問AI· Yang Zhilin cleverly used FDE to break the revenue fluctuation before going to Hong Kong?
“Lock in” big customers and bulk revenue as much as possible,It is becoming a new target after Dark Side of the Moon is rumored to be released.
On September 10, Dark Side of the Moon launched the Kimi Enterprise Partner “Moonshot Plan” and will cooperate with system integrators such as ChinaSoft International, Kingsoft Cloud, Huasheng Tiancheng, Yakang Technology, AsiaInfo Technology, etc. to build a team of frontline deployment engineers (FDE).
FDE is one of the most popular to B business models in the current AI industry. It refers to companies sending engineers to customer companies to go deep into the work site to find out business needs, and then deploy products, including large models, into the company’s core business processes to make products more usable and customers better at using them.
Regarding the “Moon Landing Plan”, Dark Side of the Moon is quite high-profile, claiming to be “the first large-scale model company in China to use the FDE model for AI terminal deployment.”
The timing of Dark Side of the Moon’s aggressive deployment of FDE coincided with rumors of its listing in Hong Kong.
At the beginning of this month, market news said that Dark Side of the Moon planned to IPO in Hong Kong this year, raising up to US$5 billion. A few days later, the “Moon Landing Plan” came out, reflecting Dark Side of the Moon’s intention to stabilize mid- to long-term income.
As one of the “Six Little Dragons of AI”, Dark Side of the Moon’s current core business model is built around cloud MaaS. It is not fundamentally different from the already-listed Zhipu and MiniMax – they both sell tokens, and they sell well and will continue to do so.
The problem is that although selling tokens is a good business, the competition is very fierce.
At this stage, based on token consumption and revenue contribution, the major customers of Dark Side of the Moon and other companies are mainly various types of AI programming, agents and office platforms, such as domestic Byte Coze/Trae, Alibaba Qoder, and Tencent CodeBuddy, foreign OpenCode, Cursor, Windsurf, etc.
These platforms are connected to a variety of mainstream models. The endless emergence of new models with soaring performance at home and abroad, the frequent adjustments of token prices, and even the special optimization of new models for certain computing power and certain types of tasks may allow users to recalculate their accounts and switch between different models with one click.
During the period of rapid expansion of AI services, the supply of tokens from leading model companies exceeds demand. In addition to the strong demand from large Internet companies and large platforms, model aggregation and distribution platforms such as OpenRouter also bring a steady stream of small and medium-sized enterprises and individual users.
But as global AI computing power continues to expand, one day there will no longer be a shortage of tokens. If model companies continue to sell tokens to Internet platforms and small and medium-sized customers as they do today, but cannot influence end users to “choose him or me,” they are likely to face fluctuations in revenue and performance.
For Dark Side of the Moon, which is rushing to go public, this potential fluctuation, although not yet apparent, also poses a long-term risk.
In this context,By focusing on FDE, we can seize big customers and old money who can afford big bucks and are not willing to replace IT infrastructure casually.It has even become the “Chinese version of Palantir”, which is the latest problem-solving method found by Dark Side of the Moon.
A
More than a month before the announcement of the “Moon Landing Plan”, Dark Side of the Moon advanced rapidly on the FDE track.
First, on July 20, Chinasoft International and Dark Side of the Moon signed an agreement. The two parties will jointly build an FDE innovation laboratory to jointly carry out the definition, research and development, delivery, polishing, testing and continuous iteration of enterprise-level Agent products. The first phase of cooperation will prioritize the key industries of energy and power.
By the end of August, AsiaInfo announced a cooperation with Dark Side of the Moon in the enterprise-level Agentic AI business.
The specific method is that Dark Side of the Moon provides complete Kimi model weights, technical documentation and back-end R&D support, and AsiaInfo establishes a dedicated FDE delivery team. Both parties jointly build an integrated delivery system covering model adaptation, parameter tuning, localized deployment, and continuous operation and maintenance.
ChinaSoft International and AsiaInfo Technologies both appeared on the first list of the “Moon Landing Plan”. Dark Side of the Moon said that the plan will continue to be open to IT service providers, software service providers and consulting organizations.
When Dark Side of the Moon does FDE, its range of motion is significantly greater than that of other “little dragons” and even surpasses that of major manufacturers.
Compared with the grand feast of Dark Side of the Moon, Friends is still in the team building or even wait-and-see stage.
Zhipu released the job requirements for FDE leaders, with a monthly salary of 60,000 to 80,000 yuan, which is average in the AI circle and cannot be compared with the generous salaries of top research institutes. MiniMax, Baichuan, Zero One Thousand Things and Stepping Stars have no public actions.
至於 DeepSeek, his ambition is not here, “Liang Sheng” obviously looks down on FDE and other mundane jobs like digging in the dirt and working for others door-to-door.
Major Internet companies have gone deeper than “Little Dragon”, but they have only taken a few more steps.
Byte had previously recruited FDEs, and the annual package was rumored to be up to 1 million yuan. Alibaba Cloud began recruiting FDEs in Beijing, Shanghai, Guangzhou, Shenzhen and Hangzhou in April this year, but only recruited 10 people, which clearly means testing the waters.
On the contrary, Tencent is slightly more interested. On the same day that Dark Side of the Moon landed on the moon, Tencent Cloud launched the “industry’s first FDE engineer certification system” and also opened recruitment for FDE partners.
Overall, Dark Side of the Moon is the AI company with the deepest involvement and the highest profile in the domestic FDE circuit. The reason is not complicated:Compared with its peers, its willingness to make money is much stronger.
The Dark Side of the Moon, which is in the process of going public, needs both a powerful model like K3 and a beautiful financial report card.
Recently, according to people familiar with the matter, Dark Side of the Moon’s annual recurring revenue has soared from US$300 million in June to more than US$1 billion in August. The company plans to increase this to US$2 billion by the end of this year.
Dark Side of the Moon is already working on this, and one of its actions is to charge licensing fees for open source models. There are reports that it is negotiating with Microsoft, Amazon and Google for a 30% revenue share.
But distant water cannot quench near thirst. While negotiating with foreign giants, Dark Side of the Moon needs to open more springs.FDE, with its clear business logic and mature business model, has naturally become one of the key choices of Dark Side of the Moon.
B
The Dark Side of the Moon wants to make money from FDE, but it also has its own little ideas:It did not recruit its own troops, but hoped that “I would provide the weapons and you would provide the troops” and that everyone would share the money they made.
The Dark Side of the Moon is known for its streamlined team. Today there are still only about 300 people, 80% of whom are R&D personnel. In comparison, there are less than 1,000 people at Zhipu, 400 at MiniMax, and 500 at DeepSeek.
When entering FDE, Dark Side of the Moon does not intend to have a large number of resident engineers. Instead, it will hand over Kimi models and Agent bases to system integrators such as ChinaSoft International and Kingsoft Cloud. The integrators will provide on-site personnel while Dark Side of the Moon only provides technology and products.
The advantage of this is that it does not need to set up sales and delivery teams across the country, nor does it need to spend energy to open up and sort out contacts. Instead, it can use the personnel and relationships of established integrators to expand its business and stay light during the launch cycle.
However, this is not too “tricky”. Actually,OpenAI and Anthropic, which seek to delve deeper into FDE, have also adopted a similar approach.
OpenAI has been coveting the FDE cake for a long time. In February this year, it launched the Frontier Alliance plan and attracted global consulting giant Capgemini as a partner. Capgemini provides a large number of certified engineers to work with OpenAI’s own FDE for customers.
On May 11, OpenAI established a new company, OpenAI Deployment Company. This company is controlled by OpenAI and has attracted 19 investment institutions including TPG, Advent, Bain Capital, and Brookfield. It has an initial capital of US$4 billion and a pre-money valuation of US$10 billion.
As soon as the new company was established, it acquired an AI consulting company called Tomoro and “recruited” 150 senior FDEs as a whole to make it their own.
Anthropic competes for the FDE market and moves even faster than OpenAI.
Just a week before the establishment of OpenAI Deployment Company, Anthropic roped in three Wall Street giants – Blackstone, Hellman & Friedman and Goldman Sachs – to invest US$1.5 billion to establish a joint venture. The new company focuses on AI-native enterprise services, embedding engineers into the core business of client companies, and deploying Claude.
Three days later, Anthropic reached a strategic cooperation with PricewaterhouseCoopers, with the latter providing a large number of certified engineers.
This forms an interesting phenomenon:When an AI company does FDE, once it gets a little deeper, it has to bring in old money to start the business together, or even set up a new company together.
Money isn’t the only reason. Even though Dark Side of the Moon has limited ammunition and is unwilling to spend too much money, Factory O and Factory A, which has repeatedly received huge financing, will never be tied up by a few billions, let alone form alliances with others just for this little money.
The real reason is that consulting firms and financial institutions that have been established for decades or even hundreds of years are better able to understand the needs and ideas of traditional large companies in finance, energy, transportation and other fields. The interpersonal network accumulated over many years also gives them the right words and budget to leverage when selling FDE solutions.
In addition, old money has a brand effect that technology companies cannot match, and large companies are very fond of this – choosing which companies to cooperate with is itself a reflection of the strength of large companies. In contrast, no matter how advanced the technology and cool the products of a new company are, they may not even be able to get in through the door of the IT director.
On the other hand, established companies have served large customers for many years and have a team of experienced field engineers. They know the know-how of on-site service, which is the essence and barrier of FDE; even if they don’t understand large models, it is just a technical shortcoming that can be solved through internal collaboration.
Old money in traditional fields is eager to catch the fast train of the AI era, and AI companies are also happy to bring old money in to make money from large companies in traditional industries.FDE at home and abroad has followed a similar path.
C
By doing FDE, Dark Side of the Moon has the opportunity to answer a question:Is there any user loyalty in the “token economy”?
Currently, model companies sell tokens through APIs, which is the main form of the so-called “token economy”. However, the supply side has always been troubled by low user loyalty: no matter how many free credits or recharge discounts are given in the early stage, users will be attracted away once they see the new “Xiao Tiantian”, and they themselves will become a helpless “Mrs. Niu”.
The ultimate solution for the model company to break the situation is to come up with a model that is so powerful that no one can match it, so that users will feel that this is the “dream model”, and it will last forever.
However, with the increasing convergence of language model technology stacks, it is normal for each company’s new model to “lead the trend for three to five days”. Even Factory O and Factory A dare not say that they have reached the top of the industry.
On the other hand, the methods and channels used by model manufacturers to sell tokens are highly consistent. Apart from adjusting prices, there is almost no room for differentiation. This also encourages users to use whichever is cheaper.
What’s more terrible is that as token distribution channels, AI application platforms and large model routing essentially do not encourage “loyalty”, but provide the greatest convenience for users to switch models at any time. Model manufacturers have almost no effective means to retain users.
FDE mode brings“Lock” token requirements and bind customers to model manufacturers的契機。
The unit price of the token has not changed, but the cost of customer churn has increased by an order of magnitude. This objectively forms “user loyalty”.
In addition, FDE can graduallyChange the user and revenue structure of model manufacturers, reach out to those traditional large companies, and get real long-term big orders.
The first five companies signed by the Dark Side of the Moon “Moon Landing Plan” mainly carry out business in industries such as communications, energy, electricity, finance, government affairs, and medical care. Their end customers include large state-owned enterprises such as State Grid, the three major operators, PetroChina, and banks, as well as various listed companies.
If Kimi can enter these enterprises, the customer structure of Dark Side of the Moon will have a chance to change qualitatively: in the past, it was mainly developers, Internet companies, and C-end individuals; in the future, it may mainly be central enterprises, major banks, operators, and energy giants.
Characteristics of this type of customer: huge contract amount, slow but stable payment, and extremely strong stickiness. Once a model company enters the supplier directory, it is difficult to replace it.
At this point, Dark Side of the Moon’s FDE ambitions have slowly unfolded.
But the challenge is also obvious: all these paper blueprints have not yet been verified.
Although Palantir has verified the feasibility of the FDE route, global Ai companies have yet to make big money from this area. Most domestic AI companies are still in the recruitment stage; Factory O and Factory A are making a lot of noise, but they have just embarked on their journey. Dark Side of the Moon is involved in FDE with a relatively clear goal; however, the specific direction cannot be solved by holding a press conference and bringing in a few partners, but requires a long period of exploration.
