Socialists Sweep NYC, China Catches Up in Coding, AI Memory Crunch, Micron's Blowout Quarter

Zohran Mamdani-aligned DSA candidates swept New York's 7th, 10th, and 13th congressional districts, defeating two seated incumbents despite Polymarket giving the sweep only a 26 percent chance before election day. All three seats are safe Democratic districts, making general election wins nearly certain. Voter turnout in the primary was approximately 17 percent, and speakers attributed the outcome to DSA's ability to exploit low-turnout elections through passionate organization, ballot harvesting, and ranked choice voting rules. Mamdani himself won the New York City Democratic mayoral primary, and speakers described him as possessing Obama and Trump level charisma while applying the Trump political takeover playbook to the Democratic Party from within. DSA co-chair Josh Block has stated the organization uses the Democratic Party as a ballot access vehicle rather than because it shares its goals, and co-chair Gustavo Grdillo openly described the goal of building independent political infrastructure inside the party, which one speaker characterized as using the party as a host organism.

DSA's platform includes abolishing the Senate, ICE, and the electoral college, granting universal amnesty, replacing the president and Supreme Court with bodies subordinate to Congress, and pursuing public ownership of major corporations. Speakers credited Mamdani's personal political talent rather than DSA's ideas as the primary driver of the movement's rise, rating him above AOC in that assessment. DSA's voting base was described as relatively wealthy, downwardly mobile white liberals, with the organization losing ground among working-class, Black, and Hispanic voters. Speakers identified housing costs, college debt, and healthcare costs as the underlying conditions creating two generations of young people who expect to do worse than their parents, which DSA is exploiting politically. Eighty percent of Democrats now disapprove of Israel, and among Americans under 50 the figure is 57 percent, a collapse from previously high bipartisan approval ratings that shaped the New York 10 race between two Jewish candidates into a straight referendum on the Gaza question.

China's Zhipu AI released GLM 5.2, a 744 billion parameter open-weight model with a 1 million token context window under the MIT license with no regional restrictions. It scored 51 points on the Artificial Analysis Intelligence Index, the highest score ever recorded for an open-weight model, trails Claude Opus 4.8 by less than one percentage point on the frontier SWE coding benchmark, and costs approximately 85 percent less than GPT 5.5 for comparable performance. Speakers described a method by which Chinese AI companies harvest reasoning traces from frontier model APIs through masked accounts on consumer devices to feed reinforcement learning and pretraining at a fraction of normal cost, and argued that once a model becomes capable enough to run its own reinforcement learning, this catch-up mechanism becomes self-sustaining and may be impossible to reverse. Z.AI's founder stated GLM 5 was trained entirely on Huawei Ascend 910B chips, though speakers noted some smuggled Nvidia chips could have been involved.

Micron reported revenue of approximately 42 billion dollars, up roughly four times year over year from 9 billion dollars, beat earnings expectations by 16 percent, and raised Q4 guidance to 50 billion dollars against a prior expectation of 43 billion dollars. Micron's entire 2026 HBM supply is already sold out, and only three companies produce high bandwidth memory: Micron, SK Hynix, and Samsung. Gavin called DRAM the single most important bottleneck in AI and predicted it will represent 30 to 40 percent of all hyperscaler capital expenditure next year. Supply chain agreements covering roughly 50 percent of Micron's revenue include floor pricing that exceeds prior cycle peak gross margins. CXMT going public in China may flood the market with cheap consumer-grade DRAM but cannot produce the HBM required for AI servers.

Gavin estimated that standing up a one gigawatt terrestrial data center costs approximately 60 billion dollars in total, and calculated that when Starship becomes rapidly reusable, putting a gigawatt of compute into orbit would cost approximately 40 billion dollars with ongoing power costs of roughly 1 billion dollars per year, making orbital compute economically competitive with terrestrial data centers. A trademark filing dated June 18, 2026 for the name Megapod covered modular data center hardware for AI computing sold as a unit. Prefab shipping container compute units can be deployed with roughly a 90-day build cycle. Distributed training efficiency drops dramatically when GPU nodes are physically separated, while inference workloads are increasingly disaggregated into pre-fill and decode stages and are well suited to distributed deployment.

Gavin estimated Anthropic at approximately 3 trillion dollars as a public company valuation, projected to end the year with well over 100 billion dollars in revenue at 85 percent gross margins on inference. Saks noted Anthropic's Fable model was rolled back after a credible jailbreak report and argued the correct response to AI cyber risk is proactive white-hat vulnerability discovery rather than model delays, warning that a labyrinthine approval process would enable regulatory capture. Cerebras signed a contract with OpenAI in December 2025 worth approximately 20 to 25 billion dollars, with financial impact not expected to be visible until around Labor Day due to a roughly seven-month production and deployment cycle, and is projected to add approximately 50 megawatts of capacity per month in 2027.