VC-Zero, Under 100 Employees, and $1 Billion in Annual Revenue — The Surge AI Approach to Entrepreneurship: Powering AI from Behind with a Focus on Quality
In 2025, one company stood out as a unique presence in the AI industry: Surge AI. Within just a few years of its founding, it achieved over $1 billion in annual revenue without accepting any external funding. Moreover, this success was not driven by flashy marketing or massive organizational expansion, but by a steadfast commitment to “data quality” and a “lean team,” choices that were consistently grounded and practical. Founder Edwin Chen, drawing on his own experience and philosophy, is rethinking where the future of AI should head. This article unravels the journey and philosophy of him and Surge AI, and why the “pursuit of quality” behind it is now determining the fate of AI.
1. What is Surge AI? — A Unique Startup Supporting the “Foundation of Data”
1-1. Background of Founding and Scaling
Surge AI was founded in San Francisco in 2020 by Edwin Chen. Chen had previously worked at major companies like Google, Facebook, and Twitter, but he says he repeatedly encountered the experience of “not being able to get data that was actually usable.”
While many AI startups threw themselves into the fierce competition for VC funding, Chen deliberately chose to “bootstrap” (founding with only his own capital). Without relying on fundraising, sales, or PR, he focused purely on the single point of “creating high-quality data.”
As a result, Surge AI achieved approximately $1.2 billion in annual revenue in 2024, less than five years after its founding. The number of employees is a small, elite group of about 110 (as of 2025).
This “small, strong, and quietly growing” strategy is in contrast to the VC-led “high-risk, high-return” corporate image, presenting a new option for how startups can exist in the future.
1-2. Services Provided — “Advanced Data Infrastructure,” Not Just Labeling
What Surge AI provides is not simple tasks like traditional “labeling images.” They generate and provide “high-quality data” that delves into the “nuance,” “human-like judgment,” “context,” and “safety” that natural language processing (NLP) and generative AI must essentially understand. Specifically, they provide the following services:
Dataset generation for RLHF (Reinforcement Learning from Human Feedback)
Data labeling by experts (legal, medical, tech, cultural domains, etc.)
Model “Evaluation,” “Safety testing (Red-teaming),” and “Content moderation”
Support for multiple languages and domains — Data that includes not just language, but also differences in culture, ethics, and systems
Through such “advanced and specialized data infrastructure,” they are contributing to the realization of “AI with human-like qualities,” which is essential for frontier AI labs.
2. “Quality is Everything” — Edwin Chen’s Philosophy and Values
2-1. Toward “Quality,” Not “Quantity” — The Resolve to Go Beyond GIGO
Chen says the following about the data that supports AI: “If the data is garbage, the model is garbage.” However, what Surge aims for is not just garbage removal. Rather, it is data that teaches AI complex and subtle nuances such as human language, thought, humor, ethics, and creativity. This cannot be measured by “checklists” or “presence of tags.” When citing poetry or literary works, he said:
“If you have a model write a poem about the moon—evaluating it as a poem just because it includes the word 'moon' in eight lines is checklist-like, and that is not the poetry we are looking for. What matters is whether it moves the heart when read, whether it can depict the transience of moonlight, whether the words pulse, whether it stirs emotions, whether it prompts thought—that is what matters.”
To capture such high-level “goodness,” simply creating massive amounts of data through human-wave tactics is never enough. Surge AI ensures the “depth” of its data by using workers with experience and expertise, as well as multi-dimensional human evaluation metrics.
2-2. The Polar Opposite Strategy for Startups — Avoiding VCs and Focusing on Value
While many startups expand their scale in the name of fundraising, hire massive numbers of employees, and grab attention through PR strategies, Chen has intentionally avoided that path. 'If you're raising money from VCs just to expand wastefully, you could fire 90% of your staff,' he says. 'With the best elite team, you can move much faster.'
This stance is not just a philosophy; it is backed by results. Delivering maximum output with the bare minimum of resources—this is truly 'lean growth.' As a result, the company has remained profitable almost consistently since its founding.
3. Why 'Quality' Matters Now — The Limits of AI and Questions for the Future
3-1. The Dangers of Benchmark Supremacy — A Warning Against 'AI Slop'
In recent years, the superiority of AI models has tended to be discussed in terms of benchmark scores. However, Chen views this as dangerous. According to him, many benchmarks are merely 'single-shot tasks' with clear, objective correct answers, and models optimized for that format are weak when faced with the complex, ambiguous reality of the real world.
Furthermore, this 'leaderboard supremacy' fosters a trend of guiding models toward 'flashy, crowd-pleasing responses.' Techniques like 'adding more emojis,' 'making markdown and visuals flashy,' or 'using long-form text to hook readers' are often prioritized over utility or accuracy—Chen calls this 'AI slop' and sounds a strong warning against it.
3-2. The Importance of RL Environments — The Path to 'Learning AI' Like Humans
Surge AI emphasizes not just providing static data, but also RL (Reinforcement Learning) mechanisms that place AI models into real-world-like 'environments' to let them learn through trial and error. For example, they train models in situations close to real-world operations, such as using tools to clear a task or taking incremental steps.
Learning that emphasizes such 'trajectories' trains the quality of a model's thought and decision-making processes, which cannot be captured by single-shot benchmarks. To cultivate essential AI capabilities—such as long-term task management, decision-making under stress, creativity, and ethics—these 'human-like learning processes' are indispensable.
4. What Surge AI Represents — Its Value as the 'Hidden Support' of the AI Industry
4-1. The Unsung Hero Behind Frontier AI Labs
Surge AI's clients include the world's most advanced AI labs, such as OpenAI, Anthropic, Google, and Meta. While these labs grab the spotlight for their large-scale models and computing power, the quality of their 'intelligence' is supported by the data quality and evaluation infrastructure of behind-the-scenes companies like this. In short, as the 'unsung hero' of the current AI boom, Surge AI is an essential presence.
4-2. A New Path for Startups and AI Companies
Surge's success has demonstrated a different way of entrepreneurship: instead of 'raising funds to scale' or 'gathering market attention,' it is about 'quietly accumulating value that the world truly needs,' 'respecting human intelligence, ethics, and culture,' and 'maintaining a lean, elite team.' This is a suggestion that many startup entrepreneurs, AI researchers, and society at large cannot afford to overlook.
Conclusion — The Hidden First Line Supporting the 'Future of AI'
The story of Surge AI and Edwin Chen is not about flashy fundraising or millions of lines of code, but about returning to the essence of AI: 'quality and reliability,' 'humanity,' and 'sustainable growth.' As AI increasingly becomes the foundation of society, the existence of such a 'hidden first line' will become even more important in the era to come.