TLDR
The All-In Podcast episode 278 covers the DSA's primary sweep in NYC signaling a socialist shift in the Democratic Party, China's open-source AI model GLM 5.2 rivaling US frontier models, Micron's blowout earnings revealing a structural HBM memory bottleneck driving 'AI-flation' in consumer electronics, and the market dynamics of massive upcoming AI IPOs including SpaceX, Cerebras, and Anthropic.
For you
PRO members can sign in and request more information — or their own version of this digest, tuned to what they care about.
Sign in to unlockSurprising Insights
GLM 5.2 was trained entirely on Huawei Ascend 910B chips, not smuggled NVIDIA GPUs, suggesting China's indigenization push is further advanced than assumed.
80% of Democrats now disapprove of Israel, making it a primary motivator for young voters and a litmus test in Democratic primaries.
Cerebras broke its IPO deal price within two days due to portfolio managers' rigid 'sell if below deal price' mandates, triggering a short pile-on unrelated to fundamentals.
Elon Musk's TerraFab could stand up a leading-edge fab in 90 days vs the typical 2-3 years, leveraging Intel partnership and SpaceX-style construction speed.
Key takeaways
DSA candidates won three NYC congressional primaries, reflecting a socialist takeover of the Democratic Party base driven by wealthy, downwardly mobile white progressives and migrant voters, with platforms including abolishing the Senate, ICE, and private property.
China's Z.ai released GLM 5.2, an MIT-licensed open-weight model matching Opus 4.8 on coding benchmarks, trained on Huawei Ascend chips, signaling China is ~6 months behind US frontier models but closing fast via distillation and indigenous silicon.
Micron's 4x revenue growth and sold-out 2026 HBM supply expose DRAM as the critical AI bottleneck; memory now represents 30-40% of hyperscaler capex, forcing Apple to raise Mac prices 14-25%.
Anthropic could be worth $3T as a public company; the market must absorb ~$4T in AI offerings (SpaceX, Anthropic, OpenAI, Cerebras) but global capital markets can handle the float if priced via auction not underwriting.
Distributed inference via modular 'megapods' (Tesla/SpaceX) and orbital compute (Starship) may bypass terrestrial power/land constraints, with SpaceX potentially achieving $40B/gigawatt in orbit vs $60-70B on ground within 3-4 years.
Notable quotes
“AI is the greatest economic leveler we'll ever find in our lifetime. It takes the world's knowledge and allows every single man, woman, and child to have an equivalent Travis Kalanick as his co-founder.”
▸ Notable quotes · 5:33“We're using the Democratic Party as a ballot access vehicle. Not because we share its goals. We build our own organization, get elected under the Democratic label, caucus with Democrats when it's useful, and push our own agenda from the inside.”
▸ Notable quotes · 10:51“The Chinese have been nine months behind our models, plus or minus three months. But when they know there's been a breakthrough around something like cyber, they can deploy more resources against that particular problem and catch up faster.”
▸ Notable quotes · 57:34“When Starship is reusable, it's going to cost $5 billion to put a gigawatt of compute into space. You're at $60 billion terrestrially. And the $25 billion that is powering compute is clearly inflationary.”
▸ Notable quotes · 72:26Chapters
1:00 — DSA Primary Sweep in NYC: Socialist candidates backed by Zohran Mamdani won three NYC congressional primaries, signaling a DSA takeover of the Democratic Party base. Candidates advocate abolishing the Senate, ICE, prisons, and private property. The shift is driven by wealthy, downwardly mobile white progressives and migrant voters.
3:00 — AI as Economic Leveler vs. Doomerism: Chamath argues AI democratizes expertise (every person gets a 'super founder' co-founder) but Silicon Valley's poor messaging and infighting have ceded narrative to doomers and socialists. Saks warns the Democratic Party faces a choice between communism and nationalism.
45:00 — China's GLM 5.2 and Open Source AI Race: Z.ai's GLM 5.2 matches Opus 4.8 on coding benchmarks, trained on Huawei chips under MIT license. Distillation from US frontier APIs accelerated Chinese progress. US regulatory delays (Anthropic's FAA-for-AI push) risk ceding lead while China scales indigenous silicon.
62:00 — Micron Earnings and HBM Bottleneck: Micron's 4x revenue growth and sold-out 2026 HBM supply confirm DRAM as the critical AI bottleneck. Memory is 30-40% of hyperscaler capex. Apple raised Mac prices 14-25% due to DRAM scarcity. Only three companies (Micron, SK Hynix, Samsung) can make HBM; China's CXMT targets consumer DRAM.
88:00 — AI IPO Wave and Market Dynamics: SpaceX IPO succeeded but float is small; employees and insiders bought heavily. Cerebras broke deal price due to rigid PM selling rules. Anthropic could be worth $3T. Global markets can absorb offerings if priced via auction. Orbital compute via Starship may undercut terrestrial data center economics.
Speakers
Jason Calacanis — Host
Chamath Palihapitiya — Co-host
David Sacks — Co-host
Travis Kalanick — Guest, Founder of Adams
Gavin Baker — Guest, Managing Partner at Atreides Management