AI Platforms In China

Forrester publishes two Waves for AI platforms. One evaluates the global market. One evaluates China.

Put the vendor lists side by side and they share almost nothing. The global evaluation covers the American hyperscalers, the data platforms, and the enterprise application vendors. The China evaluation covers Alibaba Cloud, Baidu, Huawei, Tencent, and a set of specialists most Western buyers have never encountered.

Same category name. Same underlying technology. Effectively disjoint vendor populations.

Forrester maintains separate coverage for a handful of markets in China, and it does so where the local market is not a regional variant of the global one but a different market that happens to solve the same problem. AI platforms is the clearest current example.

Why this gets its own evaluation

Three forces keep the two markets apart, and none of them are about product quality.

The first is regulation. China regulates generative AI services directly, with requirements around filing and registration of public-facing generative services, content obligations, and data handling rules that shape what a platform must do before a customer can deploy anything customer-facing. A platform built without those mechanisms is not a slightly worse option for a Chinese enterprise. It is unusable.

The second is data residency and sovereignty expectations, which run in both directions. Enterprises operating in China face requirements about where data lives and how it crosses borders. Multinationals running global platforms elsewhere frequently run something different inside China for exactly this reason.

The third is compute. Export controls have restricted the flow of advanced AI accelerators into China, which pushed domestic platforms toward locally produced silicon and toward efficiency techniques that reduce dependence on the newest hardware. That constraint shows up in platform architecture rather than in marketing, and it is one of the more consequential differences between the two markets.

The result is two ecosystems solving the same problem under different constraints, which is precisely the situation that justifies separate analyst coverage.

The lineage

Forrester has tracked this market through four names, and the sequence is worth laying out because it maps the technology shift cleanly.

The Forrester Wave: Predictive Analytics And Machine Learning In China, Q4 2020 scored nine providers against twenty nine criteria: 4Paradigm, Alibaba Cloud, Baidu AI Cloud, Beagle Data, Huawei Cloud, Inspur, JD Cloud and AI, Percent, and Tencent Cloud.

Look at that list against any 2020 global evaluation of the same market and the separation is already complete. Not one shared name.

The AI/ML Platform Landscape In China, Q4 2022, authored by Danny Mu and Guannan Lu, mapped thirty vendors, which tells you the market had considerably more depth than a nine-vendor Wave suggests.

The Forrester Wave: AI/ML Platforms In China, Q4 2023, published in October 2023, scored fourteen providers against twenty five criteria. This is the interesting edition, because Forrester used it to incorporate foundation model support into the criteria, reflecting generative AI in the Chinese market.

The Forrester Wave: AI Platforms In China, Q4 2025 is the current evaluation, published under the shortened category name.

The China edition led the rename

Forrester's own account of how it has branded this coverage is unusually explicit, and it contains a detail worth noticing.

The firm began covering this space in 2015 as predictive analytics. In 2017 it became predictive analytics and machine learning, reflecting the rise of deep learning. In 2022 it expanded to AI/ML platforms, taking a broader view with machine learning at the core.

Then, in 2023, Forrester incorporated foundation model support into the China version of the AI/ML platform Wave specifically, to reflect generative AI trends in that market.

The China evaluation absorbed foundation models as a scored capability at a point when the global category was still branded around machine learning. That is not a small detail. It suggests Forrester's analysts saw the shift landing in enterprise buying decisions in China early enough to build it into criteria, and it undercuts the assumption that the Chinese market follows the global one.

The eventual global rename to AI Platforms came later, alongside a redefinition driven by agentic AI that swapped data science tooling vendors for workflow and application vendors.

Inside The Forrester Wave: AI Platforms In China, Q4 2025

The current evaluation scores AI platform providers in the Chinese market, published in Q4 2025.

Individual vendor placements are not publicly confirmable at the time of writing. Chinese vendors announce analyst recognition primarily through domestic channels, and those announcements do not surface in English-language search the way a Western vendor's press release does. Treat any English-language claim about who led this Wave with caution unless it points at the vendor's own material.

What can be established is the shape of the field. The vendors evaluated across the 2020, 2022, and 2023 editions are the major cloud providers, Alibaba, Baidu, Huawei, Tencent, and JD, alongside AI specialists including 4Paradigm and a longer tail of firms covered in the Landscape. The competitive structure resembles the global market in one respect: hyperscalers with distribution advantages competing against specialists with depth.

It differs in that the Chinese hyperscalers also build their own foundation models at frontier scale, which collapses a distinction that still exists in the global market. Forrester separates frontier model providers from AI platforms in its global coverage, with a dedicated Landscape and Wave planned for the model builders. In China, several of the platform vendors are the model builders.

Three things that make this market structurally different

Open weights changed the economics. Chinese research labs have been among the most active publishers of openly available model weights, and the practical effect on platform economics is significant. An enterprise that can download capable weights and run them on its own infrastructure has a different negotiating position than one dependent on a metered API, and platforms serving that market have had to accommodate self-hosted deployment as a first-class pattern rather than an enterprise afterthought.

Compute constraints produced different engineering priorities. With restricted access to the newest accelerators, efficiency work that is optional elsewhere becomes necessary. Quantisation, distillation, mixture-of-experts architectures, and inference optimisation get more attention because the alternative is not available. Some of that work has flowed outward and influenced the global field.

Vertical concentration runs differently. Enterprise AI adoption in China has clustered heavily in manufacturing, logistics, financial services, and government-adjacent sectors, and the platforms reflect that in their prebuilt assets and industry accelerators. A platform strong in manufacturing quality inspection is solving a problem that features less prominently in evaluations built around North American buying patterns.

Who this research is for

Forrester states the audience for its China evaluations as enterprise clients in China or doing business in China, and that framing is the right filter.

Three buyer types get real value from it.

Multinationals operating in China, who typically cannot run their global AI platform inside the country and need to select a local one. For them this is not a comparison exercise against the global Wave. It is a separate procurement with separate criteria.

Chinese enterprises selecting domestically, for whom this is simply the relevant evaluation.

And strategists tracking where the technology is going, for whom the value is comparative. Reading the two Waves against each other tells you which capabilities are converging and which are diverging, and the China edition adopting foundation model criteria first is exactly the kind of signal that only appears if you read both.

For everyone else, the honest answer is that this evaluation will not inform a purchase. It is context rather than a shortlist.

A note on reading this market from outside

Two cautions worth stating plainly.

Coverage of Chinese technology in English-language sources is thin and uneven, and the reporting that does exist frequently arrives filtered through a geopolitical frame rather than an enterprise buying frame. Benchmark claims, deployment numbers, and capability comparisons all deserve the same scepticism you would apply to any vendor-supplied figure, in both directions.

And the policy environment moves quickly on both sides. Export control scope, model registration requirements, and data transfer rules have all changed materially within the past two years, and any of them can change again in a way that alters what a platform can offer. An evaluation published in Q4 2025 describes the market under the rules that applied then.

That is a reason to check the current position rather than a reason to discount the research, and it applies to every regulated technology market rather than uniquely to this one.

What to test

If you are actually buying here, the questions differ from the global category.

Confirm regulatory standing directly. Ask which models on the platform are cleared for the deployment pattern you intend, particularly anything customer-facing, and what the process and timeline look like for a model that is not yet cleared.

Establish the data path precisely. Where training data, inference requests, and logs are stored and processed, what leaves the country, and what documentation exists for it. If you are a multinational, this determines whether your global governance framework can accommodate the deployment at all.

Ask what hardware it runs on and what that means for you. Availability, roadmap, and portability if you later need to move a workload. This is a more consequential question in this market than elsewhere.

Test the self-hosted path if openness matters to you. Platforms differ considerably in how well they support running open weights on your own infrastructure versus consuming a hosted service, and the commercial terms differ with it.

And check integration against the software you actually run. A platform optimised for the domestic enterprise software ecosystem may connect poorly to the systems a multinational runs globally, and that gap is where deployments stall.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.