Tech Stock Shock + Solving the Mystery of OpenAI's "Blip" + Tinder's Flirt-Off
Summary
The tariff package landed as an immediate tech-valuation shock: Apple fell more than 8%, Amazon more than 7%, and Nvidia 6.5%, with Kevin Roose estimating roughly $1 trillion could disappear from major tech market caps that day. A 10% baseline tariff, plus country-specific rates including an additional 34% on China, 32% on Taiwan, and 20% on the EU, could raise a new iPhone’s price by 43%. These were, Roose said, “the biggest tariffs we have seen in basically a century.”
The policy assumes manufacturing can return to America, but the hosts see neither the capacity nor a transition plan to make that happen quickly. Apple has spent years diversifying from China, yet the vast majority of its products are still made there; US labor costs and missing manufacturing expertise compound the problem. Companies will pass costs to consumers, consumers will buy less, and “we can’t just spin up domestic manufacturing capacity overnight.”
AI’s already-difficult economics may deteriorate even if semiconductors themselves remain exempt. The announcement appeared to carve out chips, but graphics cards containing Nvidia GPUs might still be tariffed, while cooling systems, fans, racks, and other cluster infrastructure almost certainly become costlier. Frontier labs were already “burning billions of dollars,” Newton noted, so their path to profitability may have “just got a lot longer.”
The end of the under-$800 de minimis exemption directly threatens the operating models of Temu, Shein, and China-based Amazon sellers. Roose said he did not know “how these companies survive” if duty-free low-value shipments disappear within weeks. Trump-aligned tech leaders may seek carve-outs for Tesla, Starlink, SpaceX, and other exposed businesses, but the hosts called the broader strategy a “very dangerous game of chicken.”
Sam Altman’s 2023 firing emerged from an accumulated pattern of alleged “oopsie moments,” not one decisive offense. Complaints included failing to tell the board about ChatGPT’s planned release, personally owning the OpenAI Startup Fund, and telling Mira Murati that legal had said GPT-4 Turbo did not need to go through a joint Microsoft safety board, which the top lawyer denied. There were also concerns about giving people incompatible accounts. To Ilya Sutskever and Mira Murati, Altman repeatedly said “whatever it took to get what he wanted”; to the board, the repetition began to look deliberately deceptive.
The Blip was also a governance catastrophe in which Sutskever and Murati helped create an outcome they quickly abandoned. Sutskever had wanted Altman removed for years, but Hagey says Murati expected an executive coach, not a firing; the board interpreted her willingness to become interim CEO as support. Once employees threatened to leave for Microsoft, the two deputies became “the dog that caught the car.”
Altman’s durable control rested on capital formation: a tender offer valued OpenAI near $90 billion, up from $30 billion, and employees feared their expected millions would vanish without him. Hagey calls fundraising his “superpower” and Altman an “accelerating force”; investors in a room with him “have no defenses.” Her deeper read is that he “wants to matter”—to reshape society and become a “great man of history,” a characterization he rejects.
The Blip still shadows Microsoft, while Tinder’s flirting bot illustrates a smaller but telling AI product risk: solving a showcase problem instead of the customer’s real one. Microsoft publicly rescued Altman but privately built a “lifeboat” around Mustafa Suleyman; Roose predicts a “messy breakup” as OpenAI’s ambitions overlap Microsoft’s. Tinder’s Game Game produced robotic conversations and zero-flame scores, leaving Newton wondering why Tinder was shipping “shiny little novelties” instead of addressing one-word replies and ghosting.
Deep dive
1. The tariff announcement repriced tech before its second-order effects arrived
As recorded Thursday, Roose saw Apple down more than 8%, Amazon more than 7%, and Nvidia 6.5%—moves large enough that “something like a trillion dollars” might evaporate from the biggest technology companies’ market capitalizations by day’s end.
The structure paired a 10% baseline tariff on goods entering the United States with higher country-specific levies: an additional 34% on Chinese goods, 32% on Taiwanese goods, 20% on EU goods, plus measures affecting Vietnam, India, and others. “People were not expecting this much.”
The administration’s stated logic was fairness: foreign producers sell into America while some countries tariff US exports, so reciprocal pressure might level the field and stimulate domestic production. Newton’s pushback was that seeking perfectly balanced bilateral trade misreads an economy built around comparative advantage.
Roose’s immediate mechanism was simpler: manufacturers will not absorb the entire tariff while keeping prices unchanged. They will pass costs to buyers, and buyers will reduce purchases—as happened when pandemic-era supply disruptions produced higher prices and shortages.
2. An iPhone price shock exposes the fantasy of instant reshoring
One firm estimated that tariffs affecting China and Vietnam, where new iPhones are assembled, could lift a new device’s price by 43%. That was Roose’s clearest illustration that this is “not a small change in the cost of goods.”
Apple had anticipated geopolitical risk and spent years trying to diversify production away from China, yet the vast majority of its products remained made there. America lacks comparable manufacturing expertise, pays much higher labor rates, and may not have enough workers seeking electronics-assembly jobs.
Newton underlined the missing bridge: the announcement came with no plan to create domestic capacity, while new factories would take years. The administration can declare, “We’re gonna make all this stuff here now,” but neither host saw a practical route from tariff to replacement supply.
3. De minimis ends—and the tariff formula may have been chatbot-shaped
The de minimis exemption had allowed shipments worth less than $800 to enter duty-free, supporting Temu, Shein, and many China-based Amazon sellers. With that exemption expected to disappear within weeks, Roose’s blunt assessment was: “I don’t know how these companies survive.”
The hosts entertained, without evidence, the theory that the tariff formula had been “vibe coded” with ChatGPT. Roose doubted officials simply accepted a chatbot answer, noting that economists had better tools, but conceded AI might have been consulted somewhere in the process.
Newton’s narrower observation gave the joke substance: ask multiple chatbots how to eliminate bilateral trade deficits and they produce something resembling the administration’s approach. Economists may consider zeroing every deficit “insane,” but a model will still answer the requested optimization problem.
4. AI infrastructure faces costs that a semiconductor carve-out cannot contain
Semiconductors appeared exempt, suggesting US labs could still acquire Taiwanese GPUs without the new duties. Close readings created a crucial ambiguity, however: CPUs may be exempt while graphics cards containing Nvidia accelerators “may not be exempt.”
For Microsoft, Google, and Amazon, which were buying enormous volumes of advanced GPUs, that distinction could raise model-training costs sharply. Even a later clarification exempting GPUs would leave the rest of the cluster—fans, cooling systems, server racks, and supporting hardware—exposed.
Newton connected the infrastructure bill to industry economics: frontier labs were already burning billions before tariffs. Higher capital costs therefore extend an already uncertain path to profitability rather than merely trimming mature-company margins.
Trump-supporting technologists began responding with “Whoa, whoa, whoa,” Roose said. Tesla’s Chinese manufacturing and likely overseas inputs for Starlink and SpaceX make Elon Musk an obvious petitioner; based on the first Trump administration, Roose expected Apple and others to pursue company- or supply-chain-specific exemptions.
5. The Blip grew from repeated trust failures, not a single smoking gun
Hagey’s account came from more than 250 interviews, some secured only after sources declined for over a year. The Optimist began as an unauthorized biography that Altman opposed, though he eventually participated in numerous interviews; the Blip alone affected a company of nearly 800 people.
The board’s complaints included learning late about ChatGPT’s planned release, discovering Altman personally owned the independently operated OpenAI Startup Fund that was meant to be managed by the board, questioning whether GPT-4 enhancements received proper safety review, and suspecting he tried to mislead one director about another.
Sutskever’s distrust reached back to 2021. He formed a team around an idea for reasoning models, rival researcher Jakub Pachocki formed another, the teams merged under Pachocki, and Sutskever moved toward superalignment—an outcome Hagey described as a power struggle he had, in a way, lost.
Resentment deepened when Altman elevated Pachocki and told both men they could set OpenAI’s research direction. Hagey called that “telling everyone what they wanna hear”; Sutskever saw years of saying one thing, doing another, then presenting the discrepancy as an accident.
6. Murati supplied the receipts but expected coaching, not a coup
Hagey says Murati described a broader “toxic” playbook: Altman would say whatever secured the desired result, then undermine anyone who resisted. Greg Brockman added organizational friction by crashing into projects, producing brilliant work and chaos, while sitting on the board despite effectively reporting to Murati.
Sutskever and Murati collected evidence in disappearing documents, much of it screenshots from Murati’s Slack. In one example, Altman reportedly told Murati that legal had said GPT-4 Turbo did not need to go through the joint Microsoft-OpenAI safety board; OpenAI’s top lawyer replied, “I definitely didn’t say that.”
Any isolated incident could be an explainable “whoopsie moment.” The board, having experienced its own safety-related conflicts with Altman, began to view the repeated discrepancies as deliberately deceptive.
Hagey says Murati expected the board to prescribe an executive coach for a fixable leadership problem. The board instead named her interim CEO and treated her assent as endorsement; she did not initially understand how central her complaints had been to the firing decision.
7. Employees restored Altman to protect both OpenAI and a $90 billion tender
Hagey portrays Murati and Sutskever as backing away from the outcome they had helped create once employees threatened to resign en masse for Microsoft. Newton preserved the uncomfortable counterfactual: they could have publicly shared what they had seen over the prior six months, but did not. Hagey agreed they were “the dog that caught the car.”
Money made collective action easier. A pending tender valued OpenAI at nearly $90 billion, up from $30 billion; newer employees with strike prices around the lower valuation expected immediate gains and believed those millions would “go up in smoke” if Altman departed.
Some letter signatories told Hagey they “didn’t even really think about it very hard.” Because they believed the company would go to zero if Altman left, threatening to move to Microsoft did not require everyone to share a considered judgment about Altman’s conduct.
Had the firing held, Hagey thinks employees would have been poorer and AI development might not be as far along. Altman is an “accelerant,” obsessed with speed and unusually capable of raising the next round required by a business that still “doesn’t make money.”
8. Altman’s fundraising gift explains both his indispensability and the exodus around him
Newton’s puzzle was the contrast between a friendly, accessible interview subject and an industry populated by former allies: Elon Musk at xAI, Anthropic’s founders, Murati at Thinking Machines, and Sutskever at Safe Superintelligence. Hagey’s answer was that fundraising sits at the center of everything.
Hagey relayed one person’s description of fundraising: investors “have no defenses. They just give him money.” One-on-one meetings let Altman create a complete world for each listener; running an organization requires everybody to inhabit the same world, and “when people start comparing notes,” contradictions become damaging.
Roose rejected money as a sufficient personal motive because leading AI figures were already extraordinarily wealthy. Hagey’s answer: “I think Sam wants to matter”—to reshape society and become, as an early acquaintance put it, “a great man of history,” though Altman strongly denies that framing.
His ambitions extend beyond OpenAI into nuclear fusion and other moonshots. Hagey called him an “imagination hype man,” always positioning breakthroughs as imminent; his objection that her biography came too early suggested to Roose that Altman considers his historical work unfinished.
9. The board lost the communication war, and Microsoft built an insurance policy
Asked whether Altman should have been fired, Hagey answered, “Probably not.” None of the incidents was individually fireable, and the later investigation concluded that the old board was entitled to its interpretation while the new board interpreted the same facts differently.
The decisive weakness was that the directors’ personal loss of trust could not be communicated to a broader constituency despite Altman’s central role. Hagey called their messaging a case study in “what not to do”: in the current environment, “you have to flood the zone. You cannot be silent.”
Microsoft publicly rushed to Altman’s aid, but privately the weekend shook its dependence on an unstable supplier. Hagey described Mustafa Suleyman and Microsoft’s internal AI build-out as a “lifeboat”; OpenAI’s move toward a more-for-profit structure leaves Microsoft influential over compute and products.
Roose predicts a “messy breakup” as OpenAI expands from models and APIs into consumer and enterprise businesses that overlap Microsoft’s. Hagey’s less categorical conclusion was that “we are all still living in the shadow of the Blip.”
10. Tinder’s flirting bot gamified conversation without reproducing chemistry
Tinder’s limited-time US iOS experience, The Game Game, used ChatGPT to generate meet-cute scenarios, score spoken flirting, and award up to three flames. The pitch was straightforward: “Talk with AI characters and up your score using only your charm.”
Newton earned zero flames after turning a cooking date with a police officer toward police brutality, oregano, and a “very arresting” joke. Roose reached roughly two-and-a-half flames with “Are you an opioid? Because I can’t quit you,” then an anthrax request triggered minus 99 and a zero score.
Their substantive complaint was tuning: the characters compulsively asked the next question instead of revealing personality or allowing genuine back-and-forth. Hagey found ChatGPT voice mode a little better tuned, and Roose still saw value in private practice, since asking a real acquaintance for half an hour of structured flirting feedback would be embarrassing.
Roose compared the flame meter to AI coaching software that flashes instructions at call-center workers, predicting wearable systems that pulse when a dater talks too much or asks too few questions. Newton’s product critique was sharper: Tinder users need help with one-word replies and ghosting, not “shiny little novelties.”