Vera Rubin Debuts, AI Deployment Drops to 14 Days, and OpenAI Eyes a $1.4T Valuation
The 14-day AI factory model changes procurement and deployment math for GPU-heavy workloads, and the OpenAI-Anthropic financing race will set the valuation benchmark for every frontier lab. China's all-domestic cloud-plus-edge stack demonstrates a non-Nvidia path to trillion-parameter training, with direct implications for export-control debates.
AI news on October 22 split between two flagship events. In Berlin, Jensen Huang opened Nvidia's GTC with Vera Rubin, a fully liquid-cooled system that links 72 GPUs per rack into a supercomputer and is expected to deliver a 30x performance gain within two years. Nvidia also presented the DSX POD, a reference AI factory built by partners inside modular containers and integration-tested before delivery, which cuts on-site deployment from six months to 14 days. Huang told Europe to become an AI producer rather than a consumer and put a humanoid robot platform around 2027.
At iFlytek's 1024 Developer Festival in Hefei and Beijing, iFlytek launched the Spark X2.6 large model on all-domestic compute and an all-domestic cloud-plus-edge architecture. The approach keeps more tasks on local devices and sends complex jobs through edge-cloud collaboration. The accompanying exhibition showed the Feixing platform, built with Huawei on the Ascend ecosystem and billed as China's first domestic 10,000-card compute base for trillion-parameter training.
Capital and policy ran in parallel. OpenAI is advancing a roughly $30B round at a valuation above $1.4T, disclosing $16.8B in Q3 revenue and $50B annualized revenue, and wants commitments locked before Anthropic's expected mid-November IPO. Tenstorrent raised $1.5B at a $6.2B valuation. China's central party and state general offices published 17 measures on integrating technology and industrial innovation, and conferences in Nanjing, Chongqing, and Jinan released data initiatives and an AI high-value scenario map. Market worries about AI data-center capital spending and returns were also part of the week.
The 14-day deployment claim for DSX POD rests on shifting integration work into a partner's factory before the container ships; the six months that used to happen on site are gone from the critical path.
OpenAI's disclosed revenue numbers serve the fundraising pitch as much as any earnings report; they give investors a concrete anchor before Anthropic's IPO sets a public market valuation.
iFlytek's all-domestic cloud-plus-edge stack and Feixing platform turn export-control constraints into a product feature: Chinese labs are now marketing the absence of foreign chips as an advantage.
The same news cycle contained both record infrastructure launches and market questions about data-center capex returns, so expansion and doubt now move together rather than in sequence.