Original · Unique Research · 2025-11-08
Editor’s note: This complete English edition preserves the original November 2025 report and its selected Q&A. Company metrics and operational claims are attributed to the interviewee and the source, not independently audited. The headline uses “revenue”; Q3 specifies annual recurring revenue (ARR), not recognized revenue or cash received. Age and working hours describe the interview period. The Q&A is a rendition of the source’s edited interview, not a verbatim transcript. “996” refers to working 9 a.m. to 9 p.m., six days a week. The ordinary conference portrait is omitted; its visible CAAIEC conference label adds no business facts.
In today’s noisy AI era, with computing power accelerating at breakneck speed, entrepreneurship seems to have been redefined yet again. What you see are the dazzling figures on funding leaderboards, one wave of model updates after another, and posts in your social feed about who has gone to an open-source community and harvested another wave of stars. Yet behind these phenomena, what truly determines success or failure is often not the technology itself, but a rare and very real entrepreneurial mindset: pushing oneself to the limit while remaining clear-headed enough to build systems and plan for the future.
That was the deepest impression left on me by Cheng Zeyi, the 27-year-old founder of WaveSpeedAI.
Even “996” Is No Longer Enough
“From 9 a.m. to 12 midnight, 6–7 days a week.” When I casually asked about his working hours, I did not expect that answer. The pace is even more intense than “996,” but this is not something he cites to show off. It reflects the interplay of compulsion and choice while living inside a technological torrent. Cheng Zeyi is not working hard merely for the sake of working hard. He is constrained by the reality that today’s AI infrastructure remains immature and is therefore forced to handle things personally.
Yet he is not discouraged. He is even looking forward to the future: once the interface for AI coding truly reaches the mobile-phone level, he will be able to work out while conducting “patrol-style” work, letting an Agent run processes automatically while he only confirms the direction every ten minutes. He believes the future of work will no longer mean sitting at a computer for ten consecutive hours, but acting like a commander and dancing with AI.
Has AI Made Work Easier? The Answer May Be the Opposite
AI has indeed freed us from inefficiency, but it has also quietly raised the standard for what it means to be a “worker.” In the era of AI coding, being able to write code is no longer rare. What matters is whether your code is sufficiently elegant, callable, and useful as a reference. Cheng Zeyi’s standard is this: can your code become a prompt for AI, and can AI learn from, inherit, and extend it? If not, then you are merely “writing,” rather than “shaping knowledge.”
In the traditional engineering system, backend, algorithms, frontend, and data were clearly separated modules. In the face of AI coding, however, the truly valuable person is an all-rounder. As Cheng Zeyi puts it, AI forced him to evolve from a backend engineer into an extreme high performer who can write algorithms, build orchestration systems, and even modify open-source models.
This is not simply about doing everything oneself to save money. It is an extreme pursuit of efficiency when company resources are limited and the market moves quickly. A company with US$10 million in revenue and only around a dozen employees does not achieve that through thrift, but through systematic reconstruction. He uses a Monorepo to centralize the codebase and make it easier for AI to participate; low-code platforms to build workflows so that non-core personnel can contribute effectively; and a developer-friendly service system to refine the product from merely usable to genuinely easy to use.
Entrepreneurship Is Not About “Finding the Next Big Trend,” but Making Your Niche Internally Coherent
In the past, we often imagined AI entrepreneurship as a race to see who could train a stronger model first and seize the technological high ground. Cheng Zeyi clearly does not see it that way. He emphasizes not “being stronger than you,” but “being closer to the user than you.” WaveSpeedAI does not buy GPU hardware or burn capital on heavy assets. It earns users’ trust through finely tuned APIs, a Discord community with round-the-clock responses, and in-depth customer research.
In his view, real differentiation does not mean developing a few more features. It means ensuring customers “no longer need to understand the technology” when using your product. If you save them the time required to understand documentation and debug an API, you are the best possible infrastructure.
This even overturns how many people think about “open source.” He acknowledges that in the era of AI coding, large amounts of code can already be generated by models, so the value of traditional open-source repositories is declining. What, then, is the new open source? It is the open-sourcing of ideas, workflows, and the creative layer. It is the kind of openness that lets others build directly on top of you, rather than merely offering fragments of code.
Can You Go Deep and Build While Standing on the Crest of the Wave?
At the end of our conversation, I asked him: what kind of company do you want to become?
He did not offer a grand slogan about “changing the world.” Instead, he said that when anyone developing image or video products anywhere in the world thinks about the underlying foundation for AI generation, he wants WaveSpeedAI to be the first name that comes to mind.
He said it calmly, yet the idea carried tremendous force. It is the ultimate question every entrepreneur who wants to build something in the AI era must confront: are you staking out territory on top of the wave, or constructing the foundational position in people’s minds from below?
In this unprecedented technological transformation, the real barrier may be neither the most eye-catching funding figure nor the model capability at the very front of the race. It may be whether you can use “taste” to refine a system, use that system to empower customers, and use customers to validate the direction.
AI has accelerated the world, but the people who ultimately remain are still the builders. A builder’s taste, resilience, and rhythm are the true moat of this era.
Sometimes, one person’s extreme drive is not about showing others how hard he works, but about his refusal to hand over the steering wheel. While everyone else wants to “use AI to do less work,” he is more interested in “using AI to do the right work.”
That may be the form of relentless drive most worthy of respect.
Selected Interview Q&A
The Founder and His Working Life
Q1: As the founder of WaveSpeedAI, what are your current workload and working conditions like?
Cheng Zeyi: I am currently “overloaded.” I work from 9 or 10 in the morning until 12 midnight, 6 to 7 days a week. I believe this working pattern needs to change, and my last physical examination was “already a little frightening,” so I need to spend more time exercising. One reason for these long hours is that “AI coding technology is not yet mature.”
Q2: How do you view AI’s impact on programmers’ career development, particularly the so-called “age-35 crisis”?
Cheng Zeyi: The emergence of AI has actually made experienced programmers more valuable. In the past, the industry commonly believed that programmers over 35 lost value because their physical stamina declined. Now, however, AI makes programming “more dependent on experience,” so veteran engineers can create more value precisely because they have accumulated that experience. At the same time, AI has empowered engineers: it turned me from a developer focused on backend work into an extreme high performer capable of handling the entire full stack.
Company Operations and Strategy
Q3: What are WaveSpeedAI’s team size, revenue, and revenue-per-employee performance?
Cheng Zeyi: The company currently has a lean team of around a dozen people, with operations staff accounting for a relatively large share. Its annual recurring revenue, or ARR, has reached US$10 million, and ARR per employee is approximately US$1 million. This is an exceptionally high level of productivity.
Q4: With only around a dozen employees, how has WaveSpeedAI achieved such high productivity and revenue?
Cheng Zeyi: I personally handle a large amount of full-stack development, and this is one of the company’s core competitive strengths.
Workflow automation: Internally, the company uses low-code platforms such as N8N to turn parts of its work into standardized workflows, allowing non-core developers to participate effectively.
Advanced development architecture: The company uses a “monorepo” architecture, placing all code in a single repository. The advantage is that both an AI Agent and developers themselves can retrieve code very easily, dramatically improving collaboration and automation efficiency.
Q5: How does WaveSpeedAI acquire users globally and maintain a low churn rate?
Cheng Zeyi:
SEO and open-source acquisition: The company places great importance on SEO, or search engine optimization. In its early stage, the founder added a link to the official website from his well-known open-source project repository, which generated about half of the website’s initial traffic.
Hard-core customer service: Service is the engine of growth. The company provides hard-core 7x24 responsiveness and close, hands-on service to customers of every size.
Customer support that exceeds expectations: WaveSpeedAI’s user churn rate is extremely low. One distinctive practice is that we proactively study customers’ products and help optimize them, reflecting an exceptionally strong customer orientation.
Product and Market
Q6: What is WaveSpeedAI’s core competitive strength, and how does it differentiate itself from original model providers and other platforms?
Cheng Zeyi: The core strategy is “differentiated competition.” Even the same model can be sold as a differentiated offering.
Flexibility and stability: Compared with other platforms, WaveSpeedAI’s API is more flexible and stable, and can be embedded easily into automated workflows.
Breadth of product categories: The company not only provides mainstream models, but also offers many functions unavailable on other platforms, including digital humans, video super-resolution, and watermark removal, while supporting open-source models as well. This is an important channel through which it converts SEO traffic.
The goal of surpassing original providers: The company aims to become “more original-provider-like than the original providers themselves.” It takes open-source models that might be regarded as “waste” and, through its technology and services, extracts their commercial value until they become products usable at a “Lamborghini” level.
Q7: What kinds of customers does WaveSpeedAI primarily serve, and how is it different from platforms such as Pika and Runway that directly serve creators?
Cheng Zeyi: WaveSpeedAI is more Developer-Friendly, serving teams that want to develop applications with integrated AI functions. Unlike Pika and similar platforms, which mainly serve end users or Creators, WaveSpeed AI provides underlying API capabilities. It can lower customers’ development costs because, after using its services, customers can “save the cost of their own developers.”
Q8: What is your view on startups adopting an asset-heavy model, such as buying their own GPU hardware?
Cheng Zeyi: For a startup, taking an asset-heavy route is “extremely dangerous.” WaveSpeedAI has therefore never purchased heavy assets such as its own GPU hardware since it was founded.
The Future and Advice
Q9: What specific advice would you give small teams hoping to start businesses in the AI era?
Cheng Zeyi: “Taste” is important. Here, “taste” specifically means building a high-quality internal company code repository. Entrepreneurs need to make sure that “the very first line of code is beautifully written,” and organize strong practices and workflows into SOPs, or standard operating procedures. AI programming tools can then learn and work from these high-quality “examples” and SOPs, completing tasks efficiently and to a high standard.
Q10: What are WaveSpeedAI’s future plans and goals?
Cheng Zeyi: The long-term goal is to establish leadership in the minds of developers worldwide. We hope that in the future, when developers want to build with AI—or even when they ask ChatGPT for recommendations—WaveSpeedAI will be recommended to them. On the business side, the company will continue pursuing rapid revenue growth and plans to launch more products in vertical fields based on its existing capabilities, such as its already highly effective digital-human product. Ultimately, we hope to grow together with our customers and help them “grow into giants.”