China is quietly rebuilding the world's largest surveillance network — and the upgrade has less to do with adding cameras than with giving the existing ones a brain. According to an analysis of more than a dozen government procurement documents by the Financial Times, local Chinese authorities are buying a new generation of AI-enabled cameras and software that can interpret behavior in real time. Instead of merely recording, the new systems flag erratic driving, growing crowds, unauthorized entry, and even people leaning over bridges in ways that suggest suicidal intent. Beijing is pushing police forces toward what officials call predictive policing: acting on patterns before incidents occur, not after.
The overhaul is significant because the previous system, built roughly a decade ago, had become outdated. It was reactive — designed to identify specific people on a watchlist by sending footage to centralized data centers — and it struggled to interpret the intent of anyone not already being monitored. Minxin Pei, an expert on Chinese governance at Claremont McKenna College, told the Financial Times that the old network was poor at understanding what people were planning. The new generation, by contrast, runs computer vision and large language models directly on the cameras themselves, allowing officers to search video by typing prompts such as 'a woman wearing a red hat' rather than scrubbing through hours of footage.
The scale of deployment is uneven but expanding. A procurement document from Yaodu, a town in Sichuan, allocated about Rmb900,000 (roughly $132,455) for 175 high-definition cameras with an 'intelligent video analysis system.' A separate tender in Datong called for cameras that could identify a person's gender, posture, and clothing. Hikvision and Huawei, the two Chinese firms leading the upgrade, are producing devices powered by semiconductors capable enough to run AI workloads locally — a shift that makes the systems more responsive while also reducing dependence on cloud servers and trimming computing costs.
But here's the catch: this is not the lavish, ground-up build that created China's original surveillance apparatus. Analysts estimated the first generation cost Chinese authorities roughly Rmb500bn in the mid-2010s. Today's tenders are far more modest, ranging from just under Rmb1mn to about Rmb10mn per district, because authorities are layering AI onto existing infrastructure rather than starting from scratch. Some local governments are even keeping old cameras and simply swapping out intermediary servers for so-called 'AI PCs' that handle video processing on-site, cutting cloud bills further.
The political backdrop matters. In 2024, Public Security Minister Wang Xiaohong issued a directive ordering police to upgrade equipment and move toward predictive policing. The push followed a string of violent incidents in Chinese cities that experts have linked to a mental health crisis worsened by pandemic-era lockdowns and a weakening economy. The Chinese government has already used surveillance heavily in Xinjiang to monitor the Uyghur minority; human rights groups warn that smarter AI cameras could give the state unprecedented power to shape public behavior across the country. Industry insiders note that initial deployments are concentrated in dense city zones, around military facilities, and near government buildings — exactly the places where Beijing prizes control. Whether the rest of the world treats this as a warning, a model, or both will depend on debates that are only just beginning.
Imagine typing 'woman in a red hat' into a search bar — and pulling up every camera that spotted her in the last 24 hours. That's China's new normal.
China is rebuilding the world's largest surveillance network — hundreds of millions of cameras strung across its cities — by stuffing modern AI into the hardware. A Financial Times analysis of more than a dozen procurement documents found local governments quietly buying new cameras and software that don't just record footage but interpret it: flagging erratic drivers, crowd build-ups, people climbing bridge rails, or anyone wandering near a military base.
The old system, built around a decade ago, was reactive. It could match a face to a watchlist but couldn't really 'understand' what it was seeing. The new generation embeds computer vision and large language models directly on the camera, so officers can search footage with text prompts instead of scrubbing through hours of video. Companies like Hikvision and Huawei are leading the rollout, with Public Security Minister Wang Xiaohong pushing predictive policing after a wave of violent incidents in 2024.
The shift isn't 'more cameras.' It's the same cameras getting a brain transplant — moving from a recorder to an analyst.
Whatever country you live in, the technology being battle-tested in Chinese cities — on-device AI that watches, judges, and predicts — will shape global debates about privacy, policing, and what governments are allowed to know about you. UK agencies have already been stripping Hikvision cameras from sensitive sites. Schools, stadiums, and shopping malls in the US and Europe use similar tech. The questions you'll vote on as adults — facial recognition bans, predictive-policing laws, whether AI can flag you as 'suspicious' — start with cases like this one.
China has used its surveillance network heavily in Xinjiang to control the Uyghur minority, and now the same toolkit is being supercharged nationwide. Watch for three knock-on effects: export competition (Chinese AI-camera firms selling cheaper kit abroad), Western regulatory backlash, and a quieter trend — police forces everywhere realizing they can keep old cameras and just swap the brains behind them. That makes mass AI surveillance dramatically cheaper to roll out, which means it spreads faster than the politics around it can catch up.