Liberate productivity with AI infrastructure, and Bayesian computing penetrates deeply into the scientific research market.

Before the troops and horses move, the "rations" come first. In the era of large models, many practitioners believe that the server side of AI infrastructure may produce unicorns before the application side.

AI infrastructure is the intermediate layer connecting computing power and applications, including hardware, software, tools, optimization methods, etc. Therefore, regardless of which commercialization field AI applications can ultimately be verified in, AI infrastructure is indispensable for the intelligent transformation of all industries.

Especially in large-scale industrial scenarios such as scientific research and innovation, satellite remote sensing, and medical imaging, most choices will be one-stop model computing power deployment and development tool platforms, which need to be private and have standardized engineering.

OpenBayes, a Bayesian computing company, is such an AI infrastructure provider that offers ready-to-use services based on data set barriers, supercomputing clusters, and long-term engineering capabilities.

Recently, in the second season of the "Lenovo New Business Innovation Ecosystem Roadshow", Bayesian Computing was selected as one of the TOP 10 projects. Lenovo's Small and Medium Business Group's emphasis on innovation in the AI PaaS layer also reflects its consistent concept: to provide customers with one-stop services throughout the enterprise growth lifecycle and to progress hand in hand with customers.

As the core carrier for Lenovo Group to play the role of the chain leader and drive the integrated innovation of start-up enterprises, the Lenovo Star Program for Start-up Enterprises has built a three-dimensional empowerment system of technology enablement, industrial collaboration, and capital acceleration. Start-up enterprises that join the Lenovo Star Program will have the opportunity to receive all-round and multi-dimensional support from Lenovo Group, the "chain leader", including but not limited to co-creation in R & D, ecological opening, capital support, business expansion, brand and service enablement, etc. Since the launch of the "Star Program for Start-up Enterprises", it has successfully helped hundreds of start-up enterprises accelerate their development and has become an important link for Lenovo and small and medium-sized enterprises to progress together and empower each other. Under the era proposition of accelerating the cultivation of new productive forces, the Lenovo Star Program for Start-up Enterprises will continue to expand the recruitment scale of ecological partners, looking forward to building a value symbiotic community with more industrial innovation forces and jointly climbing to new heights in global technological competition.

The Cornerstone of the AI Era

In 2017, the year when Bayesian Computing was just established, big data technology also entered a booming engineering stage.

"At the beginning, we mainly provided overall solutions for AI from deployment to development for Internet and industrial enterprises. But before the explosion of AIGC, we found that many enterprises and institutions with such layouts were in fields that emphasized scientific research and innovation."

Wang Chenhan, the founder and CEO of Bayesian Computing, told 36Kr that the initial entrepreneurial experience allowed them to accumulate data processing experience for scientific research and innovation groups and find many scientific research-oriented customers who were willing to explore and grow together in the AI field.

The China Merchants Group and the School of Computer Science at Peking University are two good examples. These institutions themselves have a large number of innovation projects, high-quality data, and interdisciplinary talents. However, due to insufficient computing power resources, it is also difficult to control the costs of hardware, operation and maintenance, energy consumption, etc. required for innovation projects. Therefore, rationally allocating resources through the infrastructure services of Bayesian Computing has become a hard requirement.

In the era of the "cloud", enterprises need cloud-based services to process data, develop data applications, and open applications to end-users. And AIGC has pressed the accelerator button for the classic cloud-native architecture.

Currently, mainstream AI infrastructure services basically cover all tools and processes related to large model development and deployment. If AI applications and computing power are the two slices of bread in a sandwich, AI infrastructure is the rich filling between the "computing power" and "application" slices of bread.

In the scientific research field that Bayesian Computing "specializes in", the filling needed for this AI sandwich mainly includes data processing, model training, private deployment, etc. The highlight of Bayesian Computing itself lies in its feature of being ready-to-use without environment configuration, which can adapt to small-scale experiments in innovation projects.

In the scientific research market, another pain point is that there are relatively few talents with both industry knowledge and AI technology, and the training and cooperation mechanism for interdisciplinary talents is not yet perfect. Just in the step of data annotation, the talent cost is relatively high, and automated and engineered products can bring considerable efficiency improvement and cost optimization. Bayesian Computing is also making efforts in this regard.

In cooperation with more scientific research institutions, domestic universities, and leading industrial enterprises, Wang Chenhan further felt that the combination of basic scientific research and high-performance computing is becoming an irreversible trend.

AI for Science is also one of the relatively blue ocean core scenarios in the fields that Bayesian Computing has explored and contacted at present.

To adapt to this trend, Bayesian Computing has carried out reconstruction on the technology and product sides, focusing on instruction set migration, model structure search, and computing efficiency optimization, and clarified its positioning as a "scientific computing platform".

In 2022, after initially polishing the product matrix, Bayesian Computing officially built a product and operation team for the needs of the scientific research market. Currently, its products have reached dozens of leading domestic universities, scientific research institutes, and state-owned enterprises, and its flagship data science search engine "Hyper.AI Super Neural" has also received certain recognition.

"AI infrastructure services that can be compatible with mainstream computing software and have price competitiveness have always been of great value to the scientific research market."

Wang Chenhan believes that compared with large companies, the entrepreneurial team may have a more in - depth understanding of the needs of scientific research customers and provide more timely feedback.

From the beginning of entrepreneurship to now focusing on the niche field of education and scientific research, Bayesian Computing has been adhering to the same commercial positioning: using AI to solve technical problems, standardizing products or businesses as much as possible, and at the same time enabling customers to see the ROI of cost reduction and efficiency improvement.

Moving towards Ecological Connection

Before participating in this roadshow, the Bayesian Computing team had been paying attention to Lenovo's business and commercialization ecosystem.

As a leading industrial manufacturer with close cooperation with many universities and scientific research institutions, Lenovo itself is a potential customer that Bayesian Computing hopes to reach. In addition, since one of the core businesses of Bayesian Computing is to provide solutions for customers' HPC (High - Performance Computing) scenarios, from the perspective of technological innovation, Lenovo is also the direction they learn from.

"Lenovo is an industry senior we admire, not only because it is a global company with excellent intelligent hardware products, but also because it has occupied the leading position in the domestic and international HPC fields for many years."

Wang Chenhan said bluntly that he was already very familiar with Lenovo in terms of technology and application scenarios. But what he learned after participating in the event was that Lenovo Group also has its own venture capital fund department and has invested in and incubated hundreds of enterprises based on the group's own business.

"What I can see is that Lenovo is actively exploring and integrating internal and external resources through activities such as roadshows to build a high - quality innovation ecosystem, which shows that they are at the forefront in embracing the changes of the era and the industry."

When it comes to the similarity between his entrepreneurial concept and the Lenovo ecosystem, Wang Chenhan mentioned three keywords: innovation, pragmatism, and scenarios. "Whether it is Lenovo itself or AI entrepreneurs noticed by Lenovo, they are always looking for the combination points between traditional industries and AI, and everyone values specific application scenarios and the actual value that can be brought to users more."

In fact, in the roadshow activity, the ecological connection role of Lenovo has already let Bayesian Computing see the dawn of commercialization.

Wang Chenhan recalled that another enterprise mainly engaged in the monitoring, management, and optimization of computing power systems that came to the roadshow came to communicate and cooperate after the roadshow, and tried to use their products on Bayesian Computing's own computing power cluster. They may also jointly explore the market in the future.

"Without Lenovo's activity, it would be difficult for us to contact each other and quickly establish a trust relationship." Wang Chenhan said that the biggest feeling about this roadshow was that he communicated with many "people deeply involved in the industry". "Lenovo attaches great importance to similar roadshow activities and invites Lenovo Group's business, research institute, investment fund, etc. to participate in the scoring, which shows that Lenovo really wants to evaluate the growth prospects and cooperation possibilities of enterprises from an industrial perspective. So it not only made us gain the recognition of the industry and strengthened our confidence in doing a good job in products, but also brought us new cooperation opportunities."

The Industry Embraces Innovation

When asked what resources a high - growth startup like theirs lacks the most at present, Wang Chenhan replied thoughtfully: the first is market education in the scientific research field, which is the main customer group, and the second is talent competitiveness. These two points are also the common needs for the AI infrastructure track to move towards commercialization.

Wang Chenhan believes that if they can join the Lenovo ecosystem in the future, both of these problems can be better solved. Although leading players from the Internet to intelligent hardware giants are all making efforts to layout the AI application ecosystem, Lenovo, as a "chain leader" with a complete industrial chain, actually has very obvious advantages:

"Lenovo's marketization advantages in AI innovation lie in brand, traffic, and data. Lenovo's brand has promoted the development of intelligent hardware such as AI PCs, and enterprise service solutions have taken root in all industries. From C - end to large - enterprise customers, Lenovo can achieve a higher conversion rate by integrating digital technologies such as hardware ecosystem and AI applications, and at the same time precipitate a large amount of data for better AI iteration."

This well - developed business model makes Bayesian Computing's learning from the Lenovo ecosystem not only stay in technology but also reflect in many aspects such as business model, team building, and market strategy.

Currently, Bayesian Computing is having in - depth exchanges with Lenovo's sales team in the education field, looking forward to helping customers achieve low - cost cold start of heterogeneous computing and improve computing power efficiency through several pilot projects to verify its own value.

The Lenovo ecosystem has also paid certain attention to this innovative enterprise: in the past few years, Bayesian Computing has developed to the Series C financing stage, and one of its shareholders is Lenovo Star.

After the announcement of the Top 10 list of the roadshow, according to Wang Chenhan's description, Lenovo's "Star Program" to support startup teams immediately connected Bayesian Computing with the person in charge of Lenovo Group's ecological cooperation and continued to communicate about the enterprise's needs.

As one of the earliest enterprises to use heterogeneous chips such as GPUs to form supercomputing clusters, Bayesian Computing has reduced the training cost of large models in enterprise AI scenarios to 8.25% of the industry benchmark, and the inference cost has also been significantly optimized. In the fierce AI Infra competition, they are moving forward along the wave of technology.

At the same time, AI applications cannot be developed in isolation. Communicating more with industrial leaders like Lenovo can really improve the startup's understanding of "how to solve problems".

During the exploration period of AI applications, if infrastructure services such as AI PaaS are the cornerstone of the explosion of AI applications, then the embrace and tolerance of the industry are the cornerstone of innovation. In this process, the "mutual pursuit" between the commercialization ecosystem and innovation forces is crucial.

Now, Lenovo's Small and Medium Business Group is creating opportunities to help more teams like Bayesian Computing survive firmly and grow rapidly.