Tim Dillon

The H20 ban is about a loophole, not tariffs: Sacks explains that Nvidia kept nerfing flops to duck each export threshold (H800, then H20), but in the reinforcement-learning/test-time-compute paradigm memory bandwidth matters more than flops — and the H20 has 20% more bandwidth than the H100, so “this chip is just frankly too good.” Nvidia guided a $5.5B quarterly hit and the stock dropped 6%; Sacks’s challenge to critics: “are you against export controls in general, or you just think that we’re drawing the line in the wrong place?”

Chamath’s explosive allegation — and the most tradeable claim of the episode: roughly 47% of Nvidia’s revenue goes to China and China-related countries, and he believes shell companies in Singapore, Bhutan, Cambodia and Vietnam are way stations feeding chips to China. “Is there a Cursor of Bhutan that we did not know?” Chamath says the U.S. needs a mechanism to understand and disclose the traffic. Sacks confirms evasion is real (TSMC’s ~3M dies into Huawei’s Ascend 910C) and wants more staff at Commerce’s BIS, the rare agency he’d expand.

The bear case on the ban, via Gurley/Baker and steelmanned by Friedberg: China launched a $37B push into 3nm last year, SMIC already shipped a 7nm chip in Huawei’s Mate 60 Pro, and “they’re doing it either way” — the only question is whether controls slow them at the AI layer. Chamath disagrees on timing: the tooling is “extremely non-trivial” and denial “slows them down quite a bit.”

Trump vs Harvard escalated to the nuclear option: after Harvard’s president refused the White House’s merit-hiring/end-DEI demands, $2.2B in grants were frozen and tax-exempt status is on the table. Sacks’s legal chain — Bob Jones (1983) established that enshrined racial discrimination forfeits exemption, SFFA v Harvard (2023) found Harvard discriminating, and its new five-essay diversity application is “suspiciously similar” to DEI statements: “Harvard wants to have its cake and eat it too.”

Friedberg’s numbers make the funding fight concrete: a $53.2B endowment throwing off ~$4B a year, Harvard billing the government an incremental 69% administrative overhead on NIH grants (average ~30%), and “the majority of the capital is being used to reinvest to make new capital” — raising whether universities, not corporate labs like Bell Labs or Google, should do fundamental research at all.

Sacks’s civilizational frame: China passed the US in share of published foundational research around 2019 (English-language only — add Chinese and it’s a decade earlier), while “we’re bumbling around talking about pronouns.” His never-told story: a top mortgage-market banker walked away from a Federal Reserve dream job after being told to “play up your Indianness” — “we lost him. We meaning the American taxpayer.”

Tim Dillon’s DEI-killed-Hollywood thesis: LA executives who “believe in nothing” chased marginalized voices because “they thought there was money there,” made TV for “guilty white liberals” that nobody watched, and once it stopped being profitable “rediscovered the profit motive.” DEI, in his telling, was never altruism — it’s “the establishment trying to preserve itself” through “ceremonial optical choices.”

Science corner: three papers point to mitotherapy as a new therapeutic modality — evidence of cell-to-cell mitochondria transfer (WashU 2023), the first mitochondrial map of the human brain (Columbia), and Zhejiang University stem cells producing 854x the mitochondria at 5.7x ATP efficiency, already healing bone in a mouse osteoarthritis model.

🔗 Original source & video: Trump vs Harvard, Nvidia export controls, how DEI killed Hollywood with Tim Dillon