Fluidstack triples to an $18B valuation in six weeks on Jane Street's bet
Two AI stories worth a second look today — a neocloud jumps 2.4× in valuation between rounds, and Moonshot quietly files for Hong Kong.
Neocloud Fluidstack closed a $1.5 billion round led by Jane Street at an $18 billion-plus valuation, according to people familiar with the matter — barely six weeks after announcing a $750 million round at $7.5 billion in July. The pace is the story: Fluidstack is a UK-built, GPU-heavy hyperscaler that rents compute to AI labs, and the speed of the markup suggests someone on the buy side (Jane Street, in this case) believes the company's pipeline has shifted from "neocloud of the moment" to "infra-tier with staying power." The Jane Street lead is unusual for an AI infra round — most neocloud checks are written by Nvidia, sovereign-wealth funds, or the big private-equity shops; a market-maker fronting the cap table implies Fluidstack's GPUs are doing real work for quant desks, not just training runs for foundation labs. Watch the next round: at this trajectory, Fluidstack will be priced for an IPO within twelve months, and the order book will tell us whether hyperscaler demand has really outgrown CoreWeave-era capacity, or whether the market is just front-running another Meta-scale training buildout.
Moonshot quietly filed its A1 application with the Hong Kong stock exchange this week, formally starting the Kimi IPO process. The company declined to comment beyond "no information to disclose," but media reports peg the pre-money valuation at around $50 billion — almost exactly the figure Moonshot was reported to be raising at in July, when its $3.5 billion-plus F round closed oversubscribed and pushed the post-money to $35 billion. A confidential A1 means regulators are now looking at the deal, and a public prospectus is usually six to ten weeks behind. The interesting question isn't whether Moonshot lists — it's who anchors the offering. If the IPO draws US and Middle Eastern sovereigns alongside mainland allocators, it'll confirm the Chinese frontier-model market has a real bid outside the People's Bank ecosystem; if not, it'll be a window into how much Beijing wants to control the flow. We covered the earlier IPO-plans chatter in Moonshot AI races toward a Hong Kong IPO as the next big Chinese LLM lab.
Alibaba pushed a post-trained Qwen3.8-Max update that now sits at the top of CodeArena's WebDev leaderboard with a score of 1691, ahead of Claude Opus 5 and Kimi K3, and on the Pareto-front pricing chart it undercuts every model above $5 per million tokens — it averages roughly $5. The update is specialized (coding and "Cowork" office work), and a 22-point jump on WebDev is meaningful, but it's still a slice of the benchmark suite. The bigger story is that the Chinese labs are no longer competing on price alone: Qwen3.8-Max is now a credible frontier contender on at least one dimension where OpenAI and Claude were untouchable a few cycles ago. Our Deep Dive on Alibaba's Qwen3.8 2.4T open weights walked through why the post-training pipeline has become the actual frontier — today's update is the first concrete payoff.
What to watch: whether Fluidstack's next round is structured for a fast IPO, and how Moonshot's prospectus prices Kimi K3's enterprise traction versus DeepSeek's leaked books.
Is Fluidstack's triple-in-six-weeks valuation proof that neocloud demand is permanent, or just a market-maker pricing a trade? Tell us in the comments.
Sources: Techmeme — Fluidstack closed $1.5B round led by Jane Street · Iain Martin on Forbes · Leiphone (雷峰网) — 月之暗面启动港股 IPO · 晚点 LatePost reporting via IT之家 · 新浪科技 — Qwen3.8-Max tops CodeArena
