AI Wiki

Explore AI Wiki

Citation and evidence

Model-as-a-Service Market in China

15 min full readUpdated 26 references

This article's verification

Report a problem with this article

Edit article
More

Use this article

Raw MarkdownExplore connections

Improve this page

Revision history

Topics

AI InferenceChinese AIEnterprise AILarge Language Models

Cite this article

The model-as-a-service (MaaS) market in China is the commercial market for access to large language models and other generative models through public-cloud APIs, where enterprises and developers pay per token or by subscription instead of deploying models on their own hardware. The market research firm IDC tracks it in its China AI Software Market Semiannual Tracker and reported that MaaS revenue in China reached RMB 10.63 billion in the first half of 2026, up 889.4% from a year earlier.[1] By revenue, ByteDance's Volcano Engine led the overall market with 45.1% in that period, while Alibaba led the large-language-model segment with 46.6%.[1] The market took off after a price war that began in May 2024, when Volcano Engine, Alibaba Cloud and Baidu cut model prices steeply; by 2026 some providers, most visibly Zhipu AI, were raising API prices instead.[5][13][17][3]

Definition and scope

IDC measures the market two ways. Its token-volume measure counts the tokens enterprises generate by calling model APIs on public-cloud MaaS platforms; its revenue measure counts the fees enterprises actually pay for public-cloud MaaS services and excludes on-premises (private) deployments, which IDC tracks as a separate market for large-model platform software and licenses.[2] Usage by a cloud provider's own consumer products is excluded: Volcano Engine's figures, for example, do not include calls from ByteDance apps such as Doubao and Jimeng (Dreamina), or tokens generated inside Douyin and Lark.[4][23]

IDC splits revenue by model type. In the first half of 2026, large language models accounted for 47.0% of MaaS revenue, video-generation models for 32.7%, and other modalities for 20.2%.[1] IDC described video generation, led by models such as ByteDance's Seedance 2.0, as an important second growth curve for the market.[1] IDC groups the sellers into four types: cloud providers that bundle models with compute and tooling, large-model companies that compete on new releases and aggressive pricing, inference-service providers that aggregate many models, and telecom carriers with national compute footprints and government and enterprise sales channels.[1]

Market size

IDC's half-year series shows revenue rising almost tenfold in a year while token volume grew a little more than fivefold.[1]

PeriodMaaS revenue (RMB million)Token volume (trillion tokens)Source
2024 (full year)not in IDC's 2026 releases114.2IDC, April 2025[6]
H1 20251,075536.7IDC, 26H1 tracker[1]
H2 20251,9961,404.4IDC, 26H1 tracker[1]
2025 (full year)3,0701,944IDC, May 2026[2]
H1 202610,6312,857.9IDC, 26H1 tracker[1]

Expanded article table

IDC put year-on-year growth in the first half of 2026 at 889.4% for revenue and about 432.5% for token volume.[1] Over the same half year, MaaS rose from 5.2% of China's public-cloud platform-as-a-service (PaaS) market in the second half of 2025 to 21.0%.[1] The private-deployment market for large-model platforms reached RMB 2.88 billion in the first half of 2026, up 108.4%, with demand coming mainly from government, manufacturing, energy and finance customers.[1]

Daily consumption climbed steeply through 2025. IDC reported average daily MaaS consumption of about 1.6 trillion tokens in January 2025 and 9.6 trillion tokens by the end of December 2025.[2] An earlier IDC report had put daily public-cloud model calls at 96.3 billion tokens in mid-2024 and 952.2 billion in December 2024.[6]

In May 2026 IDC forecast that China's MaaS market would consume about 40,000 trillion tokens in 2026 and generate about RMB 18.6 billion in revenue, with a 2024-2030 compound annual growth rate of about 1,154.9%; it cautioned that this was a high-growth scenario that depended on multimodal models maturing, agent applications scaling, compute supply holding up and compliance policy staying stable.[2] IDC China research director Lu Yanxia said in 2025 that the large-model service market still had room for more than a hundredfold growth.[7]

Vendor shares

Overall MaaS revenue, H1 2026

RankVendorShare of revenue
1Volcano Engine (ByteDance)45.1%
2Alibaba25.2%
3Zhipu AI5.0%
4Baidu4.5%
5Kuaishou3.7%
6iFlytek1.8%
7China Telecom AI company1.6%
8PPIO1.0%
Others12.3%

Expanded article table

Source: IDC, China AI Software Market Semiannual Tracker, 26H1; IDC notes that rounded figures may not sum exactly.[1]

Large-language-model revenue, H1 2026

Measured only on large-language-model revenue, the order changes. Alibaba Cloud took 46.6%, more than the next five vendors in IDC's chart combined (their shares sum to 33.1%), and Zhipu ranked second, slightly ahead of Volcano Engine.[1][8] IDC attributed the dominance of language models to their technical maturity, faster-falling inference costs, and enterprise demand that is still mostly text-based, such as knowledge Q&A, code generation and agent orchestration.[1]

RankVendorShare of LLM revenue
1Alibaba Cloud46.6%
2Zhipu AI10.6%
3Volcano Engine10.2%
4Baidu6.7%
5iFlytek3.5%
6PPIO2.1%
Others20.5%

Expanded article table

Source: IDC, China AI Software Market Semiannual Tracker, 26H1.[1] Chinese financial press reported the figures as revenue-based shares, with language models counted separately from image and video models.[9]

Earlier rankings by token volume

Before 2026, IDC's widely cited MaaS rankings measured token volume rather than revenue. Volcano Engine led every period.[6][10][11]

Period (report)Total volumeVolcano EngineAlibaba CloudBaidu AI Cloud
2024 (IDC report released April 2025)114.2 trillion tokens46.4%not reported in coveragenot reported in coverage
H1 2025 (IDC report released September 2025)536.7 trillion tokens49.2%27%17%
2025 (IDC data released May 2026)1,944 trillion tokens49.5%28%10%

Expanded article table

The 2024 total excluded MaaS platforms used by Chinese companies operating overseas and calls made inside model apps.[6] For 2025, IDC listed the top five by token volume as Volcano Engine, Alibaba Cloud, Baidu AI Cloud, SiliconFlow and China Mobile Cloud, with Tencent Cloud, SenseTime, Huawei Cloud and China Telecom's Tianyi Cloud also noted.[2] The 2025 Alibaba and Baidu percentages come from press coverage of the IDC report rather than IDC's own summary.[11]

2025 by revenue

IDC's May 2026 summary gave revenue-based shares for 2025 only in outline: Volcano Engine held more than 40%, and the top five were Volcano Engine, Alibaba Cloud, Baidu AI Cloud, Zhipu and China Mobile Cloud.[2] According to KrASIA's reporting on the same IDC data, Volcano Engine's revenue share was a few percentage points lower than its token-volume share because its per-token price was below the industry average.[4]

Private deployment, H1 2026

The separate on-premises market has a different leader and is less concentrated.[1][2]

VendorShare of revenue, H1 2026
Baidu15.2%
SenseTime9.4%
China Telecom AI company7.8%
AInnovation3.7%
Zhongguancun Kejin2.6%
Transwarp1.7%
Others59.5%

Expanded article table

Source: IDC, China AI Software Market Semiannual Tracker, 26H1.[1]

Pricing dynamics

The 2024 price war

The market's early growth followed sharp API price cuts in May 2024. DeepSeek-V2 was priced at 1 yuan per million input tokens and 2 yuan per million output tokens, roughly 1% of GPT-4's price, and Chinese press later described it as having started the price war.[12] In mid-May 2024 ByteDance priced its flagship Doubao Pro model at 0.0008 yuan per 1,000 tokens, a level it said was 99% below the industry average for enterprise users.[13][5] On 21 May, Alibaba Cloud cut prices by as much as 97% on nine products built on its Tongyi Qianwen (Qwen) models, and Baidu said within hours that its ERNIE Speed and ERNIE Lite models would be free for business users.[14][13] TechNode reported that after Alibaba's cut, RMB 1 could buy processing of about two million input tokens.[15] iFlytek followed on 22 May, charging 0.21 yuan per 10,000 input tokens.[13]

In a May 2026 report on Volcano Engine, KrASIA wrote that the company made token consumption a core business-development metric and that the technologies behind its low prices were mainly prefill-decode disaggregation and KV cache reuse, which pay off mainly at large call volumes.[4] Volcano Engine president Tan Dai said that once the company saw "technology could bring costs down, it decided to cut prices all the way."[4] Volcano Engine said in October 2025 that daily Doubao usage had grown from 120 billion tokens in May 2024 to more than 30 trillion tokens by September 2025.[7]

From price cuts to price increases

By early 2026 the direction had partly reversed. Citing the Chinese outlet chinastarmarket.cn, TrendForce reported that more Chinese companies had stopped cutting prices since 2025, that Zhipu, Moonshot AI, MiniMax and StepFun had already raised some API prices in 2025, and that Alibaba, ByteDance, Tencent and Baidu had largely moved to tiered pricing.[16] On 12 February 2026 Zhipu raised prices for its GLM Coding Plan subscriptions by at least 30%, citing a rapid increase in users relying on its models for coding.[17] Zhipu's 2025 annual report said its API pricing had risen 83% from the end of 2025, and CEO Zhang Peng said at the Zhongguancun Forum on 27 March 2026 that long-term low-price competition was bad for the industry.[3]

Zhipu's interim results for the six months to 30 June 2026 showed revenue of RMB 954 million, up 399.7%, with its open platform and API business contributing RMB 825 million, or 86.5% of the total.[18][19] The company said the gross margin of the API business rose to 24.6% from -0.4% a year earlier, and attributed the change to economies of scale, higher prices and lower inference costs.[18][19] It reported that its average API selling price had risen about 101%, that token calls on its MaaS platform had increased more than 40-fold since the start of 2026, and that inference cost per token had fallen about 80% after it moved inference onto domestic Chinese chips.[18] The company still posted a loss of about RMB 2.07 billion for the period.[18][19]

Prices did not move in one direction across the whole market. IDC wrote that model launches, API price cuts, open-source releases and compute expansion all came in quick succession in the first half of 2026.[1] It argued that, as model capabilities converge, inference efficiency (inference frameworks, prefill-decode separation and large-cluster scheduling) decides how many useful tokens a provider can sell from the same compute, and so how far it can afford to cut prices; in the longer run, IDC said, access to compute at low unit cost would be the main moat.[1] In IDC's May 2026 survey data, cost-effectiveness ranked fifth among the factors affecting enterprise adoption of large models, behind model performance, security and compliance requirements, answer quality, and availability on an AI platform.[2]

Billing is also diversifying. IDC reported that pay-per-use API billing remained the main model in the first half of 2026, while subscription plans such as coding plans, reserved resources, pay-for-outcome pricing and revenue-sharing deals grew quickly.[1]

Key vendors

  • Volcano Engine, ByteDance's cloud unit, has led IDC's overall MaaS rankings by token volume since 2024 and by revenue in 2025 and the first half of 2026.[1][2][10] Its Ark platform sells ByteDance's Doubao and Seedance models alongside third-party models from DeepSeek, Zhipu and Kimi.[23] Its share of language-model revenue alone was 10.2% in the first half of 2026.[1]
  • Alibaba Cloud ranked second overall by revenue (25.2%) and first in language models (46.6%) in the first half of 2026, selling its Qwen family and other models.[1][8]
  • Zhipu AI (also known as Z.ai), which listed on the Hong Kong Stock Exchange in January 2026, ranked third overall (5.0%) and second in language models (10.6%).[1][3] IDC named GLM-5.2 and Qwen3.5 as flagship models whose gains in reasoning, coding and agent tasks pushed adoption in the first half of 2026.[1]
  • Baidu held 4.5% of overall MaaS revenue and 6.7% of language-model revenue in the first half of 2026, and led the separate private-deployment market with 15.2%.[1]
  • iFlytek held 1.8% overall and 3.5% of language-model revenue.[1]
  • Kuaishou held 3.7% of overall MaaS revenue but is not among the vendors IDC named in the language-model chart.[1]
  • PPIO, an independent MaaS and inference provider, ranked sixth in language models (2.1%); Chinese press coverage described it as the only independent MaaS provider among the ranked vendors.[1][20] The company said its platform carried more than 200 models and averaged 3.4 trillion tokens a day as of September 2026, and that more than 80 of its pull requests had been merged into the open-source SGLang inference framework.[20]
  • SiliconFlow, a MaaS provider, ranked fourth by token volume in 2025.[2] It announced a Series B round of more than RMB 2 billion in June 2026.[21]

Measurement caveats

Rankings in this market depend heavily on the metric; 36Kr wrote that each cloud provider picks the statistical definition that favours it.[22] Volcano Engine leads by token volume and by total MaaS revenue; Alibaba leads in language-model revenue and in broader AI cloud measures.[1][12] Volcano Engine's revenue share has trailed its token-volume share, and its per-token price was below the industry average.[4] Some industry insiders have also argued that MaaS figures miss customers who rent infrastructure and deploy open-source models themselves, since that spending shows up as IaaS or PaaS rather than model-service revenue.[5]

Other research firms define the market more broadly. Omdia put China's AI cloud market (including AI infrastructure, platform and model services) at RMB 22.3 billion in the first half of 2025, with Alibaba at 35.8%, Volcano Engine at 14.8%, Huawei Cloud at 13.1%, Tencent Cloud at 7% and Baidu at 6.1%.[12][11] IDC's separate 2024 study of AI public-cloud services, which covers computer vision, conversational AI, speech, NLP and machine-learning platforms, put that market at RMB 19.59 billion, with Baidu AI Cloud and Alibaba Cloud tied at 24.6%.[22][12] 36Kr counted at least seven AI-cloud market reports from IDC, Omdia and Frost & Sullivan in 2025; Alibaba ranked first in five, Volcano Engine in two and Baidu in one (shared).[22]

The figures also shift between releases. IDC's half-year token values for 2025 (536.7 trillion and 1,404.4 trillion) sum to slightly less than the 1,944 trillion full-year figure in its May 2026 summary, and IDC notes that its rounded share figures may not add to 100%.[1][2] Revenue figures also differ between IDC trackers. IDC's October 2025 release, from a tracker then called the China MaaS and AI Large Model Solutions Market Tracker, put first-half 2025 MaaS revenue at RMB 1.29 billion, up 421.2%, while the 26H1 tracker shows RMB 1,075 million for the same half, the base behind its 889.4% growth figure.[24][1] The earlier tracker had put the first-half 2024 market at RMB 250 million, and AIbase reported in April 2025 that IDC's follow-up report put full-year 2024 MaaS revenue at about RMB 710 million, up 215.7%.[25][26] Because ByteDance's own apps, other first-party products and private deployments are excluded, MaaS shares describe sales to outside customers, not total model usage.[2][4]

Continue reading

More to read

Topics:AI InferenceChinese AIEnterprise AILarge Language Models

References

  1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16 ^17 ^18 ^19 ^20 ^21 ^22 ^23 ^24 ^25 ^26 ^27 ^28 ^29 ^30 ^31 ^32 ^33IDC China. "中国MaaS市场实现近9倍增长：高增速下厂商加速涌入，未来竞争壁垒在哪？" (China's MaaS market grows nearly ninefold). IDC blog, 29 October 2026. idc.com/...%82%e5%95%86%e5%8a%a0%e9%80%9f%e6%b6%8c
  2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12IDC China. "中国MaaS市场进入高速增长期，Token经济从概念走向规模化实践" (China's MaaS market enters a high-growth period). IDC blog, 7 May 2026. idc.com/...%b5%e8%b5%b0%e5%90%91%e8%a7%84%e6%a8%a1
  3. ^1 ^2 ^3Jiemian News. "智谱上市后首份年报亏损47.18亿元，股价累计上涨近5倍" (Zhipu's first annual report after listing). 2026. m.jiemian.com/...14191900
  4. ^1 ^2 ^3 ^4 ^5 ^6Cheng Zi. "How ByteDance's Volcano Engine holds nearly half of China's MaaS market." KrASIA, 13 May 2026. kr-asia.com/...s-nearly-half-of-chinas-maas-market
  5. ^1 ^2 ^336Kr English. "Who leads China's AI cloud race? Alibaba, ByteDance, and Baidu all stake a claim." KrASIA, 27 October 2025. kr-asia.com/...tedance-and-baidu-all-stake-a-claim
  6. ^1 ^2 ^3 ^4China.com. "2024公有云大模型调用量，火山引擎中国第一！" (coverage of IDC's 2024 public-cloud large-model call-volume report). April 2025. m.tech.china.com/...042025_1659002
  7. ^1 ^2Yicai. "字节跳动最新披露：豆包大模型日均tokens使用量超30万亿" (ByteDance: Doubao daily token use exceeds 30 trillion). 16 October 2025. yicai.com/...102866248
  8. ^1 ^2Leiphone. "大语言模型'收入榜'出炉，阿里云以46.6%份额居首" (Alibaba Cloud leads LLM revenue ranking with 46.6%). 29 October 2026. leiphone.com/...VsFzLhOgbV95UM6v
  9. ^Zhitong Finance via Futu News. "阿里云占中国大语言模型MaaS市场份额46.6% 位居第一" (Alibaba Cloud ranks first with 46.6% of China's LLM MaaS market). 29 October 2026. news.futunn.com/...nked-first-with-a-46-6-share-of
  10. ^1 ^2National Business Daily. "Token经济崛起，大模型竞争进入'调用量'时代，火山引擎如何拿下半壁江山？" (coverage of IDC's 2025H1 report). 15 October 2025. nbd.com.cn/...4091641
  11. ^1 ^2 ^3PingWest. Coverage of IDC's May 2026 China enterprise MaaS market landscape analysis. May 2026. pingwest.com/...313519 ; Sina Finance. "AI时代，只剩火山和阿里两朵云？" 12 May 2026. finance.sina.cn/...detail-inhxrshc1273498.d
  12. ^1 ^2 ^3 ^4The Paper. "大厂云的'定语'战争" (The qualifier war of big-company clouds). 26 September 2025. m.thepaper.cn/newsDetail_forward_31686488
  13. ^1 ^2 ^3 ^4Fortune. "China's AI race heats up as ByteDance, Baidu, Alibaba trigger LLM price war." 22 May 2024. fortune.com/...nts-bytedance-alibaba-baidu-iflytek
  14. ^Bloomberg via Caixin Global. "Alibaba Comes Out Fighting in China AI Price War With Drastic Discounts." 22 May 2024. caixinglobal.com/...th-drastic-discounts-102198812
  15. ^TechNode. "Price war in China's AI field expanded as Alibaba Cloud and Baidu have drastically reduced model prices." 22 May 2024. technode.com/...e-drastically-reduced-model-prices
  16. ^TrendForce. "Rising Costs and Demand Drive China's LLM Price Jump: Zhipu GLM-5 Hikes 30% in First 2026 Increase." 16 February 2026. trendforce.com/...-hikes-30-in-first-2026-increase
  17. ^1 ^2Reuters via Yahoo Finance. "Chinese AI startup Zhipu hikes prices for coding plan as demand rises." 12 February 2026. finance.yahoo.com/...startup-zhipu-hikes-003904634
  18. ^1 ^2 ^3 ^4Sina Finance. "智谱中报：收入涨近4倍，但真正变化是API成了主收入" (Zhipu interim report: API becomes main revenue). 7 September 2026. finance.sina.cn/...detail-iniqymaa6342962.d
  19. ^1 ^2 ^3EqualOcean. "Zhipu AI's H1 Revenue Jumps Nearly 400% as API Business Takes 86.5% of Sales." 2 September 2026. equalocean.com/...as-api-business-takes-86-5-sales
  20. ^1 ^2NetEase Tech. "IDC报告：上半年中国MaaS市场规模达106.3亿元" (IDC: China's H1 MaaS market reaches RMB 10.63 billion). 29 October 2026. 163.com/...LADPSKFQ00098IEO
  21. ^The Paper. "大模型造就'Token工厂'，硅基流动近一年营收同比增长超十倍" (SiliconFlow revenue grows more than tenfold). June 2026. m.thepaper.cn/newsDetail_forward_33389828
  22. ^1 ^2 ^336Kr. "中国AI云市场，究竟谁是'第一'？" (Who is really first in China's AI cloud market?). 2025. 36kr.com/...3485776450346114
  23. ^1 ^210jqka (Tonghuashun). "上半年中国MaaS营收破百亿，火山引擎45%份额中国第一" (Volcano Engine statement on IDC's 26H1 data). 29 October 2026. news.10jqka.com.cn/...c681238524.shtml
  24. ^IDC China. "IDC：大模型市场双线爆发，开源+多模态释放AI商业化潜能" (China MaaS and AI large model solutions market tracker, 2025H1). IDC press release, 28 October 2025. idc.com/...a%e6%94%beai%e5%95%86%e4%b8%9a%e5%8c%96
  25. ^IDC China. "增长规模超100%，大模型市场进入增长快车道" (China MaaS and AI large model solutions market tracker, 2024H1). IDC press release, 24 October 2024. idc.com/...%9e%e9%95%bf%e5%bf%ab%e8%bd%a6%e9%81%93
  26. ^AIbase. "2024年中国MaaS与AI大模型解决方案市场迎来爆发式增长" (coverage of IDC's 2024 full-year MaaS figures). 29 April 2025. aibase.com/...17674

Improve this article

Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.

1 revision · v1 · 3,059 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Market sizes, vendor shares and growth rates checked against IDC's October 2024, October 2025, May 2026 and October 2026 releases, along with Chinese and English press coverage of the price war and Zhipu's interim results, October 2026.

Cite this page: AI Wiki. "Model-as-a-Service Market in China." aiwiki.ai, updated 29 Oct 2026, fact-checked 29 Oct 2026. CC BY 4.0. https://aiwiki.ai/wiki/china_model_as_a_service_market

Edit article