← Back to stories Detailed view of organized electronic circuit boards in a production setting.
Photo by Andrey Matveev on Pexels
凤凰科技 2026-04-19

Tech giants are snapping up all the chips

Fundraising rumours and a delayed flagship

It has been reported that DeepSeek (深度求索) has quietly kicked off its first external financing round since founding, reportedly seeking a valuation of at least $10 billion and aiming to raise no less than $300 million. The Financial Press first circulated the claims after The Information’s coverage; Reuters and other global outlets followed but Reuters said it could not independently verify the story. DeepSeek has not issued an official response, leaving investors and the market to parse leaks and code hints on GitHub about an awaited next‑generation model — a release that has been rumoured, postponed and repeatedly denied in recent months.

From training cost miracle to multi‑front competition

One reason DeepSeek remains centre stage is historical: the company’s R1 model shocked the market last year by claiming extremely low training costs — a “cost‑deflation” story that allegedly undercut rivals and briefly rattled chip stocks such as NVIDIA. But the AI landscape has moved fast. It has been reported that a number of former DeepSeek researchers, including prominent agent developers, have migrated to ByteDance (字节跳动), underscoring how talent and priorities have shifted toward multi‑modal models, video and agentic systems. In short: raw SOTA text benchmarks are no longer the whole game. Investors now ask a different question — can DeepSeek translate back‑end training thrift into affordable, high‑frequency inference (token) economics?

Chips, cloud price rises and the token battleground

That question matters because hardware and cloud economics are tightening. Amid U.S. export controls and broader trade frictions that have constrained supply of top‑end accelerators, it has been reported that H100 rental rates climbed from about $1.70/hour/GPU in October 2025 to $2.35 in March 2026, and Chinese cloud providers — Alibaba (阿里巴巴), Baidu (百度) and Tencent (腾讯) — have applied double‑digit price adjustments to AI compute. Alibaba has even set up an Alibaba Token Hub (ATH) under CEO Wu Yongming (吴泳铭) to manage token creation and distribution, signalling that token consumption — not just chip count — will shape winners. So can DeepSeek rekindle the industry’s appetite by delivering a new round of “token deflation”? If it can, investors may look past delayed model launches; if it can’t, the market’s hunger will be satisfied elsewhere.

AISemiconductorsE-Commerce
View original source →