From a Tripartite Balance to a Quadrilateral Standoff

There have been new changes in the cloud service sector.

Recently, Volcano Engine has comprehensively upgraded its third-generation cloud servers, combining Volcano Engine's full-stack self-developed, hardware-software integrated DPU2.0 architecture, and self-developed virtualization full offloading technology, achieving zero loss of computing power.

This means that Volcano Engine can provide higher cost-effective computing power.

Combined with the Doubao large model, Volcano Engine can meet the urgent needs of various industries to reduce costs, increase efficiency, and improve quality under the wave of AI, helping enterprises achieve sustainable high-quality growth through digital transformation.

Reuse Inside and Outside, Highest Cost-Effectiveness

Cloud services have entered an era of competition among giants.

With the entry of internet players of all sizes, cloud services once became a standard configuration for the industry. However, the "cloud war" is not only a price war but also a competition of core competitiveness. The size of small and medium players is not on the same level as the top players, resulting in infrastructure and AI capabilities falling behind, and the industry has undergone a major shakeup.

More importantly, as competition intensifies, the gunpowder smell between top players has become increasingly strong, not only competing for the incremental market but also fighting fiercely for the existing market.

Against this background, Volcano Engine's emergence as a dark horse is particularly noteworthy.

On June 22, 2020, Volcano Engine officially launched, making it the latest entrant among domestic cloud service providers. However, its performance does not resemble that of a latecomer, but rather it quickly opened up the situation and successfully established a foothold.

The reason for this is closely related to its early flag planting and positioning.

In fact, although Volcano Engine was relatively late in external exposure, it has always provided technical services to ByteDance's ecosystem internally. With both internal and external efforts, it has accumulated rich cloud service experience, data, and a large server scale and computing power resources.

For example, Volcano Engine's real-time data warehouse provides a complete set of real-time data for Douyin E-commerce, including real-time dashboards, real-time analysis, real-time alerts, and real-time marketing.

In this way, Volcano Engine has entered a positive cycle mode through its approach of reusing both internal and external resources.

An insider told Zinc Dimension: "Volcano Engine's actual competitive advantage lies in its resource scale, which is among the best in the country. Volcano Engine and ByteDance's businesses such as Douyin and Toutiao are internally and externally pooled, able to provide customers with the highest cost-effective cloud services, thus gaining a price advantage."

More crucially, the Doubao large model is in place.

Currently, large models are moving from "demonstration" to "practical use" and will become an indispensable driving force for cloud services in the future, thus becoming the focus of industry competition.

According to the latest FlagEval evaluation ranking, in the "objective evaluation" of closed-source large models, the Doubao large model ranked second with a comprehensive score of 75.96, second only to GPT-4, and is the highest-scoring domestic large model. In the "subjective evaluation," the Doubao large model also ranked second.

It is not difficult to see that the Doubao large model has filled the last gap for Volcano Engine, giving it the confidence to compete with top players.

The Doubao Large Model, Technical Cost Reduction as the Key

To catch up from behind, using price to exchange for market share is a mainstream approach, and the Doubao large model is no exception.

Public data shows that the inference input price for the Doubao general model pro-32k is only 0.0008 yuan per thousand tokens. At that time, the pricing of similar models on the market was typically 0.12 yuan per thousand tokens, making the Doubao general model 99.3% cheaper.

In short, Volcano Engine has propelled large models into the "cent era" and fired the "first shot" in the price reduction of mainstream large models since May 2024.

It is worth noting that traditional large models, due to their relatively early investment and accumulation, follow a path of scaling cost reduction, while the Doubao large model focuses primarily on technical cost reduction.

Volcano Engine President Tan Dai said in an interview with the media: "Price reduction is achieved through technological optimization of costs. If it is just subsidies, using losses to exchange for income is not sustainable. Volcano Engine will not take this path."

On the one hand, there is computing power optimization.

Volcano Engine's third-generation general-purpose instance g3i can achieve up to 122% higher computing power compared to the previous generation. It performs even better in business scenarios such as high-performance computing, database deployment, web applications, and audio-video processing, especially achieving significant technical breakthroughs in the field of AI inference.

For example, performing SDXL-Turbo text-to-image model inference can achieve second-level image generation; in conversational text generation scenarios, performing large language model inference with 8 billion parameters can control the first packet delay within 1 second.

On the other hand, there is scheduling optimization.

An industry insider told Zinc Dimension: "There is a significant tidal phenomenon in the invocation of large models, with some demands during the day, some at night, and even some in the early morning. The peak and valley demands may also differ, resulting in spatial and temporal imbalance, requiring load mixing for the invocation of large models."

To address this, Volcano Engine has built a robust elastic resource pool and pioneered the g3i instance elastic reservation mode in the industry, featuring a pay-as-you-go pricing model with "free advance reservation and automatic delivery at the appointed time."

In this way, users can flexibly access computing power resources, satisfying both high-performance, high-response, and high-computing power demands during peak periods and low-cost demands during trough periods, achieving a cost optimization of more than 27% compared to traditional billing methods.

Anchoring on computing power optimization and scheduling optimization, Volcano Engine has successfully chosen a technical cost reduction path that suits itself.

In addition, the Doubao large model is also noteworthy in its implementation, focusing on the two major areas of mobile phones and automobiles, fully unleashing the productivity of large models, and truly serving B-end enterprises.

For example, Volcano Engine has partnered with intelligent terminal manufacturers such as OPPO, vivo, Honor, Xiaomi, Samsung, and ASUS to jointly establish the Intelligent Terminal Large Model Alliance. Xiaomi's "Xiaoai Classmate," OPPO's Xiaobu voice assistant, Honor's intelligent office assistant, and others have successfully integrated Volcano Engine's large model services, bringing a more intelligent AI interaction experience to massive mobile phone users.

From a Tripartite Balance to a Quadrilateral Standoff

As can be seen above, Volcano Engine has greater imagination.

Canalys data shows that global cloud service spending increased by 21% year-on-year in the first quarter of 2024, reaching $79.8 billion; among them, spending in mainland China increased by 20% year-on-year, reaching $9.2 billion.

In other words, various industries are still accelerating their adoption of cloud services, and cloud services represent a vast incremental market.

As everyone knows, competition in the existing market presents a situation of one rising and another falling, while competition in the incremental market has the potential for all to progress together. This is why, despite the "tripartite balance" of Alibaba Cloud, Huawei Cloud, and Tencent Cloud, the market is still eagerly speculating about who will become "China's fourth cloud."

In fact, there is no winner-takes-all scenario in cloud services.

Most small and medium-sized enterprises are price-sensitive, and when price reduction becomes a keyword in the industry, they have more options. Large enterprises, on the other hand, will not put all their eggs in one basket to avoid being "manipulated" and losing bargaining power.

In addition, different clouds have different characteristics, and enterprises will give priority to more suitable "clouds."

Taking Volcano Engine as an example, its Creative Cloud combines audio-video AI technology, copyrighted content, virtual human technology, and a content resource pool to provide enterprises with one-stop intelligent content production services. Its Video Cloud is the first service provider in the industry to apply H.266 on a large scale, saving 50% on encoding compared to H.265, and innovating in the entire AIGC chain. Old movies such as "Project A" restored by its 4K technology have been showcased at film festivals in Beijing, Cannes, and Shanghai.

In addition, large models have become a new race point.

For example, traditional autonomous driving solutions suffer from delays and inaccuracies in human-car interaction. However, after equipping with the Doubao general model lite, low-latency dialogue can be achieved, with an effect that is more than 50% better than traditional voice processing, thus better adapting to intelligent cockpit scenarios.

Currently, large models have not yet truly landed on a large scale. Volcano Engine needs to continue honing its strength, seizing the opportunities to implement them well, expanding the market, and leveraging its differentiated competitiveness to help enterprises achieve value growth. It is possible that it could catch up and become "China's fourth cloud."

As a result, the competitive landscape of cloud services is shifting from a tripartite balance to a quadrilateral standoff.