Goodbye TikTok, Ni Hao RedNote? + A.I.'s Environmental Impact + Meta's Masculine Energy
Summary
- The Supreme Court’s decision to uphold PAFACA pushes TikTok toward a US blackout, but its future now depends on political intervention or a sale, not just the court ruling. ByteDance said Americans would lose access on January 19, although the law technically targets distribution through Apple and Google; Kevin Roose sees Donald Trump’s non-enforcement authority or a trusted-buyer deal as the remaining wildcard. A deadline rescue looked unlikely because ByteDance did not seem interested in selling, Beijing did not seem willing to authorize one, and “no one could put together a deal quickly enough.”
- Elon Musk is the only prospective buyer the hosts treated as meaningfully plausible, precisely because China may retain leverage over him. Bloomberg and The Wall Street Journal reported, based on anonymous sources, that contingency discussions appeared to be happening among Chinese officials rather than TikTok, while Musk wants X to work more like TikTok and Tesla gives Beijing leverage unavailable with a buyer such as Frank McCourt. Casey Newton could see either a real transaction or “a nothing burger,” but called the idea of Musk acquiring a second beloved platform uniquely alarming.
- The RedNote migration suggests social-media users can abandon platforms without moving to the obvious incumbents. Xiaohongshu, a non-ByteDance Chinese app commonly called RedNote, became the No. 1 free iOS app as self-described TikTok refugees chose an even more explicitly Chinese service, joking, “I would fly to China and hand my Social Security number to Xi Jinping before I would ever use Instagram Reels.” That is a warning against assuming TikTok’s engagement automatically transfers to Meta or Google.
- Meta and Google nevertheless stand to receive one of their greatest competitive gifts if TikTok disappears. India’s TikTok ban was followed by explosive growth for Instagram Reels and YouTube Shorts, and Instagram is already on many TikTok users’ phones. Kevin’s countercall is that generational identity includes rejecting older people’s platforms, leaving room for RedNote, a new US app, or protocol-based networks instead.
- The per-query environmental cost of generative AI is real but obscured by extrapolations that companies can easily dispute. Sasha Luccioni rejected a universal “10 times Google” figure or 500 milliliters of water per conversation, arguing that hardware, latency, model size, and data-center location produce ranges rather than one number. Her consumer rule is task matching: use “the smallest model” that works, avoid LLMs for calculators, consider whether conventional search is sufficient for a simple lookup, and demand that providers “give us the range.”
- The investment-scale climate issue is aggregate infrastructure demand, not guilt over one useful prompt. US data centers rose from 1.9% of annual electricity consumption in 2018 to 4.4% in 2023 and could reach 6.7%-12% by 2028; Google and Microsoft acknowledged that AI helped knock them off their own sustainability targets. Nuclear agreements, including Microsoft’s Three Mile Island recommissioning and Google’s Kairos partnership, are being presented as responses to the need for continuous power, but transmission and grid fragmentation remain constraints.
- Efficiency gains may expand AI’s total footprint rather than shrink it, while Meta’s masculinity turn raises a separate governance risk. Luccioni invoked Jevons Paradox: cheaper, more efficient inference encourages companies to put LLMs into more products, potentially overwhelming per-use savings. Meanwhile, Zuckerberg’s call for less “neutered” corporate culture landed alongside the end of Meta’s DEI program and plans to cut 5% of “low performers,” leading Casey to hear a message that “women are [not] welcome at Meta.”
Deep dive
1. The Supreme Court clears the path to a TikTok shutdown
The Court upheld PAFACA and rejected ByteDance’s challenge after surprising observers by taking the case at all. Casey expected more First Amendment deference, but the justices treated divestiture as preserving users’ speech: sell the platform and “all of the speech on the app remains.”
ByteDance said it would block US access on January 19, leaving existing apps unable to refresh or populate new content. That goes beyond the statute’s immediate mechanism, which directs Apple and Google to remove TikTok; the hosts saw the shutdown threat as a possible effort to mobilize angry users and create leverage.
Kevin still assigned some probability to Donald Trump intervening after TikTok CEO Shou Chew received an inauguration seat among VIPs. A new administration could decline enforcement or bless a trusted-buyer transaction, letting Trump present himself as having “heroically” saved an app important to young Americans.
2. A sale looks unlikely on schedule, with Musk the wild card
Frank McCourt proposed buying TikTok even without its recommendation algorithm, financing the deal through private equity and wealthy investors including Kevin O’Leary. MrBeast also said billionaires had contacted him, but both hosts treated that prospect as unserious: “I don’t think we have to think about it that hard.”
Bloomberg and The Wall Street Journal reported that senior Chinese officials had discussed Musk as a contingency, according to anonymous sources describing confidential discussions, while TikTok called the story “pure fiction.” Kevin’s reason for taking it seriously was that TikTok might genuinely be outside the conversation: Beijing, whose permission ByteDance would need to divest, could be deciding the asset’s fate.
Musk’s strategic fit is legible—he has said X should operate more like TikTok—but so is China’s potential leverage through Tesla’s operations there. Casey called the idea of turning Musk into “kind of a soft Chinese agent” far-fetched, then conceded that Musk is consistently careful and “almost never says anything remotely critical” about Beijing.
Kevin’s deadline case had three parts: ByteDance did not appear willing to sell, China did not appear willing to permit a sale, and no buyer could close by January 19. Casey added that TikTok’s reported largest market was Indonesia, so ByteDance might conclude it can “give up on America” and keep making money elsewhere.
3. RedNote converts a ban into a consumer rebellion
Xiaohongshu, commonly called RedNote, reached No. 1 among free iOS apps as Americans protested the loss of one Chinese platform by joining another. It is not owned by ByteDance, and its mechanics resemble TikTok’s For You feed, but its pre-migration content was mostly Chinese-language material about China. The app’s literal translation, “Little Red Book,” also evoked the book of Mao quotations.
Kevin’s first three recommendations captured the service’s rapid Americanization: a Modern Family clip, a Chinese-language video of a dog having its anal gland expressed, and latte art depicting Luigi Mangione. Casey’s verdict: “In just three videos, you’ve captured a shocking amount of the zeitgeist.”
Self-described refugees posted introductions, tried Chinese makeup styles, asked for followers, and admitted, “I can’t read shit on this app.” The result looked less like organized resistance than an improvised cultural exchange between bewildered Americans and established Chinese users.
Users turned privacy warnings into protest theater, enthusiastically agreeing to app tracking and promising China all their data. The companion “goodbye to my Chinese spy” meme had Chinese people—possibly from anywhere in the world—pretending to farewell their surveillance targets, sometimes advising them to call their mothers, an irreverent rejection of Washington’s security framing.
4. TikTok’s users may enrich incumbents—or route around them
Kevin saw the migration as evidence of platform fragility: TikTok wanted users to fight for it, but many simply packed up and moved. That behavior makes social feeds look “interchangeable and commoditized,” even if creators must bring their audiences elsewhere.
Casey’s own position remained conflicted. He accepts “good reasons” to restrict adversarial foreign ownership, yet hates the mechanism and fears its precedent: a future administration could target an American platform, demand new ownership, and cite the Court’s conclusion that divestiture creates no speech problem.
The incumbent bull case comes from India, where YouTube Shorts and Instagram Reels surged after TikTok’s ban. Removing Gen Z’s defining platform could let Meta extend a portfolio in which Facebook skews older and Instagram serves Millennials, while Google further entrenches itself in short-form video.
Kevin’s pushback—worth keeping—is that fleeing to an obscure Chinese-language app reveals “how badly we don’t wanna be on Instagram.” Casey countered that most TikTok users already have Instagram and will likely look there, while also noting growth in the Fediverse and protocol-based products born from frustration with billionaire-controlled apps.
5. AI’s viral climate statistics outrun the available evidence
The discussion began with a widely shared post claiming one ChatGPT search uses 10 times a Google search’s energy, while training one model emits as much carbon dioxide as 300 New York–San Francisco round trips and five times a car’s lifetime emissions. The hosts presented these as culturally powerful claims to test, not settled measurements.
Luccioni’s first principle was skepticism toward individual culpability: people operate inside systems and sometimes must use resource-intensive technology for work. Her preferred response is to “ask for accountability, ask for transparency” because caring is optional, but people need credible information if they want to act on that concern.
The 10-times comparison traces back to an old estimated Google-search footprint and a similarly extrapolated ChatGPT query using assumed hardware and latency. Luccioni argued that the useful output is a range across comparable models; chasing one exact number lets providers dismiss the entire question by saying, “That’s not the exact number.”
Smaller AI companies are often more compute-constrained and therefore do more with less—sometimes with only “100 GPUs to work with.” Their motive may be frugality rather than climate protection, but lower computation and energy use also reduce operating costs.
6. Water use is local, variable, and easy to misuse
Data centers consume water primarily to cool hardware, with efficiency commonly related to liters per kilowatt-hour. Luccioni described visiting one as “an overwhelming experience”—the noise, heat, and “buzz of electricity”—with some heated water evaporating and the remainder requiring cooling before reuse or being returned to nature.
The familiar 500-milliliter-per-conversation claim extrapolates energy measurements from an open-source model into estimated cooling demand. It is “definitely not systematically 500 milliliters,” but usage is non-negligible, and geography matters: a new data center can strain a nearby town already experiencing water shortages.
Luccioni’s pet peeve is using ChatGPT as a calculator: it is bad at arithmetic and consumes orders of magnitude more energy than a tool that needs no cooling water. Her own defensible use is asking for playful titles after writing an abstract—a bounded task where the model supplies something she finds genuinely useful.
Kevin admitted sending Raycast four or five questions daily, including low-stakes lookups such as Billy Crystal’s age, without caring enough to verify the answer. Casey argued that useful AI is tiny beside driving, flying, or eating meat; Luccioni replied that those are incomparable choices, and users should instead compare two ways of completing the same task.
7. Data-center demand is becoming a grid-level constraint
Lawrence Berkeley National Laboratory estimated that US data centers’ electricity share more than doubled from 1.9% in 2018 to 4.4% in 2023. With AI a major contributor, the report projected 6.7%-12% by 2028—a macro footprint far more consequential than any isolated query.
Big technology companies have been major buyers of renewable-energy credits and long-term power-purchase agreements, earning Luccioni’s qualified acknowledgment that they had prepared. Yet Google and Microsoft reported in 2024 that AI growth had pushed them off sustainability targets established around 2018 or 2019; their renewable-energy offsets were no longer covering demand.
Nuclear then “entered the chat”: Microsoft moved to recommission Three Mile Island, while Google partnered with Kairos. Data centers require continuous power, while wind and solar vary, so “you can’t just yeet a bunch of solar panels” and expect them to follow an intensive computing load.
Generation is only half the bottleneck. A data center is a concentrated “energy sink,” and grid operators must physically deliver additional megawatt-hours to whichever rural site hosts it; America’s patchwork of providers and differently sized grids makes modernization a collection of difficult local transmission problems.
8. Better chips do not guarantee lower total consumption
Luccioni rejected the umbrella claim that AI’s climate benefits will outweigh its costs as a false dichotomy. The most energy-intensive systems—large language models—“have yet to prove their utility in fighting climate change,” while models already helping with climate change are not driving most of the sector’s energy and carbon burden.
Algorithmic improvements, distillation, and better chips can reduce energy per task, but Jevons Paradox threatens the aggregate result. As nineteenth-century coal use and fuel-efficient driving illustrated, greater efficiency lowers effective cost and encourages more consumption: “We’re gonna put AI into even more things.”
The hosts’ macro summary was therefore cautious but clear: data centers strain grids, their number is increasing, and expanding usage may absorb inference savings. Luccioni agreed that the environmental claim must be taken seriously even while individual prompts become cheaper.
9. Transparency is the actionable point of agreement
Kevin’s micro synthesis matched Luccioni’s view: individual LLM use may not move the climate needle, but users should choose the smallest adequate model and reserve generative AI for tasks where its advantage matters. People need not “tear their hair out” over a genuinely valuable interaction.
Luccioni emphasized consumer power: providers possess average-energy figures or at least defensible ranges, so users should say, “Stop bullshitting us.” Adding AI to familiar footprint comparisons would support informed choices instead of “feeding them shit and keeping them in the dark.”
Luccioni also proposed user controls such as a toggle for Google’s generated summaries, letting people retain a core service while choosing a mode coherent with the values or resource constraints they want to optimize.
10. Meta’s masculine makeover masks a governance argument
Mark Zuckerberg told Joe Rogan that culture had overcorrected against masculinity and produced a “somewhat more neutered” corporate environment; martial arts helped him recognize the absence. Kevin placed the interview alongside Meta’s efforts to become more palatable to the political right and the incoming Trump administration.
Casey answered with satire: make the Like button display a bulging vein and grunt, turn conference rooms into octagons, rename finance “Mixed Martial Accounting,” and give 4chan content moderation. Other proposals included cooling data centers with Mountain Dew Code Red and handing women’s hackathon ideas to male executives—“what kind of energy is more masculine than taking credit for a woman’s idea?”
Beneath the jokes, Casey connected Zuckerberg’s rhetoric to killing off Meta’s DEI program and cutting 5% of employees labeled low performers. With Meta employing roughly two men for every woman, Casey’s concern was not masculinity itself but the institutional message: “The message is not, uh, women are welcome at Meta.”
Kevin realized that Silicon Valley’s earlier “founder mode” campaign was similarly gendered: its prominent advocates were men asserting that “we’re big boys” entitled to run companies accordingly. He contrasted Zuckerberg’s performance with Jeff Bezos discussing fear and vulnerability at work—emotional openness that looked more secure than changing one’s values “wholesale almost overnight” while controlling platforms used by billions.