Different Capital (不同资本) VP: AI Hardware Expedition in the Era of Large Models — Will Software Fees Become the Core of Future Growth?

Lead: shifting center of gravity from chips to fees

In an interview with Huxiu, the vice president of Different Capital (不同资本) framed the current phase of China’s AI boom as an “expedition” into hardware driven by large models — a costly, infrastructure‑heavy stage that may soon give way to software‑first monetization. He argued that as model training and inference scale up, the real durable margins could migrate from selling chips and machines to charging for model access, customization and ongoing software services. Questions for investors follow: do you back silicon or subscriptions?

Hardware realities and the economics of large models

The VP laid out the economics succinctly: GPUs and other accelerators remain the bottleneck for many Chinese developers, and cloud compute bills now dominate project budgets as model sizes climb. Leading Chinese cloud and AI builders — including Baidu (百度), Alibaba (阿里巴巴) and Tencent (腾讯) — are racing to domesticate large‑model deployments, but it is expensive work. It has been reported that firms facing constrained GPU supply are increasingly focused on inference optimization, compression and middleware — areas that naturally lend themselves to recurring software fees rather than one‑time hardware sales.

Investment implications: where returns may lie

From a venture perspective, Different Capital’s VP signalled a clear preference: invest in software layers that capture long‑term value — model orchestration, verticalized LLMs, fine‑tuning platforms and billing systems — rather than in commoditized hardware distribution. Will enterprises pay steady fees for domain‑tuned models and SLAs? The interview suggested yes, particularly for regulated sectors (finance, healthcare) where compliance and reliability matter more than raw GPU access.

Geopolitics and the long view

Geopolitical dynamics complicate the picture. US export controls and ongoing trade tensions have reportedly accelerated China’s push for chip self‑reliance, but domestic alternatives remain a multi‑year project. That reality may actually strengthen the software fee thesis: if hardware supply is strategic and tight, software that squeezes more value from every cycle becomes more lucrative. So which wins out — silicon independence or subscription economics? For Different Capital, the smart money is on fees as the sustainable core of future growth.