AI / LLMMulti-region
← Sticker Wall
JOURNEY · DETAIL
Creao AI: Rebuilding Enterprise AI Middle Office with Agent OS, Breaking Through with Tens of Millions of USD in Financing
Founded: Anonymous (Core team members from major tech giant AI platforms) · Creao AI
JOURNEY
Key Fields
FIELD STAMPSIndustryAI / LLM
RegionMulti-region
ScaleMid-size
ChannelOther
Origin
The founding team originated from the AI platform department of a leading tech giant. During the process of serving major clients, they discovered that companies wanting to use AI were bottlenecked by system integration, data compliance, and multi-model scheduling. General-purpose large models could not directly meet enterprise demands for on-premises deployment, auditability, and multi-Agent collaboration. The team judged that the future entry point for work is not a chat box, but an Agent operating system capable of scheduling underlying computing power, models, and business systems, prompting them to leave and start a business building Agent OS.
Milestones
2023
Seed Stage Failure
Initially, the team provided vertical-industry large model fine-tuning services, taking on custom projects for 3 financial clients. However, they found that every project required rewriting engineering code, with delivery cycles lasting up to 4 months, a single project gross margin of less than 20%, and full-year 2023 revenue of less than 5 million RMB. The team faced a capital chain fracture and was forced to lay off half of their frontend engineers.
2024
Direction Adjustment Turning Point
During a review, the founders found that all clients were repeatedly asking for the same set of components: model routing, permission management, audit logs, and Agent orchestration. The team decided to stop taking custom projects, abstracted their engineering capabilities into a product, invested 6 months into developing the first version of Agent OS, and released a closed-beta version in March 2024 focusing on private deployment.
2024
PMF Validation PMF
After the closed-beta release of Agent OS, a leading brokerage firm was willing to pay 1.2 million RMB for private deployment, activating 200 Agent seats. Within the next half-year, the team secured 5 financial and 3 manufacturing clients, achieving a full-year 2024 ARR of approximately 8 million RMB, with subscription revenue from the private deployment version accounting for over 60%, validating product-market fit.
2025
Scaling and Competitive Pressure Growth
In the first half of 2025, Creao AI added over 20 new clients, with ARR breaking 30 million RMB mid-year, and the team expanding to 70 people. However, concurrent rapid iterations of open-source and free platforms like Dify and Coze directly eroded the SME market, extending Creao's contract signing cycle from 3 months to 5 months, and pushing the lost-order rate up to 35%, forcing a strategic shift from SMEs to mid-to-large-sized enterprises.
2026
Financing and Strategic Upgrade Inflection Point
On April 14, 2026, Creao AI announced the completion of tens of millions of USD in financing, restructuring the work entry point in the AI-native era with Agent OS. Post-financing, the team primarily allocated funds toward underlying computing power adaptation and multi-model scheduling engine upgrades, while establishing ecological cooperation with domestic chips and cloud vendors to provide end-to-end private AI middle-office solutions for key industry clients, with newly added contract amounts exceeding 40 million RMB in the first half of 2026.
Turning Points
- In early 2024, abandoned custom project delivery, abstracted engineering capabilities into a standardized Agent OS product, transitioning from a project company to a product company.
- In mid-2025, with the rise of open-source platforms Dify and Coze, Creao was forced to exit the SME market and strategically focus on the private deployment needs of mid-to-large-sized enterprises.
- In April 2026, completed tens of millions of USD in financing, adapted underlying engines with domestic chips, and entered the customer base of state-owned enterprises with self-organized computing power.
Failures & Pitfalls
- In 2023, vertical-industry large model fine-tuning outsourcing, a 4-month delivery cycle, and gross margins under 20% brought the capital chain close to collapse, forcing a 50% headcount reduction.
- In 2024, the first version of Agent OS only supported the CUDA ecosystem; an Xinchang client required domestic chip adaptation, and the team spent 3 months rewriting code before delivery.
- In 2025, small-to-medium clients were replaced by open-source platforms like Dify, causing Creao's single-quarter lost-order rate to rise to 35% and quarterly revenue to drop 15% quarter-on-quarter.
关键成功要素
- Core product Agent OS solves the standardization problems of model routing, permissions, auditing, and multi-Agent orchestration in enterprise private deployments.
- ARR grew from the 8 million to the 40 million level between 2024 and 2026, validating the willingness of mid-to-large enterprises to pay for a private AI middle office.
- The team's early hands-on engineering delivery experience provided authentic client demand inputs for product abstraction.
- Following the tens of millions of USD in financing in 2026, deep integration with domestic chips and cloud vendors built an integrated closed-loop from computing power to models to the middle office.
Lessons
- Building an AI middle office must start with standardized products rather than custom projects, otherwise delivery costs will consume gross margins and drag down cash flow.
- The SME market is easily covered by open-source and free platforms; private AI middle-office startups should focus on the compliance and custom depth of mid-to-large enterprises.
- Domestic adaptation is not a bonus feature but a mandatory requirement; ignoring the Xinchang ecosystem will directly lead to losing state-owned enterprise orders.
- Extended contract cycles after product launches are often signals of intensifying competition, and responses should prioritize client segmentation over price cuts.
Core Data
- 2024 Annual Recurring Revenue：Approximately 8 million RMB (publicly disclosed figures, independent review unverified)
- Mid-2025 Annual Recurring Revenue：Approximately 30 million RMB (publicly disclosed figures, independent review unverified)
- H1 2026 Newly Added Contracts：Exceeding 40 million RMB (publicly disclosed figures, independent review unverified)
- Team Size：Approximately 70 people (as of mid-2025) (publicly disclosed figures, independent review unverified)
- Financing Amount (April 2026)：Tens of millions of USD (publicly disclosed figures, independent review unverified)
Competitors / Peers
In the private AI middle-office track, Creao AI's main competitors include open-source or lightweight platforms such as Dify, Coze, and LangChain, which rapidly penetrate the SME market with free or low-cost offerings, creating strong pressure on Creao. Simultaneously, cloud-native Agent platforms from major tech giants like Volcano Engine, Alibaba Cloud Bailian, and Zhipu MaaS are also expanding upstream, holding cost advantages in computing power scheduling and model supply chains. Additionally, some traditional enterprise SaaS vendors are embedding AI capabilities into their existing product lines, forming lateral competition against private middle-office services.
Sources
- https://news.qq.com/rain/a/20260414A02P4P00
- https://news.qq.com/rain/a/20250716A05GNO00
- https://36kr.com/p/3924625732221059
- https://neodrop.ai/post/psv3RbMVkRj
- https://36kr.com/p/3821594633769346
- https://yunpan.plus/t/27279-1-1
- https://neodrop.ai/post/xXyP5MTQIAI
- https://liuwei.blog/2026/03/31/%e5%9b%9b%e4%b8%aamit%e5%b0%91%e5%b9%b4%e7%9a%84%e7%bc%96%e7%a8%8b%e5%b8%9d%e5%9b%bd/
- https://www.lukuai.com/25773.html
- https://www.36kr.com/p/3856646387137412
- https://neodrop.ai/post/_xPpXHvefgk
- https://163.com/dy/article/KVPL1NTL05568W0A.html