A Chinese‑born AI start‑up backed by Silicon Valley heavyweight Benchmark quits China amid U.S. scrutiny—triggering criticism in Beijing and a policy test in Washington.

In an era when artificial intelligence is both an industrial project and a geopolitical contest, few deals have crystallized the fault lines like the one between Manus and Benchmark. Manus—the buzzy agent‑style AI start‑up that some investors breathlessly touted as “the next DeepSeek”—has been forced into damage control on two fronts. In Beijing, the company was pilloried online for “abandoning” China after it moved to strike its name from local social media, laid off staff in the capital and shifted its corporate center to Singapore. In Washington, its Silicon Valley backer, Benchmark, has faced questions over whether a $75 million investment ran afoul of America’s new outbound‑investment regime for Chinese AI. Caught in the middle, Manus says the decision was commercial, not political: overseas users, it argues, offer the most sustainable path to profit.

The uproar underscores how sharply the regulatory environment has turned since the brief honeymoon for cross‑border AI builders in 2023–24. The U.S. Treasury is reviewing the Benchmark‑led funding under rules designed to stop American capital and “intangible benefits” from turbocharging advanced AI in countries of concern. Those measures—sometimes called reverse CFIUS—are still being defined in practice, but they already cast a long shadow over venture deals. For investors, the message is simple: exposure to Chinese AI is no longer just a portfolio call; it is a national‑security decision, subject to legal risk and public controversy.

On the Chinese side, the blowback has been just as fierce. Nationalist influencers and some state‑linked outlets framed Manus’s relocation as a betrayal timed to cash a U.S. check. The company’s rapid erasure from domestic platforms—where its video demos once attracted huge followings—was interpreted as proof that its future lies outside the mainland. That reading ignores the business reality: Manus’s product is a general‑purpose AI agent whose most engaged user base has been overseas, and whose reliance on foreign foundation models, including Anthropic’s Claude and Alibaba’s Qwen in various configurations, complicates pure‑China operations. Yet the politics of “decoupling” rarely leave space for nuance.

Benchmark’s calculus was more straightforward. The storied venture firm, known for early bets on Uber, Instagram and Snap, saw an opportunity to back a team with viral product‑market traction, a fast shipping cadence and an opening in the emergent market for agentic AI assistants that execute tasks end‑to‑end. The investment valued Manus’s parent, commonly referred to in filings as Butterfly Effect, at just under half a billion dollars—lofty for a pre‑revenue company but not outlandish in today’s AI land‑rush. The firm argues that censoring American capital will only push promising founders to reorganize abroad without improving security outcomes.

To understand why the reaction has been so intense, consider the DeepSeek effect. The Chinese model’s surprise leap in price‑performance this spring demolished assumptions about where the cutting edge must live, and it revived Silicon Valley’s interest in scouting China’s AI scene. Manus benefitted from that halo. Its demos—automating résumé screening, building websites, even basic research tasks—fed a narrative that Chinese‑born teams could iterate faster on practical agent products than well‑funded U.S. rivals. That narrative, in turn, collided with a bipartisan American push to cordon off capital and know‑how and with Beijing’s own tightening around data and model governance.

The facts of the case are not in serious dispute. Manus shut down its China‑facing app presence, announced a shift of key functions to Singapore and began recruiting outside China. Benchmark confirmed it led a $75 million round. U.S. officials said the transaction was under review; critics in Congress called for a hard line, citing the risk that U.S. investors could indirectly subsidize AI talent pipelines that serve the Chinese state. Beijing’s criticism focused on loyalty and “national feelings,” with commentators branding the founders “defectors.” Between those poles lies a narrow band of corporate pragmatism: Manus aims to sell to global customers, and moving may smooth compliance and hiring, even as it invites a storm at home.

For founders, the episode is a case study in the new playbook for “de‑Chinafication.” First, establish an offshore holding company with governance and data policies aligned to Western buyers; second, wind down public‑facing operations in China to limit regulatory and political risk; third, court capital that can survive regulator scrutiny, preferably from investors with strong compliance track records. Each step carries costs. Talent churn is significant, domestic brand value is sacrificed, and regulators may still demand guarantees on source code access, model weights and customer data handling.

For investors, the lessons rhyme with previous technology flashpoints. Due diligence now reaches beyond cap tables and IP to geopolitics: where are the model weights hosted; who controls access; can the company deliver clean separation between any China‑based R&D and commercial products abroad? Deal structures are being redesigned with contingency in mind: springing governance terms if rules tighten; capital tranches tied to relocation milestones; and “kill switches” for knowledge transfer. Some limited partners, especially public institutions, are advising managers to avoid any Chinese nexus on frontier‑AI deals, period.

Regulators are also learning in real time. Washington’s outbound‑investment review is still a pilot. Treasury will have to decide what constitutes a prohibited “covered transaction” versus an allowed passive stake with guardrails. Beijing, for its part, has been championing homegrown champions—DeepSeek most of all—while signaling that firms who look westward at sensitive moments risk public shaming or worse. Neither side appears eager to draw bright lines that would give start‑ups certainty. The result is a gray zone where headlines move faster than the rulemaking.

What happens next for Manus will depend on execution. Reincorporating in Singapore offers friendlier business infrastructure and a recruiting base that is both international and familiar with Chinese‑language markets. But the company must prove staying power in a crowded agent field that now includes Big Tech incumbents and a swarm of venture‑backed competitors. Its reported reliance on outside foundation models lowers capex but raises platform risk. And the longer the Treasury review lingers, the harder it becomes to hire senior talent or sign enterprise accounts wary of future sanctions.

The bigger question is whether this saga chills cross‑border AI investment or simply channels it into new shapes. If Benchmark’s stake is deemed compliant—with conditions around information flows and governance—other U.S. funds may quietly follow, backing founders who have de‑risked their China footprint. If it is blocked or heavily penalized, the message will be decisive: the cost of touching Chinese‑origin AI outweighs the upside, however global the company’s ambitions. Either way, the Manus affair clarifies an uncomfortable truth for entrepreneurs on both sides of the Pacific: in 2025, building category‑defining AI is as much about regulatory diplomacy as it is about model quality and product velocity.

For now, Manus is neither pariah nor savior—just a young company trying to outrun politics long enough to build. It may yet justify the “next DeepSeek” hype, or it may become another cautionary tale about geopolitics swallowing the start‑up dream. The lesson for the rest of the ecosystem is stark: in AI’s hottest markets, you don’t just raise capital; you raise flags, and you should know which ones are flying over your cap table before the outrage cycle begins.

Sources

• Financial Times, “Manus and Benchmark: The AI deal that upset China and the US,” Aug. 11, 2025.
• Semafor, “US Treasury examining Benchmark Capital’s ties to Chinese startup Manus AI,” May 9, 2025.
• Bloomberg, “The New Third Rail in Silicon Valley: Investing in Chinese AI,” July 14, 2025.
• Reuters, “Beijing boosts AI startup Manus as China looks for the next DeepSeek,” Mar. 21, 2025.
• Business Insider, “China’s newest AI agent Manus draws DeepSeek comparisons,” Mar. 2025.