A SenseTime headquarters visit puts a delegation inside one of the companies that defined China's artificial-intelligence industry. The Beijing programme runs at SenseTime's Haidian site in the Ideal International Mansion — a showroom walk, a themed briefing by a company expert, and hands-on time with the systems the company is shipping today. What a management team takes home is not a showcase tour. It is how an organisation built on computer-vision research repositioned itself for the large-model era, and what computing power, models and applications actually look like when held together inside one company. The company's larger Shanghai campus operates separately.
Key takeaways
SenseTime's Beijing showroom is the company's core window onto its national computing layout, smart-city operations, SenseAuto driving business, healthcare and education units, and the web end of its SenseNova large model.
The reason to go is the full stack. SenseTime combines the SenseCore AI computing base, the SenseNova model family and industry applications, rather than acting as a pure software vendor or a model developer renting outside compute.
The headline experiences are the SenseAuto smart-cabin demonstrator — a working cabin rather than a display panel — and SenseNova's AIGC tools, including image generation and the AI digital-human platform.
The business story runs from the academic lab to a Hong Kong listing in late 2021, the AI computing centre coming online in 2022, and the SenseNova launch in 2023 that moved the company decisively into generative AI.
We arrange the programme end to end, including site scheduling on the Chinese side, transfers within Beijing and English interpretation on site. The Beijing visit is normally built as a compact half-day module.
Key visit details
Where the visit is held
Location — SenseTime's Beijing site on the 6th floor of the Ideal International Mansion, 58 North Fourth Ring West Road, Haidian District. The exact meeting point is confirmed after the company approves the booking. Delegations normally gather at the mansion's south entrance rather than the underground parking.
Getting there — the site is an office building rather than a company campus, and it has no dedicated visitor parking. Guests driving themselves should allow time to find a nearby public car park. We normally transfer the group by coach and handle the planning in advance.
Format and duration
Duration — A compact half-day, about one and a half hours on site: roughly forty minutes in the showroom and a one-hour themed briefing, followed by questions and a group photograph.
Format — A guided showroom walk covering the seven display areas, then a presentation such as 'Large-Model Trends and New Productive Forces' delivered by a senior company expert. Deeper technical workshops can be added when the delegation has the time.
Who the programme suits
Best suited to — Enterprise and public-sector teams studying large models and compute strategy; automotive and smart-cabin businesses; smart-city, healthcare and education organisations; AI product and algorithm teams; and executive groups benchmarking China's AI leaders.
Main themes — Large-model technology, AI computing strategy, generative-AI applications, intelligent driving and the cabin, and the industry rollout of AI across finance, healthcare and government.
Booking, group size and site rules
Booking — A client booking form is submitted around five working days ahead, and SenseTime confirms after internal review. Groups needing the public-holiday or peak-season window should start earlier, and the programme is best suited to groups of 10–20 participants.
Site rules — The office-floor reception is limited to roughly 15 to 30 visitors per group. Photography follows the host's rules on the day, and access is by the building's controlled lifts, which is why meeting logistics are confirmed before you travel.
Why a SenseTime visit is worth a delegation's time
The scale of the company alone justifies the trip. SenseTime employs around two and a half thousand people, and it has filed more than nine thousand patent applications, with research spending of roughly three point eight billion yuan in 2025. Its cumulative spending on research and on AI infrastructure has each passed the ten-billion-yuan level, and generative AI now accounts for more than seventy percent of its revenue. It serves organisations from State Grid and CNNC to Industrial and Commercial Bank of China and Huawei.
The more useful object of study is how the company answered the single biggest threat it has faced. Before generative AI, SenseTime's identity was discriminative — vision models that recognised and classified. The arrival of ChatGPT changed the question overnight, from perception to generation, and the response was the SenseCore-plus-SenseNova combination now on display. Reading that pivot in a report tells you the outcome. Standing in the showroom where the computing layout, the model and the applications are presented together tells you how the pivot was actually managed.
The compute question is the part delegations consistently underrate. SenseCore is not simply a server room for internal training — it supplies capacity to outside institutions and companies, which turns capital expenditure into a platform. The 'bullwhip effect' of AI compute, where small shifts in end demand produce large swings upstream, is the subject most likely to change how a visiting leadership team reads the rest of the market. See the Beijing tech tour page for combining SenseTime with other AI stops, and the Shanghai SenseTime visit for the company's larger campus.
Inside the Beijing showroom
The showroom presents seven areas in roughly forty minutes, and four of them carry most of the weight for a visiting team.
SenseNova and the web-end model
SenseNova large model. The web-end demonstration covers dialogue, code and copy through the SenseChat interface, together with industry use cases from finance to government. For a visiting team the useful question is not the benchmark number on the wall but how the model is embedded inside real workflows, and what that changes in staffing.
AIGC: Miaohua and the digital-human platform
Generative content. Miaohua turns a short text prompt into a finished poster, and the digital-human platform clones a speaker from a short video into a usable avatar. The exhibit is the clearest demonstration in the room of content production being compressed from days to minutes, and it is the stop most requested by marketing and creative teams.
SenseAuto and the smart cabin
SenseAuto cabin. The demonstrator is a working cabin rather than a display board: multimodal perception tracks gaze and expression, the cabin warns on fatigue, and the system reads objects outside the window as a guide. The useful question for an automotive visitor is how vision and the large model are deployed on automotive-grade chips at an acceptable cost.
Smart city, healthcare, education and mixed reality
Industry applications. The smart-city wall presents the history of the urban operations business; the healthcare and education areas show clinical and classroom deployment; and SenseMARS demonstrates the mixed-reality reconstruction used at major cultural sites. Together they answer the question a visiting executive asks first: where the model actually earns money.
The themed briefing
The one-hour session is the core of the programme, delivered by a SenseTime expert rather than a general guide. It runs in three parts.
New productive forces in the AGI era
Part one frames what new productive forces and the 'AI plus' agenda mean in practice, rather than as policy vocabulary. It sets the language the rest of the briefing uses.
Model trends and the compute bullwhip
Part two covers large-model development and the bullwhip effect across the compute supply chain — the segment most likely to change a delegation's planning assumptions for infrastructure.
From models to industry AIGC
Part three moves from the model to AIGC deployment, working through cases in finance, government, healthcare, education, intelligent driving and smart cities. The briefing ends with time for the delegation's own questions.
What different teams take away
For enterprise and public-sector teams — how large models and compute strategy are planned together, including what to build, what to rent and where the real cost sits.
For automotive and smart-cabin teams — how multimodal perception and in-cabin interaction are deployed on constrained hardware, and what fatigue and driver-monitoring systems actually detect.
For smart-city, healthcare and education teams — how AI projects are structured, measured and scaled inside regulated environments, beyond the pilot stage.
For AI product and algorithm teams — how a research-led organisation manages model iteration and compute allocation at the same time. The same showroom should look different to each group, and the briefing is shaped around whichever one your delegation is.
Sample visit programme
| Time | Session | What to expect |
|---|---|---|
| 20 min | Arrival and briefing | Coach transfer to the Haidian site, gathering at the Ideal International Mansion south entrance and check-in on the 6th floor. |
| 40 min | Showroom walk | Seven display areas: national computing layout, smart city, SenseAuto cabin, healthcare, education, SenseNova web end and the chess AI, with a company guide. |
| 60 min | Themed briefing | Large-model trends, the compute bullwhip and industry AIGC cases, delivered by a senior company expert, followed by Q and A and a group photograph. |
The point of the visit is not the screens on the wall. It is seeing the compute, the model and the application running as one system.
Practical notes for the day
Plan the entrance, not the car park. The site is an office building with no visitor parking, and the lifts require an access card. We confirm the south-entrance meeting point and transfer logistics in writing before the day.
Keep the group compact. The office-floor reception works best within roughly 15 to 30 visitors, and a larger group should be split across staggered time slots rather than arriving together.
Bring the right questions. The expert briefing is the core of the visit, and groups arriving with two or three specific questions around compute, deployment or industry rollout leave with answers they can act on.
Interpretation matters. The vocabulary of model architecture, compute scheduling and multimodal systems does not survive casual translation, which is why English interpretation on site is handled by people who work on these visits.
Questions before a SenseTime Beijing visit
Where is the visit held? At SenseTime's Beijing site on the 6th floor of the Ideal International Mansion, 58 North Fourth Ring West Road, Haidian District, with the gathering point at the south entrance.
Is this different from the Shanghai site? Yes. The Beijing programme is the Haidian showroom and themed briefing; the larger Shanghai campus operates as a separate visit, and both can be arranged.
What does the SenseAuto demonstrator show? It is a working smart cabin rather than a display board: multimodal perception that tracks gaze and expression, fatigue warning, and object recognition outside the window.
What does the themed briefing cover? Large-model trends, the compute bullwhip and AIGC industry cases, delivered for about an hour by a senior company expert with time for your own questions.
How large can the group be? The office reception works best within 15 to 30 visitors. A group of 10–20 participants is usually the best fit; larger groups are staggered or split. We charge per group, so larger groups usually offer better value per person. Individuals and smaller groups are welcome to enquire at a higher per-person rate.
How far in advance should we book? A booking form is normally submitted around five working days ahead, with earlier planning for peak periods, because the site needs internal approval.
Does the programme always run as described? It is built around the elements above. Facilities and areas opened on the day are subject to what the site opens, and we confirm the final scope in writing during planning, before you commit.
Is English interpretation provided? Yes. English interpretation on site is part of how the programme is run, including during the expert briefing.
Arranging your SenseTime visit
We arrange the visit for you end to end: the introduction to the host, the scheduling on the Chinese side and English interpretation on site. Tell us your dates, your delegation size and whether you also want Shanghai, and we will confirm what can be arranged. Start at chinatechvisits.com/company-visits.