Socialists Sweep NYC, China Catches Up in Coding, AI Memory Crunch, Micron's Blowout Quarter
Summary
New York’s Democratic-socialist sweep showed how disciplined activists can capture safe seats and pull an entire party left. Polymarket put the three-candidate sweep at just 26% before election day, yet Mamdani-backed challengers won in NY-7, NY-10, and NY-13, including two defeats of established incumbents. David Sacks’s key investor-relevant point was organizational leverage: the DSA views Democrats as “a ballot access vehicle,” so even members who survive may alter votes and rhetoric to avoid primaries.
The panel attributed socialism’s momentum less to working-class demand than to downward mobility, institutional failure, and exceptional political communication. Gavin Baker described the emerging base as relatively affluent, highly educated white liberals whose outcomes differed sharply from peers in industry, while Chamath Palihapitiya emphasized housing costs, college debt, healthcare dysfunction, and inflation as proof to younger voters that “the system’s rigged.” Baker’s changed view was explicit: he once considered AOC the Democrats’ standout communicator, but now calls Zohran Mamdani “one of the most talented politicians I’ve ever seen.”
The proposed cure for youth radicalization exposed a fundamental split between child protection and anonymous speech. Chamath Palihapitiya argued that under-16 social-media bans could reduce early exposure and produce “a far less radicalized youth”; Travis Kalanick agreed that social media is “brain rot for real” but said age verification is really a mechanism to deanonymize adults and construct “a full-scale censorship regime.” Baker supported the child-safety objective but judged Kalanick’s free-speech cost “really high.”
Israel has become a primary-election fault line whose intensity is increasingly explained by age rather than party alone. Sacks presented Brad Lander’s defeat of pro-Israel incumbent Dan Goldman as close to a single-issue test, citing 80% Democratic disapproval of Israel and 57% disapproval among Republicans under 50. Chamath insisted critics must separate Jews, Israelis, the state of Israel, and Benjamin Netanyahu: collapsing them into one object of blame is “insane.”
GLM 5.2 turns China’s open-model progress from a strategic warning into an immediate commercial constraint. The 744-billion-parameter, one-million-token-context MIT-licensed model scored 51 on Artificial Analysis, beat GPT 5.5 on Frontier SWE, trailed Claude Opus 4.8 by under one point, and was described as 85% cheaper than GPT 5.5 at comparable performance. With Opus 4.8 rolled back and GPT 5.6 navigating approvals, Sacks warned that America is “on a shot clock”: stopping domestic deployment will not stop China.
Open weights may compress frontier-model margins while expanding the infrastructure opportunity. Baker expects enterprises to run a “council of LLMs,” routing perhaps 85% of work through customized open models and escalating only difficult tasks to frontier systems; his rough current split was open models processing 80%-plus of tokens while frontier tokens capture about 90% of economic value. His conclusion was not that frontier labs disappear, but that composability shifts value toward chips, memory, hosting, and routing.
Micron’s quarter confirmed that DRAM—not exotic components—is the binding AI bottleneck and a source of consumer inflation. Revenue rose from roughly $9 billion to $42 billion, Q4 guidance reached $50 billion versus $43 billion expected, and 2026 HBM supply was already sold out; Baker expects DRAM to absorb 30%-40% of hyperscaler capex next year. Apple’s $699 MacBook Neo moving to $799 and Mac Studio pricing rising 25% illustrated the spillover: “Inflation has come to the desktop.”
Scarce terrestrial power strengthens both orbital-compute economics and the value of large energized sites, while public markets remain able to fund the buildout. Baker put a one-gigawatt terrestrial data center at $35 billion of semiconductors plus $25 billion of power and cooling, versus approximately $40 billion in orbit if reusable Starship lowers launch to $5 billion. He separately valued Anthropic at roughly $3 trillion and argued that public markets need absorb only the offered slice, but Cerebras showed why execution matters: once an IPO breaks deal price, mechanical selling and opportunistic shorts can turn disappointment into “a pile-on.”
Deep dive
1. The DSA sweep converted low-turnout organization into congressional power
Mamdani-endorsed candidates went three-for-three: Brad Lander unseated two-term incumbent Dan Goldman in affluent NY-10, the NY-13 challenger defeated a five-term incumbent backed by Hakeem Jeffries, and Claire Valdez won NY-7. Polymarket had priced the full sweep at only 26%.
Jason’s demographic observation was deliberately barbed: the candidates performed especially well with younger, college-educated, high-income voters—“people who can afford to be socialists.” Because all three districts are safely Democratic, the primary victories likely determine the eventual officeholders.
Sacks read the DSA platform as a constitutional reconstruction project: abolishing the Senate, carceral state, ICE, and Electoral College; granting universal amnesty; subordinating the executive and judiciary to Congress; expanding the House; imposing proportional representation and ranked-choice voting; and pursuing public ownership of major companies.
More consequential than any single seat was the organization’s declared strategy. One DSA co-chair described Democrats as “a ballot access vehicle,” useful for winning office and caucusing when convenient, while calling the establishment “an obstacle, not a home.”
2. Primary threats may move more Democrats than DSA seat totals suggest
Sacks argued that the DSA thrives in contests like New York’s primary, where turnout was roughly 17%: activists are passionate, highly organized, and practiced at exploiting low-participation electoral systems. He expects similar tests in Los Angeles, where he said the DSA holds something like half the city council seats.
His forward mechanism was straightforward: after strong incumbents lost, every Democrat in a deep-blue district must consider the possibility of being primaried. Members therefore may shift rhetoric, votes, and positioning toward DSA priorities even when relatively few formal DSA candidates win.
Jason compared the strategy with Trump’s takeover of the Republican Party: build a large enough tent, dominate the communication channel, then force established officials to follow the new base. The DSA’s own formulation was equally direct: Democrats cannot represent both “the billionaire class and the working class.”
Mamdani’s communication skill made the threat harder to dismiss. Jason described being briefly captivated by a Knicks speech despite opposing Mamdani’s politics: “It was like getting hypnotized.” Sacks’s response captured the vulnerability—“That’s all it takes. We’re screwed.”
3. Economic frustration supplies the audience, but charisma supplies the leader
Baker said the old Democratic coalition sought opportunity for working-class, Black, and Hispanic Americans; in his reading, the DSA is losing those voters while gaining relatively wealthy white liberals who are downwardly mobile. Many moved from elite schools into an NGO-and-nonprofit system rather than industry.
His sharper institutional claim invoked the “Curley effect”: politicians can support policies that worsen constituent outcomes yet drive out rivals, then award well-paid nonprofit jobs to allies. Baker cited homelessness spending per capita more than doubling in New York and roughly quadrupling in California while outcomes deteriorated.
Chamath supplied the demand-side explanation: younger generations face expensive housing, college debt, healthcare dysfunction, and inflation, and increasingly doubt they can outperform their parents. With little conception of socialism, communism, or the historical examples Sacks cited, “democratic socialism” arrives in an attractive wrapper—“the Coke Zero of socialism.”
Baker nevertheless credited the movement’s ascent principally to Mamdani, not its ideas. His updated judgment: “I used to think AOC was by far the most talented Democratic politician,” but Mamdani is a singularly capable speaker, interviewer, and “chameleon” with no obvious establishment rival.
4. AI was pitched as capitalism’s leveler—and Silicon Valley as its worst ambassador
Chamath called AI “the greatest economic leveler we’ll ever find in our lifetime.” Search gathered the world’s information, but AI converts that information into expertise and action, giving anyone the functional equivalent of a brilliant co-founder who can “out-engineer people” and “outthink people.”
In that framing, access is no longer gatekept; outcomes depend on a person’s ability to direct the available intelligence toward something valuable. Yet Silicon Valley’s public infighting allowed narratives about job losses, water consumption, and doom to dominate, leaving capitalism’s alternative poorly represented.
Chamath said congressional races in Utah and New York became practical referenda on AI, with Anthropic-funded anti-AI groups opposing what he described as OpenAI-funded pro-AI groups. The pro-AI side “held the line,” but only narrowly.
Kalanick offered a deeper social diagnostic: “Truth and justice is the immune system for society.” When truth loses to distorted media or crimes carry no consequences, the immune system is suppressed and social pathologies flare; the trajectory of those two variables predicts whether society improves or worsens.
5. Social-media age gates split the panel over censorship
Chamath argued that if political stability and predictability emerge in Canada, the UK, and Australia after their under-16 bans, that could be evidence for the bans; he pointed to Florida’s existing action and suggested considering broader U.S. adoption. His theory was that a healthier information diet could yield less-radicalized future voters.
Kalanick mounted a full counterargument while conceding the harm: social media is bad for children and adults, “brain rot for real,” potentially worse than cigarettes. But he said child protection supplies the justification for forcing adults to identify themselves, enabling authorities to suppress disagreement as “harmful content.”
Baker agreed that an under-16 ban would be beneficial in isolation, but said the price Kalanick identified was substantial. Anonymous accounts and free speech on X, in his view, prevented a much worse information environment; he supported age restrictions only if anonymity could survive.
Sacks paired free speech with civic formation. Hearing overtly anti-American classroom material during COVID helped move some Democrats toward Trump, he said; schools should teach national failures honestly while still explaining why America is “about as good as it gets” and an unusually successful multicultural society.
6. Israel now divides younger voters across both parties
Sacks treated Lander’s victory over Goldman as unusually clean evidence. Both were veteran white Jewish New York politicians with broadly progressive credentials, but Goldman defended Israel while Lander visited a mosque and denounced what he called genocide in Gaza; Lander won handily.
The polling Sacks cited was stark: 80% of Democrats disapproved of Israel. Among Republicans under 50, disapproval was 57%, whereas older, more establishment and Fox-oriented Republicans remained much more supportive—making age the clearest divider.
Chamath compared the authority granted Israeli leadership after October 7 with America’s response after September 11, including passage of the Patriot Act. He understood the initial mandate to “set the table right,” but argued that criticism of Netanyahu had become wrongly conflated with hostility toward Israelis or Jews.
Jason diagnosed a communications vacuum: Israel needed a sub-35-year-old American-Israeli spokesperson fluent in English and social media, with comparable representatives in France, Germany, and other countries. Instead it often deployed admirable older figures who lacked the medium’s native language while “taking body blows every second.”
7. GLM 5.2 brought open weights to the coding frontier
China’s Z.AI released GLM 5.2 under the MIT license with 744 billion parameters and a one-million-token context window. It is downloadable, self-hostable, forkable, free of regional restrictions, and usable commercially with only the license attribution.
The reported performance placed it beside closed frontier systems: 51 on the Artificial Analysis Intelligence Index, the highest open-weight score cited; better than GPT 5.5 on Frontier SWE; and less than one percentage point behind Claude Opus 4.8. API use was described as 85% cheaper than GPT 5.5.
Baker said the quality challenged some of his prior beliefs, though he considered large-scale distillation certain. His analogy was an enormous device farm querying cloud APIs through masked accounts, harvesting full reasoning traces, then feeding those traces into reinforcement learning and possibly pre-training.
Distillation is effectively a “cheat sheet,” but GLM 5.2 may now be capable enough to conduct its own reinforcement learning. Baker hedged the conclusion because he had not seen the next OpenAI or SpaceX models; the frontier gap might widen again.
8. Regulation cannot preserve a lead against a competitor outside US jurisdiction
Sacks emphasized the timing: Opus 4.8 had been rolled back after a reported jailbreak, while GPT 5.6 was navigating additional approval hoops. That left a Chinese open-weight model competitive with the strongest currently available OpenAI and Anthropic systems.
Jason asked whether Dario had deliberately provoked the outcome to create a regulatory moat. Sacks noted that Dario had advocated an “FAA for AI,” so Anthropic had obtained the approval structure it sought, but probably did not welcome losing Opus 4.8—possibly being “hoisted on his own petard.”
Sacks opposed rewarding that strategy with a labyrinthine regulator. Once the jailbreak is resolved, Anthropic should return to market, and OpenAI should not face unnecessary delay: “We do not have months to give away in this race.”
Cyber risk, in his account, requires acceleration rather than suppression. White hats need frontier tools to find vulnerabilities, distribute patches, and force a broad upgrade cycle before adversaries exploit them; China, historically around nine months behind “plus or minus three,” can close faster after a known breakthrough.
9. Composable models shift the profit pool toward infrastructure
Baker’s enterprise architecture is a “council of LLMs”: a router sends routine tasks first to a company’s tuned open-weight model, then invokes a frontier system for harder work, verification, or added reasoning. He argued Grok should be one of the frontier choices because it will tell an owner politically inconvenient truths.
He suggested open systems might process 85% of enterprise queries, even though frontier tokens currently capture roughly 90% of economic value while open-source systems already handle more than 80% of tokens. Those ratios could persist; the models perform different economic jobs.
Open weights are therefore not bearish for the AI complex as a whole. They transfer margin from frontier laboratories to compute, memory, hosting, and routing providers—making Nvidia, in Sacks’s phrase, “the American open-source champion,” capable of releasing a GLM 5.2-class model when incentives permit.
Nvidia’s restraint may reflect channel conflict, but OpenAI’s newly announced Jalapeno chip, reportedly built with Broadcom, could alter that calculation. Jason’s speculation: if frontier labs enter chips, Nvidia may become more willing to compete directly in models.
10. China is preparing an exportable AI stack, not merely a benchmark model
Z.AI claimed the GLM-5 family was trained entirely on Huawei Ascend 910B clusters; DeepSeek V4 had attracted a similar claim. Sacks kept the hedge explicit—China might be lying or using smuggled Nvidia hardware—but viewed the broader indigenization push as credible.
GLM 5.2 inference is optimized for Huawei chips. Sacks expects China to package Huawei hardware with optimized models as “AI in a box” and sell it globally at a fraction of US cost, following its familiar formula of better, cheaper, faster—or at least almost as good.
That prospect reinforced his case for US exports: America could have supplied friends and partners before Chinese substitutes matured, yet “invented reasons not to sell abroad.” He expects China to be globally competitive within one or two years.
Hosted Chinese models still refused Jason’s Taiwan and Tiananmen Square prompts, but Sacks did not regard censorship as fatal to open weights. American companies such as Perplexity had already forked Chinese models and restored excluded content, separating the base capability from the hosted political layer.
11. DRAM became AI’s tollbooth and consumer electronics’ inflation channel
Micron’s revenue rose roughly fourfold, from $9 billion to $42 billion, beating expectations by 16%; Q4 guidance reached $50 billion against $43 billion expected. Its full 2026 HBM supply was sold. Jason said Micron’s stock was up about 10x, while separately noting that Baker’s earlier HBM-maker call was up 14x.
Baker dismissed the “bottleneck bros” searching for obscure components: “The bottleneck that matters is DRAM.” Capacity and bandwidth underpin every model, and only Micron, SK Hynix, and Samsung were described as capable of producing the advanced memory needed for AI servers.
Micron’s supply agreements now include price floors and ceilings and cover about half its revenue with only four customers. Baker said the new floor pricing exceeds previous cycle peaks on a gross-margin basis, potentially transforming an industry historically valued as cyclical.
HBM’s difficulty comes from stacking dies—eight high in HBM3, moving toward 12 and 16—and packaging them reliably. New capacity takes years, while data centers “hoover up” memory; Baker expects DRAM to consume 30%-40% of total hyperscaler capex next year.
12. Scarce power favors orbital compute, modular deployment, and execution-led valuations
Baker’s one-gigawatt terrestrial bill was $35 billion of Nvidia semiconductors plus $25 billion of power and cooling. Reusable Starship could cut launch to $5 billion, producing roughly $40 billion of orbital cost versus $60 billion on Earth today—and perhaps $70 billion terrestrially within three or four years.
Jason said about 40% of data centers had been contested since 2021 and expected that share to rise. Chamath separately described energy supply as very meager against effectively infinite demand, arguing that large energized terrestrial sites could become “Hope diamonds,” while modular pods could compress build cycles to approximately 90 days.
Training still requires tightly co-located chips—even two kilometers can destroy efficiency—but distributed inference is more plausible. Separating memory-capacity-heavy prefill from bandwidth-heavy decode could extend H100s and A100s for seven, 10, or 12 years; Kalanick cited Targon offering H200s permissionlessly for $3-$4 per hour.
Baker put Anthropic at roughly $3 trillion today, said it would end the year well over $100 billion, and argued it could be very profitable at inference scale, with people reporting 85% gross margins. For Cerebras, however, the operative metric was energized megawatts: adding 50 MW monthly in 2027 could imply a roughly $9 billion cloud run rate, but breaking IPO price triggered mechanical selling—another argument for leaving pricing room or, as Jason argued, using an auction.