What This Is
TechCrunch published an in-depth report this week. Their reporters spent a week in Beijing and Shanghai, interviewing founders and engineers at more than a dozen leading Chinese AI startups. This is not product news. It is an observational note on "what China's AI scene is thinking right now."
What matters is the report's restrained lens: it neither proclaims "Chinese AI has overtaken the US" nor dismisses the field as "cheap and worthless." It captures a middle state: pragmatically building models while anxiously hunting for capital.
Industry View
The recurring judgment: under compute constraints, Chinese teams have carved out a path of "doing more with fewer GPUs, building models that actually work." DeepSeek's training costs, Zhipu's open-source strategy, and the rock-bottom inference pricing from several firms are all products of this logic. Most engineers quoted acknowledge this "pragmatism born of pressure."
But there are dissenting views. Several investors caution: if "low-compute efficiency" becomes a long-term competitive moat rather than a bet on the next architecture, Chinese AI companies risk falling behind in the paradigm shift of 2026-2027. In short, today's cost edge may become tomorrow's liability.
Commercialization is the recurring awkward topic. Multiple CEOs concede: the model itself is not the business; what sells is workflows embedded into business operations. And converting models into ARR (annual recurring revenue) is something Chinese AI companies are collectively still catching up on.
Impact on Regular People
For enterprise IT: open-source Chinese models have improved in quality, with deployment costs at about one-third of last year's levels. Self-built AI customer service and internal knowledge bases have moved from "pilot project" to "deploy now."
For individual careers: AI-related roles are growing, but role definitions are getting fuzzier. An emerging title is "AI Workflow Designer"—not the person who writes prompts, but the one who slots existing models into concrete business processes.
For consumer markets: models are cheaper, and the price war has begun. What regular users will feel: more free AI tools, but fewer that can sustain value over time—products that can't afford to burn cash will disappear first.