Megvii at 15: Prioritize Business First Before Achieving Profitability

In late October, market rumors spread that Megvii is considering a Hong Kong IPO, planning to raise 500 million to 1 billion US dollars. A Megvii spokesperson responded: "We have no specific IPO plans." It has been nearly two years since Megvii withdrew its IPO application from the Sci-Tech Innovation Board. Counting from its first submission of listing documents in 2019, the company's long journey to go public has lasted for more than seven years, with two failed attempts to land a listing and no success yet.

But what is more noteworthy about Megvii recently is not the IPO rumors, but the internal letter issued by CEO Zhao Kang on the occasion of the company's 15th anniversary on October 8. In the letter, Zhao Kang summed up the core actions of the past year in four words: strategic convergence. Megvii no longer tries to be a company covering all AI scenarios, but focuses on the main track of "physical space operation", taking POA OS as the base, and deploying agents in five scenarios: urban governance, transportation operation, business intelligence, work safety, and workplace management.

Zhao Kang's new positioning for Megvii is: "We are no longer satisfied with being an AI technology company that 'sees accurately', we want to be the commercial value company that 'gets things done'."

Translated, this sentence means: In the past 15 years, Megvii sold algorithm capabilities; from now on, Megvii will sell replicable operation results. This is an identity shift from a "technology supplier" to an "operation service provider", and it is also the latest answer given by a company that re-answered "what exactly do we rely on to make money" after the IPO window was closed repeatedly.

The Quality of Turnaround from Loss to Profit

The most concerned information in the internal letter is that Megvii expects to achieve full turnaround from loss to profit in 2026.

Supported by several sets of growth data: the total number of model releases in the first half of 2026 increased by 180% year-on-year compared with the same period in 2025; the contract amount in the business intelligence field is expected to increase by more than 100% year-on-year; both the proportion of agent-related revenue and gross profit margin have increased significantly. As disclosed in the internal letter, Megvii's technical R&D personnel still account for more than 60% of its total staff.

However, all these growths are based on a very small base. Megvii recorded a cumulative loss of 14.634 billion yuan from 2018 to the first half of 2021, and the cumulative unrecovered loss reached 16.596 billion yuan as of the end of June 2021. Since then, Megvii has not publicly disclosed complete financial data, but its number of employees has been drastically reduced, and its business lines have been significantly shrunk. The value of its turnaround from loss to profit needs to be understood under the background of "slimming down first, then hematopoietic".

The four measures proposed by Zhao Kang all point to the same core problem: no wavering in direction, concentrating resources on the five major scenarios, not chasing short-term hot spots; convergent operation methods, breaking through replicable benchmarks first before pursuing scale, and each project must answer three questions: "Can it be replicated? Can customers renew their subscriptions? Can customers take the initiative to ask you to develop the next scenario?"; aligned organization, realizing collaboration among R&D, product, delivery and sales teams, and shifting the evaluation criteria from "who signed the order" to "who successfully delivered the project for the customer and realized its replication"; incentives are tilted towards teams that can deliver tangible agent results.

An industry insider close to Megvii gave a straightforward interpretation to 36Kr: The core of these four requirements is to transform from project-based delivery to replicable product and service revenue. "Customized delivery for government and enterprise customers has always been the bottleneck of AI companies' commercialization. What Megvii is trying to solve now is to turn every project into reusable assets."

The Replication Challenge

"Replicability" is what Megvii has lacked the most in the past 15 years. The common dilemma of the "Four Little Dragons of AI" lies in the diseconomies of scale of project-based delivery: every customer requires re-development, the payment collection cycle is long, half of the gross profit is taken away by system integrators, the larger the business scale, the more employees are needed, which essentially makes them more like engineering contractors than product companies. Megvii's "strategic convergence" essentially means that after admitting this model is unworkable, it tries to use standardized hardware, standardized systems and standardized applications to transform its business model from "earning a sum of money for one project" to "earning money many times for one product".

The problem is that the balance between standardization and customization is a difficult problem faced by all government and enterprise AI companies. Among the five scenarios, urban governance, transportation operation and work safety have highly scattered customers with vastly different demands, leaving limited space for standardization by nature. Business intelligence and workplace management are relatively closer to standardized products, but the competition in these two tracks is also the most fierce. Whether Megvii can find the intersection between "replicable" and "customer demand satisfying" is the key to the sustainability of its profit turnaround.

The Founders' Divergence

Another background of Megvii's 15th anniversary is the complete separation of its founding team.

Co-founder Yin Qi no longer holds any position in Megvii. In 2024, he stepped down as the legal representative, and in the following two years he successively served as the chairman of Qianli Technology and StepFun, the former focusing on vehicle manufacturing and the latter on large language models. In January 2026, StepFun completed a B+ round of financing of over 5 billion yuan, and Yin Qi officially took the position of chairman, responsible for promoting the commercialization of AI large models on end devices. Another co-founder, Tang Wenbin, founded embodied intelligence company Forcelink Intelligence in March 2025, whose core team is composed of former Megvii members. In June 2026, Forcelink Intelligence acquired a logistics robot company and received investments from Zhipu AI, SenseTime, StepFun and other companies. Co-founder Yang Mu, together with Yin Qi and Tang Wenbin, pledged all their Megvii equity to Chongqing Industrial Investment Mother Fund in January 2025, which was interpreted by the outside world as the founding team's complete farewell to their former company.

Megvii used to be the most technology-focused company among the "Four Little Dragons of AI" — its founders graduated from the Yao Class of Tsinghua University, it owns the Face++ platform, and it has a high market share in face recognition solutions for the Android camp. However, its technology label has not been converted into a closed commercial loop. According to media reports, Yin Qi once said a sentence after leaving: "Technology can change the world, but it cannot change an enterprise. The top priority of any enterprise is always to form a closed commercial loop."

This sentence carries great weight in the context of Megvii's 15th anniversary. It does not negate technology, but admits that the real moat of an AI company does not lie in the algorithm itself, but in the depth of the algorithm's integration into the customer's workflow. Algorithms will be open-sourced, caught up with, and integrated into larger platforms. What can really retain customers is the company that masters the interface that customers have to open every day, and thus has the pricing power.

Megvii's current POA OS and five major scenarios essentially aim to become that "interface". But this is a path that Yin Qi and Tang Wenbin have chosen not to take within Megvii's system. The person who stays to move forward on this path is Zhao Kang.

The Business Test

Megvii's 15 years have exactly covered the rise and fall of the first full cycle of China's AI industry.

When it was founded in 2011, Megvii represented the belief that "algorithms can change the world". When it first sprinted for IPO in 2019, it represented capital's fanaticism for AI unicorns. When it withdrew its Sci-Tech Innovation Board IPO application in 2024, it represented the dilemma of the project-based AI business model. By its 15th anniversary in 2026, it represents a more simple yet harder-to-answer question: Can an AI company that has raised tens of billions of yuan in financing, lost over 100 billion yuan, and failed two listing attempts survive first with a replicable small business?

There is one sentence in Zhao Kang's internal letter that is the most down-to-earth: "Excellent technology is only the admission ticket. To deeply integrate technology with scenarios, make customers willing to renew their subscriptions, and realize sustainable revenue is the real capability."

This sentence is correct. But it also shows that Megvii has spent 15 years distinguishing between the "admission ticket" and the "real capability". If the turnaround from loss to profit can be realized in 2026, it will be an observable milestone — it proves that Megvii has the ability to keep its cost under its revenue. But from turning a profit to sustainable profit, there is a harder threshold to cross: whether each of the five major scenarios can really create replicable benchmarks, and make customers take the initiative to renew their subscriptions and expand scenarios after the benchmarks are successfully implemented.

Megvii is no longer waiting for an IPO. It is waiting for a more fundamental thing: to prove that it is a company that can make sustainable profits, not just a company that once had cool technologies.

This article is from the WeChat official account "Tech Business", written by Tech Business, and authorized for release by 36Kr.