Episode in Brief
Xi heads to Washington. Xi Jinping is set for his first US state visit since 2015, capping his busiest travel stretch since 2019. Expectations for the Trump-Xi meeting itself are low — stabilization, not a reset — though a Chinese business delegation is plausible and worth watching for who is (and isn’t) on it.
AI drama goes long-form, regulation follows. Mango TV has released what it calls China’s first full-length AI-generated drama, a 30-episode sequel made in five months instead of a year-plus.
A labor dispute goes global. Auto-lighting supplier Xingyu rescinded roles for about 100 new graduates; when a domestic complaint stalled, the graduates took their case straight to Xingyu’s European automaker clients, triggering investigations at Volkswagen and Mercedes-Benz.
Moonshot moves toward an IPO. Kimi-maker Moonshot is reportedly preparing a dual Shanghai/Hong Kong listing. Robert discusses how China’s frontier AI labs are unusually collaborative with each other because they see the competition as global, not domestic.
Anthropic disclosure meets Moonshot dispute. A new Anthropic threat report details a China-based dating-app scam network built on Claude, landing alongside an unresolved dispute over Moonshot allegedly rerouting Kimi queries to Claude.
Xi’s State Visit to Washington
Low Stakes, High Symbolism
This has been Xi Jinping’s busiest travel stretch since 2019: stops in Kazakhstan and Egypt and at the BRICS summit in New Delhi, with a first state visit to Washington since 2015 still to come later this month. Robert Wu is skeptical that the Trump-Xi meeting itself will produce much: “the expectation is not very high.” He argues its main function is simply to keep principal-to-principal contact happening, not to reset the relationship. Unlike last year’s meeting in South Korea, which came amid an active trade dispute and drew intense speculation, this year’s meeting draws relatively little domestic Chinese attention — Wu says even policy circles are only lightly engaged, because China feels it has “found its own pace” in dealing with Washington. He does not rule out a narrow trade agreement or selective easing of controls, but cautions against reading Xi’s other stops as building toward the US visit; Chinese leaders typically bundle overseas trips together for logistical reasons rather than strategic sequencing, citing Xi’s 2013 stops in Mexico and Costa Rica ahead of a US visit as precedent.
Will Xi bring business leaders?
One open question is whether Xi will bring a delegation of Chinese executives, something he has mostly avoided since around 2020. Reuters and SCMP reporting suggests he might this time, and Wu thinks it plausible given that Trump’s own May visit to Beijing included Elon Musk, Tim Cook, and Jensen Huang. Nothing is confirmed. Wu specifically hopes to see China’s frontier AI founders included, though he calls this “very unlikely,” and flags that ByteDance founder Zhang Yiming was a conspicuous absence from the equivalent delegation during Trump’s Beijing visit despite ByteDance’s high stake in the relationship. Looming over the visit: Huawei’s criminal trial has just opened in New York, and Shenzhen hosts APEC in November — China’s largest hosting moment in over a decade.
“I think right now it’s just good that the leaders of two countries get to meet every now and then in person and have this principal-to-principal communication. This really can help stabilize the relationship.”
— Robert Wu
China’s First Full-Length AI Drama Lands
Appetite for more than micro
Mango TV has released what it calls China’s first full-length AI-generated drama: a 30-episode sequel to Journey to the West, produced in five months versus the year-plus a conventional production would take. Mango’s stock rose in the days after release. Olivia frames this against the prior episode’s discussion of Niu Lai, an unpolished, human-made viral film that succeeded partly because audiences craved something authentic — raising the question of whether audiences now want the opposite.
Robert doesn’t see AI and human content as being in conflict. He compares the current spread of AI-generated micro-dramas — now visible everywhere on China’s subways and in elevators — to the early mass-adoption phase of short video in China. He doesn’t think the government opposes the trend, reading it instead as an emerging industry Beijing is willing to let develop even as it disrupts incumbents: “many small-time actors and actresses lose a lot of their jobs” as some production sites lose work. His forecast is a split market: a large, cheap, always-available content tier that AI fully occupies, alongside formats built on the audience’s ability to physically meet or interact with real performers — concerts, tours, fan events — that AI cannot replace. He expects a middle category of AI-enhanced human performance to grow too, describing a friend’s account of a human-acted, choice-driven interactive drama filmed like a branching game, where the actors resisted being replaced by AI specifically because they wanted in-person contact with fans; he speculates AI could eventually make branching-script productions like this commercially viable by handling the otherwise unmanageable number of narrative paths.
Olivia connects this to a physical-world parallel: a reconstructed palace complex she visited in Ordos, Inner Mongolia (formerly nicknamed a “ghost city”), built at large expense purely as a filming location — the kind of capital investment that could increasingly be replaced by virtual, AI-generated sets, with data centers taking the place of studio backlots.
“There will be a mass market, really cheap, you-can-access-anytime type of content, and that will be completely occupied by AI. But then there’s also some work, some projects where humans are actively involved.”
— Robert Wu
A Labor Dispute at Xingyu Goes International
Don’t test these recent grads
Xingyu, one of China’s largest automotive lighting suppliers and a vendor to global automakers, hired roughly 400 new university graduates and then reportedly moved to terminate about 100 of them — offering the option to resign voluntarily or accept lower-paid assembly-line roles. A domestic complaint to China’s labor bureau reportedly went nowhere, so the affected graduates took their case directly to Xingyu’s European clients. Volkswagen Group China has opened a special investigation; Mercedes-Benz has confirmed it referred the matter to a specialist team. Xingyu has apologized, suspended its HR director, and offered a compensation package including a job-search subsidy and accommodation support. The company is already listed on the Shanghai Stock Exchange and is reportedly planning to add a Hong Kong listing.
Robert calls the underlying dynamic simply sad: trust between Chinese employers and employees is low right now, and he says it runs in both directions — he separately describes a friend, also a founder, whose signed offers to two strong candidates were declined at the last minute, partially because the candidates didn’t trust the company to honor them. He’s unambiguous that Xingyu’s conduct was wrong (“clearly not ethical, not a professional thing to do”), but he’s struck by how effectively the affected graduates organized: escalating to the company’s biggest overseas clients was, in his words, “really smart,” and the incident became national news specifically because of how well-organized the response was.
The bigger pattern, per Robert, is that Chinese companies now selling into global supply chains carry global reputational exposure that most of them have not internalized.
“You can see that young people today are very innovative and creative in how they seek their own rights. Like this going-to-the-EU thing — it’s really smart, I have to admit.”
— Robert Wu
Moonshot’s IPO and a Rare Truce Among China’s AI Labs
Competition looks different these days
Moonshot AI, maker of the Kimi model family, is reportedly preparing a dual Shanghai and Hong Kong listing. The episode cites Moonshot running at roughly $1 billion in annualized revenue currently, targeting a $2 billion annualized run rate by year-end (figures as stated in the episode, not independently verified here) — still small next to the revenue levels reported by OpenAI and Anthropic, but Moonshot’s pitch, per Olivia, is growth and the durability of Kimi K3 as a commercial platform. Robert reads the implied valuation as roughly in line with China’s other already-listed frontier model companies, Zhipu and MiniMax, which he says now function as a market benchmark for what a Chinese frontier lab is worth, at “only a fraction” of US peers’ valuations. He notes one of those peers announced a large secondary fundraise just last week even with its share price underperforming — in his reading, every Chinese lab is racing to build a capital war chest now, in case “the AI bubble bursts in a few months or a few years.”
Asked to compare competitive intensity with the US, where Robert calls the frontier-lab rivalry “almost petty,” he makes an unexpected argument: China’s AI labs are, if anything, less rivalrous with one another than in previous domestic “tech wars” (food delivery, bike-sharing, and earlier super-app rivalries such as Alibaba and Tencent blocking each other’s links). His explanation is that AI is the first Chinese tech sector born global rather than domestic, so Chinese labs see Anthropic and OpenAI as the competition, not each other. He describes talent and even informal support flowing between Chinese labs and their founders’ former employers in a way that would have been unthinkable during the super-app era, crediting this partly to the founders’ generation (millennials and Gen Z, globally oriented from day one) and partly to the industry’s global-facing origin.
On Moonshot’s business model, Robert points to a licensing arrangement attached to its open-weight releases: large commercial users monetizing above a certain threshold are asked to pay Moonshot a licensing fee, and he believes compliance is improving because bigger companies increasingly care about their reputation among researchers and peers. Some cloud providers reportedly want to support the arrangement directly because stronger underlying models improve their own businesses — in his framing, an ecosystem-building logic rather than a purely transactional one.
“This is probably the first ever industry in China where, when it was first born, it’s a global competition.”
— Robert Wu
Anthropic’s Scam Report
Time to get off the apps
Anthropic’s newest threat-intelligence report includes a case the hosts single out: a China-based app studio accused of using Claude to build and operate more than 20 dating apps, all marketed as staffed entirely by real people. The report describes nearly 5,000 distinct AI personas that messaged roughly 25,000 real users over a two-week window in April, with real gig workers mixed in at roughly a 3-to-1 AI-to-human ratio to cover tasks the AI could not yet handle convincingly, such as live video calls used to pass authenticity checks. Anthropic says it banned the associated accounts and worked with other AI labs to shut the network down. These figures are as reported in Anthropic’s own disclosure; the hosts do not independently verify them.
Robert connects this to China’s broader, worsening scam landscape, illustrating the government’s aggressive countermeasures with a personal anecdote: two of his personal investors, who are neighbors, had their bank accounts automatically frozen for several hours after each sent him a similar amount of money the same day — an anti-scam algorithm flag he had to spend time getting unwound. He argues telecom and romance scams (China’s “pig butchering” schemes) are especially hard to stop because they exploit psychology rather than force, and because the more sophisticated operations are cross-border — he cites the earlier Myanmar scam-compound cases as the precedent that first forced international enforcement cooperation. He doesn’t expect the trend to fade and thinks user education, not enforcement alone, is the realistic response, since scam operations can now be produced at a scale no police force can individually chase down.
On Anthropic itself, Robert is candid about being conflicted: disclosing client usage behavior is something he feels Anthropic “maybe shouldn’t disclose,” even though making a genuine scam network public plausibly serves the public good. He links this report to a separate, earlier Anthropic accusation that Moonshot secretly routed some Kimi user queries to Claude and presented Claude’s answers as if Kimi had generated them — by Anthropic’s own count, a large number of chats, though nothing close to Kimi’s total user base. Robert’s read is that this is not equivalent to stealing a rival’s underlying model or algorithm — “you can only send the questions and get the answers” — and is better understood as a shortcut to save on data costs than outright theft. But he thinks Moonshot should have anticipated the exposure and reputational cost: “they should have known that Anthropic would use this as a means of attack.” Given the timing — Anthropic’s scam disclosure and its routing accusation both landing as Moonshot pursues its IPO — Robert suspects deliberate strategy on Anthropic’s part, describing what he sees as a form of commercial warfare. He separately mentions an unconfirmed claim circulating that someone possibly linked to the PLA uploaded sensitive material into Kimi without realizing it could be rerouted to Claude — he is explicit that he does not know the details and cannot confirm this account, so it should be read as unverified.
“It does feel to me that this is very strategic. That there is some fierce commercial warfare.”
— Robert Wu
