chair openly admits using the Democratic Party as a ballot access vehicle [1] — David Sacks "We're using the Democratic Party as a ballot access vehicle. Not because we share its goals. We build our own organization, get elected und…" 10:30 .
Mamdani-endorsed DSA candidates went 3 for 3 in NYC congressional primaries, unseating major incumbents in a 26% probability upset. The DSA has openly declared it's using the Democratic Party as a ballot access vehicle while pursuing an agenda to abolish the Senate, ICE, prisons, and seize corporate assets.
Mamdani-endorsed DSA candidates won all three NYC congressional primaries on Tuesday, defeating two multi-term incumbents in a major upset.
According to Polymarket, the chances of a Mamdani sweep across all three congressional primaries was just 26% before Election Day.
AI gives every person on Earth access to a genius co-founder with no gatekeeping — but Silicon Valley has done a catastrophically poor job of communicating this. The messaging vacuum has allowed DSA to fill it with socialist rhetoric, and the pro-AI groups barely held contested congressional races.
When truth and justice erode, society's immune system fails and radicalism fills the void. Travis Kalanick's framework explains why COVID misinformation, unchecked crime, and social media rot lead inevitably to political extremism.
Communism — the desire to be lazy and get something for nothing — is hardwired into human nature. Every person has felt it. The danger is when societies build ecosystems that reward those impulses without consequences. That's what we're watching scale in American cities.
California quadrupled per-capita homelessness spending with worsening outcomes — because the money flows to NGOs run by allies, not to solving problems. This is the Curley Effect: pursuing policies designed to fail while driving out opponents and enriching a political class.
Chapter 3 · 22:50
Z.AI releases GLM 5.2 — the highest-scoring open-weight model ever — trained on Huawei chips, 85% cheaper than GPT-5.5, with founder claiming fable-level capability before Q1 2027 [1] — Jason Calacanis "GLM 5.2: 744B params, 1M token context, MIT license: China's Z.AI released GLM 5.2, a 744 billion parameter open-source model with a 1 mill…" 45:20 .
The real goal of under-16 social media bans isn't protecting children — it's forcing adults to de-anonymize so governments can build censorship regimes. What gets labeled 'harmful content' is just content that disagrees with whoever holds power.
Gavin Baker cited an uncontested study showing that electing a Republican DA causes all-cause mortality for young Black men in that city to drop by 7%.
Chamath cited a statistic that 0.1% of the population commits approximately 70% of violent crimes, suggesting targeted interventions could near-eliminate crime.
Polling shows 80% of Democrats now say they disapprove of Israel, a dramatic reversal from previously high bipartisan approval ratings.
Among Republicans under 50, Israel's disapproval rating is now 57%, reflecting a major generational divide within the GOP.
Chapter 4 · 45:12
Debate over whether Dario Amodei engineered Fable's rollback to gain a regulatory moat [1] — Gavin Baker "Anthropic has spent years advocating for an FAA-style approval process for AI. When a jailbreak report from a trusted partner triggered Fab…" 55:37 . David Sacks warns the US is on a shot clock — slowing down won't slow China, which operates outside US jurisdiction.
China's Z.AI released GLM 5.2, a 744 billion parameter open-source model with a 1 million token context window under the permissive MIT license.
GLM 5.2 API usage costs are 85% cheaper than GPT-5.5 for comparable performance, making it a highly cost-effective alternative.
The founder of Z.AI told Elon Musk that open-weight frontier-level AI capabilities will be available from China before Q1 2027.
China runs farms of tens of thousands of devices querying Claude and GPT APIs through masked accounts, harvesting every reasoning trace and feeding them back into their own models. This distillation cheat sheet has brought them to within a tick of the US frontier — for a fraction of the cost.
The future is composable models: a router sends 85% of queries to a cheap open-source model fine-tuned on your own data, and only the hardest tasks go to frontier models like Claude or GPT. Open source doesn't kill frontier labs — it shifts economic value to infrastructure and makes the whole system more powerful.
Anthropic has spent years advocating for an FAA-style approval process for AI. When a jailbreak report from a trusted partner triggered Fable to be rolled back, Anthropic got its regulatory moat — while simultaneously getting hoisted on its own petard. The question is whether this was calculated.
Chamath argued that despite China being 24 months behind on silicon and 6 months behind on models, they are only a few months behind in total AI capability.
DRAM is the single most important AI bottleneck. Micron revenue is up 4x to $42B, its entire 2026 supply is sold out, and DRAM will likely be 30–40% of all hyperscaler CapEx next year. The side effect: Apple prices are up 14–25% as data centers hoover up memory supply.
Micron reported revenue of $42 billion, up 4x year over year from $9 billion, beating expectations by 16% and raising Q4 guidance to $50 billion.
Chapter 5 · 1:01:46
Micron revenue up 4x to $42B [1] — Jason Calacanis "Micron revenue up 4x YoY: $9B to $42B: Micron reported revenue of $42 billion, up 4x year over year from $9 billion, beating expectations b…" 1:01:40 , entire 2026 supply sold out. DRAM is projected at 30–40% of hyperscaler CapEx. Apple forced to raise Mac prices 14–25% as AI data centers hoover up memory supply [2] — Jason Calacanis "Apple MacBook Neo price up 14–15%: Apple raised the price of the MacBook Neo from $699 to $799, a 14–15% increase, driven by the AI memory …" 1:03:00 .
Gavin Baker made a call on HBM makers like Micron as the best performing asset in the 2025 prediction show; Micron is up 14x since that time.
Apple raised the price of the MacBook Neo from $699 to $799, a 14–15% increase, driven by the AI memory crunch tightening DRAM supply.
Gavin Baker estimated that DRAM will likely represent 30–40% of all hyperscaler capital expenditure next year as AI infrastructure demand intensifies.
Chapter 6 · 1:10:17
Tesla's Megapod trademark filing suggests modular AI data centers at Supercharger stations. Travis describes putting compute in Adams restaurant kitchens. Debate over distributed training vs. inference efficiency constraints [1] — Travis Kalanick "Distributed training across physically separate locations is orders of magnitude less efficient — even 2 kilometers of fiber kills performa…" 1:12:14 .
A terrestrial 1-gigawatt data center costs $60 billion ($35B chips + $25B power/cooling). A reusable Starship puts the same compute in space for $40 billion. As terrestrial costs inflate and launch costs deflate, space wins. Running costs in orbit: about $1 billion per year for power.
Standing up a 1 gigawatt terrestrial data center costs approximately $35 billion in NVIDIA semiconductors plus $25 billion in power and cooling equipment.
Distributed training across physically separate locations is orders of magnitude less efficient — even 2 kilometers of fiber kills performance. But everything that kills distributed training is actually fine for inference. Distributed inference clouds are coming, and they could recycle unused compute from restaurants to homes.
With a reusable Starship, it would cost approximately $5 billion in launch costs to put a gigawatt of compute into orbit, making the total around $40 billion versus $60 billion terrestrially.
Chamath estimated that roughly 40% of all new data centers have faced regulatory or community contestation since 2021, a number he expects to rise.
Chapter 7 · 1:24:42
Gavin Baker values Anthropic at $3T [1] — Gavin Baker "Gavin Baker believes Anthropic would trade at $3 trillion as a public company, with revenue on track to hit well over $100 billion by year …" 1:27:48 . SpaceX lockup fear may be overstated as employees bought on IPO. Cerebras breaks deal price triggering systematic sell-off [2] — Gavin Baker "When Cerebras broke IPO deal price, giant fund managers with an ironclad rule to sell regardless of fundamentals triggered a doom loop. Sho…" 1:29:25 — Gavin advises always price IPOs to never break deal price.
Gavin Baker believes Anthropic would trade at $3 trillion as a public company, with revenue on track to hit well over $100 billion by year end. The market can absorb this because it's simply a shift from private to public — and the float will only ever be a small slice of those totals.
Gavin Baker stated his belief that Anthropic would trade at approximately $3 trillion as a public company, with revenue projected well over $100 billion by year end.
Cerebras stock broke below its IPO deal price within two days of its first earnings report as a public company, triggering systematic selling by funds that exit any stock that breaks deal price.
When Cerebras broke IPO deal price, giant fund managers with an ironclad rule to sell regardless of fundamentals triggered a doom loop. Shorts pile in to push a stock through deal price, price-insensitive sellers dump, and the stock goes far lower than any rational valuation would suggest. The lesson: price your IPO so you never get there.
Gavin Baker recalled that when he was at Fidelity, he priced Uber at $14 billion in an early round and was pilloried in the press; the round ultimately closed at $17 billion.
No indexed bits in this chapter.
China runs farms of tens of thousands of devices querying Claude and GPT APIs through masked accounts, harvesting every reasoning trace and feeding them back into their own models. This distillation cheat sheet has brought them to within a tick of the US frontier — for a fraction of the cost.
A terrestrial 1-gigawatt data center costs $60 billion ($35B chips + $25B power/cooling). A reusable Starship puts the same compute in space for $40 billion. As terrestrial costs inflate and launch costs deflate, space wins. Running costs in orbit: about $1 billion per year for power.
This episode
Factual claims made this episode, and whether a source was named.
Polymarket assigned only a 26% probability to a full Mamdani sweep of three NYC congressional primaries before Election Day.
Electing a Republican DA causes all-cause mortality for young Black men in that city to drop by 7%, according to an uncontested study.
Approximately 0.1% of the population commits around 70% of violent crimes.
GLM 5.2 achieved the highest score of any open-weight model ever on the Artificial Analysis Intelligence Index with 51 points.
GLM 5.2's API usage cost is 85% cheaper than GPT-5.5 for comparable performance.
Z.AI's founder told Elon Musk that open-weight fable-level AI capabilities will be available from China before Q1 2027.
GLM 5.2 and the GLM-5 model family were trained entirely on clusters of Huawei Ascend 910B chips.
Micron's revenue increased approximately 4x year over year, from $9 billion to $42 billion, beating expectations by 16%.
Micron's Q4 guidance was raised to $50 billion versus prior expectations of $43 billion.
Apple raised the price of the MacBook Neo from $699 to $799, a 14–15% increase, and raised the Mac Studio price by 25%.
Setting up a 1 gigawatt terrestrial data center requires approximately $35 billion in NVIDIA semiconductors plus $25 billion in power and cooling equipment.
With a reusable Starship, the launch cost to put a gigawatt of compute into orbit would be approximately $5 billion.
80% of Democrats now disapprove of Israel, according to recent polling.
Among Republicans under 50, the disapproval rating for Israel is now approximately 57%.
DRAM will likely represent 30–40% of all hyperscaler capital expenditure next year.
Anthropic would trade at approximately $3 trillion as a public company, with revenue projected well over $100 billion by year end 2026.
China has been approximately 9 months behind US frontier AI models, plus or minus 3 months depending on capability.
California's per-capita spending on homelessness has quadrupled while outcomes have gotten worse.
Almost half of SpaceX employees bought shares in the SpaceX IPO.
Approximately 40% of all new data centers have faced regulatory or community contestation since 2021.
This episode
NYC Mayor and DSA leader whose endorsed candidates swept all three congressional primaries in a major political upset.
Referenced as the driver behind Terafab memory fab, SpaceX orbital compute, and a potential Tesla-SpaceX merger; Z.AI founder addressed him directly about frontier AI timelines.
Anthropic CEO accused by Gavin Baker and David Sacks of deliberately engineering a regulatory moat by priming officials to roll back Fable after a jailbreak report.
DSA-aligned candidate who defeated two-term incumbent Dan Goldman in New York's 10th Congressional District primary over the Israel issue.
House Democratic leader whose endorsed candidate lost the NY-13 primary; now faces the prospect of managing DSA members if Democrats take the House.
Pro-Israel two-term incumbent congressman who led Trump impeachment efforts but lost New York 10th District primary to Brad Lander primarily over the Israel issue.
Far-left political organization using the Democratic Party as a ballot access vehicle to advance a radical platform including abolishing the Senate and seizing corporate assets.
AI lab whose Fable model was rolled back after a jailbreak report; discussed as potentially worth $3 trillion as a public company with revenue over $100B by year end.
Elon Musk's aerospace company that recently completed its IPO; central to the orbital compute thesis with Starship providing potential $5B launch cost for 1 gigawatt of space-based compute.
One of only three companies globally that manufacture high-bandwidth memory (HBM); reported revenue up 4x YoY to $42B with entire 2026 supply sold out.
Dominant AI chip manufacturer whose GPUs are central to data center buildouts; discussed as a potential beneficiary of open-source AI and possible future model competitor.
US frontier AI lab whose GPT-5.5 model was benchmarked against China's GLM 5.2; also launched its own Jalapeño chip via Broadcom partnership.
AI chip company that recently went public and broke below its IPO deal price within two days of its first earnings report, triggering a systematic sell-off.
Chinese AI company that released GLM 5.2, the highest-scoring open-weight model ever, trained on Huawei Ascend 910B chips, with plans for fable-level capability before Q1 2027.
Consumer electronics company forced to raise prices on Macs by 14–25% as the AI-driven DRAM crunch tightens memory supply for consumer devices.
Chinese technology company whose Ascend 910B chips were reportedly used to train GLM 5.2, central to China's AI indigenization push.
AI inference chip company mentioned as central to the composable model future and disaggregated inference; acquired by NVIDIA, invested in by Chamath.
Elon Musk's planned semiconductor fabrication facility focused on DRAM production, being developed with Intel partnership to address AI's memory bottleneck.
South Korean semiconductor company and one of only three global manufacturers of high-bandwidth memory (HBM) for AI data centers.
China's Z.AI frontier-class open-source AI model with 744B parameters, MIT license, highest-ever Artificial Analysis Intelligence Index score, trained on Huawei chips.
Stats
We use essential and analytics cookies to run Vuci. To understand how the site is used: Privacy Policy.
Install Vuci on your phone
Add it to your home screen for a faster, app-like experience.
Install Vuci on your phone
Tap the Share button, then “Add to Home Screen”.
A new version is available
Reload to get the latest Vuci.