Since November 30, 2022, when OpenAI in the US released ChatGPT, the AI application became popular all over the world and became one of the fastest-growing consumer apps in the world — according to DevDay, the inaugural developer conference, OpenAI has now attracted more than 2 million developers, including over 92% of Fortune 500 companies. Today, ChatGPT has around 100 million active users every week.

At the same time, it is showing great productivity to unlock potential, rapidly empowering multiple industries, and spawning a variety of new business models in a very short period of time.

Recently, Google, which is ready to fight back, finally released the “killer weapon” that has been rumored for a long time. It claims to be the most powerful universal AI model in history and surpasses GPT-4 — the multi-modal Gemini. It is the largest and most powerful Google model so far. Its display results in various fields such as text, video, and voice have crushed GPT-4, which is already very powerful.

It is easy to see that the development momentum of AI can already be described using terms such as incredible and beyond imagination, and the AI competition that has taken the world by storm and is in full bloom has begun to show a pyramid of multi-level comparisons and competition.

The upper tier is where tech giants compete for underlying technology, algorithms, and computing power. The further down the level, the closer to the user level, the AI applications used to solve actual pain points and problems.

After more than 200 days of running wild, from big models to AI applications, we finally have a clearer vision and answers.

As a large model of next-generation infrastructure, it itself does not directly generate value. The future of the intelligent era will not be the big model itself, but the systematization, scenario, and verticalization of the big model ecosystem.

In the AI era, for the business world, it is more about a story of cognitive disruption and mindset change.

According to the minimalist investment mentality proposed by the famous investor Mr. Li Guofei, only companies with a win rate of 95% or more are worth paying attention to.

What he said was simple and extremely correct. Most of the companies that were able to reach the final finish line had already fully demonstrated a sufficiently high win rate and advantage in the early stages.

Just like now, although many companies are promoting their close ties with AI, in reality, there are very few listed companies that actually make good use of AI technology to earn back “real money.”

In response to this situation, the author proposes a simple and practical two-tier screening criteria: 1) positive profit+higher profitability, and 2) the business model can be deeply integrated with AI.

From the above list of listed AI companies with large models and a good degree of integration, it can be seen that apart from Baidu, a well-known Internet giant, the emergence of Bairongyun has undoubtedly surprised most people.

For Bairongyun itself, it's probably just an ordinary thing that is logical and reasonable.

This low-key company, quietly betting on AI decisions, not only made it “wake up early in the morning” but also “catch up with the bazaar.”

01 Big Model's Profit Path: Thicker in the First Half and Wider in the Second Half

As we all know, the performance of large models is mainly affected by three key factors: algorithms, computing power, and data. Computing power, as a computing infrastructure integrating software and hardware, is generally obtained through the purchase of cloud services. On the basis of computing power, through extensive data training, AI has obtained powerful large-scale model algorithms. By combining stacking algorithms, computing power, and data, large models of fairly large-scale parameters can be obtained.

Therefore, we can see that many large model companies with huge parameters have sprung up in the market, but despite this, there are few that can actually achieve profit.

The most essential reason is that the value of a large model depends not only on the scale of the model's parameters, but also on the actual effect of the model and the adaptability of the scene. Simply put, the value of a large model depends on both model performance and “commercial depth.” This is the key for Bairong Cloud to be the first in the industry to achieve profit — the company not only has forward-looking technical reserves, but also continues to adhere to the “long-term principle” of creating commercial value.

Technology goes without saying much; every player's path is very different. For example, the AI intelligent voice robot (Chatbot) independently developed by Bairong Yunchuang was first commercialized in 2017. It is the same origin as ChatGPT's underlying technology. It is a generative AI content production method that can provide a “realistic” interactive experience in terms of interaction effects; its BR-LLM industry model is based on a deep learning transformer framework, combined with NLP, intelligent voice, etc., and can support 10 billion level of parameter training through deep fine-tuning.

As for how to create commercial value and achieve commercial depth, we have seen that each player has many different paths.

What is unique about Bairongyun is that it adheres to the core strategy of “land first, then expand”, focusing first on the depth of the industry, then on the breadth of the industry, and gradually expanding its business to a wider range of industries.

The first vertical industry to land was the financial industry. Financial institutions have conditions such as massive data, diverse business scenarios, and increased demand for digital transformation, making finance a natural AI implementation scenario.

02 In the first half of the big model, Bairongyun focused on “deepening the industry”

As one of the first technology platforms to understand and use AI for generation and decision-making, Bairongyun has accumulated rich know-how and high-quality data while deeply cultivating vertical industries such as finance. Compared to the trend of simply pursuing the scale of model parameters, the company cleverly combines technology and industry advantages and grafts them together onto the BR-LLM large model based on the deep learning transformer framework.

This large-scale model is highly suited to industry needs, is mainly driven by scenarios, and focuses on creating commercial value in actual scenarios. In the process of applying it to actual business scenarios, the big model can continuously learn and adapt to changes during implementation, and complete self-iteration.

Haitong Securities International stated in a research report on Bai Rongyun, “We believe that in the vertical field, model self-iteration in actual application scenarios is the key to large-scale commercial implementation of large models, rather than just focusing on various authoritative evaluations and rankings, and can reverse build a moat of the company's hierarchical AIGC capabilities.”

This positive cycle, driven by commercial implementation and self-iteration, has made Bairongyun build high barriers in the field of big financial models, making it difficult for competitors to break through.

At the same time, the company is also continuously reusing AI capabilities for a wider range of financial scenarios: from bank credit card business, to consumer loans, then to inclusive finance and wealth management; from the banking industry's retail system, to banking consulting, to visual management and the operation and cross-marketing of a large number of bank users... Deep barriers and highly reusable capabilities, these two major factors have created a reliable foundation for Bairongyun's continued profitability and laid the company's lower limits.

03 Entering the second half of the industry, what determines the upper limit of AI companies?

Feng Yong, vice president of Bairongyun, said that the company has always adhered to the corporate DNA of pragmatism and delivery of business value. “We have now implemented applications in various productivity scenarios such as digital employees, digital assistants, digital people, programming assistants, and self-service data analysis.”

He said that this kind of corporate DNA and commercial layout can be replicated. Products and services based on the “decision-making AI screening+voice robot access” model have been successfully applied to core business links such as marketing acquisition, repayment reminders, and customer visits in the financial industry. This is also applicable to many other industries. If success in the financial industry can be repeated in a new industry, the company's upper limit will not be limited.

In fact, broadening the industry has been the company's strategic focus in recent years. In the first half, Bai Rongyun kept the big model far ahead by continuously polishing and honing the big model by going deep into the vertical field early on. Today, ChatGPT has set off a global AI wave. The big model industry chain is progressing rapidly upstream and downstream, and the production efficiency of big models has also improved markedly, and the industry has entered the second half. At the beginning of the second half, the company is fully equipped to promote the big model in more industries, making the big model “deep and broad”.

Judging from the results, the company has moved from the financial sector to a more diversified industry and has successfully expanded from depth to breadth. Up to now, nearly one-third of the company's customers have come from fields other than finance. Customers include more than 2,000 non-financial institutions in the fields of e-commerce, automobiles, recruitment, travel, logistics, ticketing, takeout, and tourism.

Continued improvements in financial performance also indicate that the company is on the right track. In 2022, Bairongyun achieved total revenue of 2,054 billion yuan, an increase of 27% over the previous year; adjusted net profit of 294 million yuan, a sharp increase of 108%. In the first three quarters of this year, Bairongyun's revenue reached 1,983 billion yuan, an increase of 33% over the previous year. It is already close to the total revenue for the whole year of last year.

04MaaS and BaaS open the era of AI industry

In the last stage of artificial intelligence technology, model parameters are small and generalizability is poor. Usually, one scenario requires a specific model. It is difficult to apply customized methods to various long-tail application scenarios, so the production model is similar to a “handmade workshop style”.

In a new stage of development based on big models, the entire AI industry has entered the “industrial production” model. Unlike traditional AI development models, large models have high general capabilities after being pre-trained with massive data sets, and only a small amount of data can be fine-tuned to significantly improve results. This means that the production efficiency of large models has been greatly improved, and they can better face diverse and fragmented application scenarios.

It is worth noting that although many AI companies claim to focus on big models, in reality they are old wine in new bottles. They still apply the traditional business model, often corresponding to one model or even one customer per scenario. However, we need to see that new business models are rapidly taking shape, and some companies have already found the door to a new world.

MaaS and BaaS are two new ecosystems for new AI that currently have high expectations. Both may turn large models into services, greatly simplifying the usage process and application costs of complex models.

MaaS (Model-as-a-Serivce, model as a service) packages models as services so that developers do not need to worry about complicated model download, installation, and management, but can easily obtain model output results through an API interface; BaaS (Business as a Service) goes one step further in the MaaS business, providing customers with more convenient and faster integrated AI solutions.

For example, on Bairong Cloud's decision-making AI-driven MaaS cloud platform, customers can freely deploy various models through the MaaS cloud platform's standardized API according to their own inquiry requirements, including calling ready-made model products for direct use in industrial applications; or “fine tune” their own products based on the big model, and quickly evaluate users with KYC (know your customers) and KYP (know your products), and ease of use is greatly improved.

For Bairongyun, it is only necessary to develop a large pre-trained model, use it as a general platform, and then fine-tune it to provide more convenient and diverse services for various users in the AI field, quickly meeting the ever-changing needs of thousands of people.

In essence, MaaS and BaaS are highly in line with the “All For Everyone” concept. Simply put, compared to previous-generation AI products and services, MaaS and BaaS are easier to use and more affordable. These two services are more likely to be paid according to performance and usage. This flexibility and high performance make it easier for enterprises to try and accept procurement-related services.

At present, all commercial implementation of OpenAI is carried out in the form of providing MaaS services. Under the leadership of OpenAI, global tech giants have begun to invest in MaaS layout. Basically, these manufacturers are focusing on the general sector. However, to build a strong MaaS and BaaS (Business as a Service) ecosystem in a vertical industry, you first need to have an extremely strong underlying industry model, which is an extremely challenging task.

Feng Yong, vice president of Bairongyun, said that charging for results is the best business model, but this model requires very high product and technology requirements, and requires a deep understanding of customer needs and industry needs. Only by truly creating value for customers can we get a share of value. Therefore, MaaS and BaaS ecosystems, which are relatively mature and continue to be profitable, such as Bairongyun, are still scarce resources in vertical industries. This market is still a huge blue ocean market, waiting for those with abilities to enjoy it.

05 Conclusion

A new wave of AI is on the rise, and the competitive landscape of the industry in the past is being disrupted.

For the previous generation of AI companies, it had its own lucky and unlucky sides. The lucky reason is that the vast majority of these companies have not achieved profit. They can finally change their fortunes through the new technological revolution, and finally continue the new round of capital boom to “prolong their lives” through a new round of financing.

What is more unfortunate is that in the new era, the competitors they will encounter are new species such as Bairongyun, which has a better degree of integration with AI and has even achieved better internal circulation. The probability that they will break through and survive is imaginable.

However, among the companies that have emerged in the new generation, there are still few companies whose business models have worked or are even profitable. Whether they can escape the fate of the previous generation of AI companies is still unknown.

However, the easiest way to find the ultimate winner is to find players who have maintained a winning streak.