AI in Shanghai's SMBs: What We Actually See on the Ground

Start with the industry-side numbers. Xinhua reported in late April 2026 that Shanghai's 394 above-threshold AI enterprises — those large enough to enter official industry statistics — posted combined output of more than 637 billion yuan in 2025, up 39.5% year on year. A separate Xinhua report in May 2026 covered Moshu Space, China's first innovation community dedicated to large-model startups, opened in September 2023 on Xuhui's West Bund: its resident companies grew from just over a hundred in 2024 to more than three hundred, and in April 2026 it entered the Ministry of Industry and Information Technology's first batch of excellence-grade incubators. Citywide, 169 large models have cleared regulatory filing, and intelligent computing capacity has passed 160,000 petaflops.

So the industry side runs hot. But if you own a company of a few dozen people in this city, you probably feel something else: the Shanghai AI in the news and the daily life of your company look like two different worlds. That temperature gap is the subject here — what AI adoption actually looks like right now among the Shanghai SMBs we deal with. To be clear: these are qualitative observations, not a survey, and we are not going to dress them up with invented percentages.

State one: most companies are stuck at scattered individual use

Employees use AI; the company does not — by far the most common picture. A salesperson polishes follow-ups, an operations hire drafts copy, each on their own tool and their own intuition. Good usage stays in individual hands, uncirculated and unaccumulated. The boss knows "everyone is using it" but cannot say in which steps of which process, or to what effect. Strictly speaking, this is not a company applying AI; it is a company that happens to employ some individuals who use AI.

State two: content production and support replies get through first

Among the companies we see, the scenarios that actually cross from personal use into process use cluster in two areas. Content production — product listings, official-account posts, short-video scripts — where inputs and outputs are clear, mistakes are cheap and volume pressure is constant. And customer-facing replies — high-frequency, repetitive, backed by existing material. Manufacturers, cross-border sellers and local service businesses differ in almost everything else, yet converge on these first scenarios. For a company not yet started, that regularity is a ready reference — see Which Scenario Should Your First AI Workflow Target?.

State three: owners' hesitations concentrate on two things

Talk to decision-makers and the concerns are remarkably uniform. Data security: where does customer and pricing information go once pasted in? And return on spend: how do we verify the money was worth it? Both concerns are legitimate and deserve serious answers. But the response we observe most often is a third thing: unable to settle the two questions, the company simply waits. Concerns that should lead to boundaries, small stakes and acceptance criteria lead instead to standing still — arguably the largest hidden loss going right now.

State four: the missing piece is not a tool but an owner

Tools have never been the shortage. What is missing is the person who turns tools into process: who picks the scenario, sets the data rules, builds the role templates, and keeps usage alive? Big companies have digital departments; in an SMB it is usually nobody's actual job, so everything stays at the level of personal initiative. The few SMBs we have seen get real process traction share almost exactly one trait: someone is explicitly in charge — not necessarily senior, but given time and authority by the boss.

Three practical moves for a Shanghai SMB

Start from a high-frequency small scenario, not from a grand platform. Pick a step that happens every day and can tolerate error, get a visible result, then talk about expansion.

Use the local ecosystem — the support here is unusually concrete. Municipal policy has moved from the "Mosu Shencheng" AI implementation plan issued in late 2024 (roughly, "shaping Shanghai with models") to the March 2025 service-sector measures that established compute vouchers, model vouchers and corpus vouchers, with communities like Moshu Space adding matchmaking resources on top. Policy details keep evolving, so have whoever owns this check official channels periodically — but trial-and-error cost in this city can be kept genuinely low.

Count the labour hours before sizing the spend. Work out what the candidate scenario costs today in human time; whether an AI option is worth it, and at what scale, then has a reference point — see how SMBs should plan their AI budget.

The gap is headroom, not distance

Put the hot industry side and the quiet company side together, and the conclusion is not pessimistic: the infrastructure — models, computing, tools, policy support — is now cheap and good enough, which was not true two years ago. For most Shanghai SMBs, what is missing is not another year of technology maturing but a few internal steps: an owner, clear boundaries, one small scenario to start. Industry heat does not convert itself into your company's productivity; that last stretch you walk yourself.