Produced by | RoboIsland

In May 2026, 01.AI marked its third anniversary. Li Kaifu forwarded an all-staff letter on his personal WeChat official account with a long and straightforward title—"To the 01.AI Team: Being the First Profitable AI2.0 Company and Leading You to Win Matters More Than Any Valuation." (The original content had been deleted as of press time.)

The letter opens with a story spanning 45 years. The young man who first delved into artificial intelligence in his doctoral application letter has now become the helm of his third entrepreneurial venture. He believes AGI is a once-in-many-lifetimes opportunity and the clearest mission of his current life.

The letter explains the company's name, 01.AI, is derived from the Dao De Jing: "The Dao gives birth to One, One gives birth to Two, Two gives birth to Three, Three gives birth to all things."

Reading this letter, it's hard not to be moved by its emotional intensity. A year and a half ago, during the darkest moment, external criticism was relentless. A close friend advised Li Kaifu to cut the team to ten people, suggesting that the funds on hand could sustain a laid-back existence for over a decade. Li Kaifu felt that quitting would be like turning off the lights just before dawn.

During that time, a third of the company's workstations stood empty. On the first day of the Lunar New Year, Li Kaifu stood in the deserted office and said to his colleagues nearby, "Please look at me and give me a smile."

This letter is not just a look back but also includes self-critical adjustments and tangible commitments.

01.AI's current positioning is clear: an AI implementation company expanding from China to the global market, benchmarking against Palantir, with the goal of securing real contracts from major clients. The letter even proactively offers a new label: "Don't call us one of the 'Six Little Tigers'; call us the 'Money Leopard.'"

The company simultaneously announced three incentives: issuing an additional 20 million stock options, initiating the selection of the first batch of partners, and establishing a CEO-specific incentive program starting at 1 million yuan with no upper limit, heavily rewarding teams that secure the largest contracts, highest gross margins, and best annual recurring revenues.

This is not a conventional corporate anniversary letter but more like a pre-battle mobilization. Between the lines, there is a recalibration of the company's positioning, a clear-eyed restructuring of the team, and a determined resolve to pull itself out of the mud and charge forward once again.

Anyone who reads this letter will find it hard to deny the belief and resilience of Li Kaifu. Personally flying to Africa to meet clients, descending into mines on New Year's Eve, and after peel off (divesting, context suggests stepping away) from his role as an AI evangelist, he immersed himself in the industrial depths for three years. Such flexibility and endurance are rare among entrepreneurs.

The letter is touching, but it is, after all, just a letter.

Li Kaifu states that 01.AI's growth does not rely on loss-leading volume but on high-gross-margin, high-repurchase 2B ARR, claiming that in a certain quarter next year, it "will become the first AI 2.0 company in China to achieve profitability."

These sentences carry weight on paper, but what truly matters is not the wording but the data in the upcoming quarterly results, project-by-project progress, and renewal contracts.

Can an AI company that no longer owns its foundational models build a competitive moat through customer relationships and delivery capabilities? Can the high-labor model of deeply serving major clients maintain healthy profit margins during scaling? Are the co-creation system and DRI paradigm proposed in the letter the true starting point for organizational transformation or just temporary measures in a predicament (predicament, context suggests a difficult situation)?

These questions are given direction in the letter but have yet to be answered.

I. The Shelf Life of Idealism

When 01.AI set out, everyone was looking in the same direction.

In May 2023, the large model craze ignited by ChatGPT swept the globe, and China's venture capital circle quickly assembled a star-studded formation to rival OpenAI.

01.AI was among the most haloed. Li Kaifu, with his triple identity spanning academia, industry, and investment, assembled a team of top talent from Google, Microsoft, and Alibaba. Just eight months after its founding, its valuation exceeded $1 billion.

At that time, 01.AI indeed lived up to such expectations. In November 2023, it released the Yi-34B pre-trained model supporting a 200K context window, topping the HuggingFace open-source model rankings; in March 2024, it open-sourced the Yi-9B model with outstanding coding capabilities; in May of the same year, it launched the Yi-Large closed-source model with hundreds of billions of parameters; in October, it released the lightweight flagship model Yi-Lightning, ranking sixth globally.

As parameters grew larger and rankings climbed higher, the team's confidence soared. An early member recalled a sense of excitement in solving problems, believing that once technical benchmarks caught up with international leaders, the rest would fall into place.

In October 2024, when Li Kaifu was asked in public whether he would persist with self-developed large models, his response was firm: "We will never abandon pre-training." At the time, this was interpreted as a dual commitment to the team and investors.

But commitments have a shelf life.

Li Kaifu later explained that as a startup, competing with tech giants in burning money would never succeed. Training a trillion-parameter-scale super-large model once could cost tens of millions of dollars in computational resources. Moreover, the open-source community's catch-up speed far exceeded expectations.

Coupled with the formal release of DeepSeek-R1 in January 2025, which performed close to leading closed-source models in multiple benchmark tests, the moat around foundational models effectively became a public pool.

In early 2025, Li Kaifu made a decision that few entrepreneurs would dare: while peers were still climbing, he was the first to turn and head down the mountain.

He merged most of the pre-training team into Alibaba Cloud, shifting to train faster and cheaper medium-parameter models to build profitable applications. However, he kept a small flame of pre-training alive; 01.AI did not entirely abandon it but no longer pursued super-large models.

II. Can 1.5 Billion in Orders Buy Back Valuation?

After descending the mountain, Li Kaifu chose a path almost entirely opposite to his previous approach.

At 01.AI's inception, he had explicitly favored the To C market and firmly refused to do loss-making To B business. But by March 2025, the strategy was redefined as a full commitment to To B, shifting the business focus from chasing parameters and rankings to securing clients and projects.

Li Kaifu himself also transformed. No longer frequently appearing at industry forums and ranking discussions, he boarded planes to Africa to secure orders, defining his role as Chief AI Strategist—not just selling products but helping enterprise clients calculate: Which number in your financial report can AI improve?

He called this approach the "First Leader Project."

Because AI transformation is not a conservative, error-free endeavor but a risky change in a company's fate that only a first leader can undertake—a founder personally engaging, deeply understanding client businesses, and helping rewrite their profit statements.

Orders did indeed surge. Starting from over 100 million yuan in revenue in 2024, orders reached 500 million yuan and 250 million yuan in audited revenue in 2025, then exceeded 1.5 billion yuan in the first five months of 2026, forming a steep upward curve.

But to assess the true quality of these 1.5 billion yuan in orders, one must look beyond the total and examine the structure.

From the client mix, large enterprise clients and government projects contributed the bulk of the orders. Li Kaifu mentioned in interviews that there are currently over a hundred clients, with revenue primarily driven by strategic major clients, including more than one order exceeding 100 million yuan.

The company also undertook sovereign AI projects in places like Kazakhstan, providing overall solutions for AI infrastructure to national and regional governments, which he described as a major growth area.

From the product foundation, 01.AI draws inspiration from the U.S.-based Palantir, building digital foundations for enterprises through deep industry understanding.

The company's WanZhi Enterprise Large Model Platform adopts a "horizontal platform, vertical industries" approach, co-creating scenario-based solutions with key clients in government, finance, and industrial sectors, then refining industry know-how to feed back into the platform. The goal is to build an enterprise-grade AI operating system for standardized, scalable AI implementation.

But a critical question lurks: Will deep service become a barrier to scaling?

Currently, each major client order requires a high density of manpower, from algorithm engineers stationed on-site to business process research and data integration, all demanding the company dispatch teams to dive in.

The core logic of this approach is to exchange service depth for client stickiness. The idea itself is sound, but how much additional manpower is needed for each new 100-million-yuan order? Under the hands-on service model, each project must start from scratch, and the experience from one project may not directly apply to the next.

Thus, the true value of these 1.5 billion yuan in orders ultimately depends on one metric: whether the standardization level of the WanZhi Platform and the reuse rate of industry know-how can continuously improve.

If so, growth will gradually shift from stacking manpower to stacking products; if reuse rates stagnate, more orders will require more personnel, flattening profit margins.

III. Leopards Don't Need an Audience

For over a decade in China's AI circle, Li Kaifu has played the role of evangelist.

Writing bestsellers, delivering transcontinental speeches, and using the most accessible metaphors to depict how AI would change the world. With the public trust he built over these years, he established the initial valuation and team foundation for 01.AI.

But after 2025, the public saw a different Li Kaifu.

He hoped that 01.AI would no longer be called one of the "Six Little Tigers" but the "Money Leopard"—a telling signal.

When a figure known for technological idealism rebrands his company from tiger to leopard, he has largely redefined himself.

A tiger is a beast observed by all; a leopard must hunt to survive.

Everyone must confront the same question: Can AI make money?

Li Kaifu's judgment is that after consolidation in the large model track (track, context suggests the large language model sector), there will no longer be so many companies simultaneously building foundational models. The opportunities for small and medium-sized companies lie in applications, implementation, and helping clients truly earn money.

When delivering this judgment, Li Kaifu's tone was calm, as if stating a verified fact rather than a debatable viewpoint.

The three-year trajectory of 01.AI can be plotted as an intriguing graph: the vertical axis represents valuation and halo, the horizontal axis represents time, with a noticeable dip in the middle.

From 2023 to 2024, the curve soared upward, with valuations exceeding $1 billion, model rankings topping charts, and industry spotlights shining brightest; in early 2025, it suddenly plummeted, with the abandonment of super-large models nearly unanimously interpreted by the outside world as falling behind; then, starting in late 2025, the curve quietly rose again, carrying 1.5 billion yuan in orders, a goal of breaking even, and a 2027 IPO plan, returning to center stage.

Behind this rise and fall looms a larger question for 01.AI: Where is the competitive moat for an AI company that does not own its foundational models?

Li Kaifu's answer: customer relationships, industry know-how, end-to-end delivery capabilities. And one metric he cares about most: renewal rates—clients must be willing to renew to truly trust you.

During the AI1.0 era, a batch of star companies burned through financing and collapsed before reaching profitability. Li Kaifu vowed to break this curse, and now, less than two years remain until his self-imposed validation deadline.

Can the WanZhi Platform truly function like an operating system, plug-and-play? Are the manpower costs for each new major order increasing or decreasing? How are renewal rates faring among the first flagship clients? These questions remain up in the air.

Descending the mountain is not the endpoint but the starting point of another journey. On this path, there are no rankings to endorse you, no parameters to support you—only judgments that must be validated with real money.

The clarity of taking an early step forward may be harder than persistently struggling with oxygen deprivation at the mountaintop.

Li Kaifu made a less glamorous choice, and how this choice will ultimately be defined will not be found in any report or press conference.

The answer lies in clients' renewal contracts.

RoboIsland. All rights reserved. No reproduction without authorization.