The Evolution of Cloud Giants' Ecosystem: From Cloud Computing to the Era of Large AI Models
-
"The only constant is change," a statement from the 2023 letter to all employees by Wu Yongming, CEO of Alibaba Cloud, the cloud provider that underwent the most significant transformations this year. For Alibaba Cloud, which has long led China's top three cloud providers, public cloud and integration have been the two major changes since the second half of 2023 and will remain key focuses in 2024.
Similarly, Alibaba Cloud's transformation reflects the overall strategic positioning of domestic cloud providers in 2023. It corresponds to the subtle shift in the role cloud providers are expected to play in the ecosystem of the future large AI model era. Now, cloud providers have gained another strategic lever: general-purpose large models.
"Applying the 63 analyzed generative AI technologies across various industries could add $2.6 trillion to $4.4 trillion annually to the global economy." This prediction comes from McKinsey's report The Economic Potential of Generative AI: The Next Productivity Frontier.
A consensus view is that 2023 was a year of rapid development for large models, while 2024 will see an explosion of AI applications in the industrial sector. Against this backdrop, corporate demand for generative AI has surged dramatically, and this rising demand will directly impact the cloud giants providing AI computing power. In the traditional cloud computing era of the past decade, cloud vendors have undergone constant changes in their ecosystem strategies. They transitioned from being "general integrators" to being "integrated," shifting from handling all business operations themselves to delegating more applications to ISV vendors with independent R&D capabilities.
Now, AI has further propelled this strategy forward.
Several indicators highlight this trend: In December 2023, Cai Yinghua, a key figure responsible for advancing government and enterprise business, left Alibaba Cloud, signaling a full embrace of the public cloud strategy. Additionally, Tencent Cloud reiterated its focus on cost efficiency while strengthening its ecosystem and in-house product strategy. Huawei Cloud has also accelerated strategic collaborations with various enterprises. These signals collectively reflect that in the new era of transitioning from cloud to AI large models, the traditional all-in-one model is being rapidly abandoned. Leading cloud vendors like Huawei Cloud, Alibaba Cloud, and Tencent Cloud are all seeking greater ecosystem openness.
On the other hand, ISVs and software vendors are actively joining the cloud vendors' large-model ecosystems to reinvent their applications. According to incomplete statistics, over the past few months, as cloud vendors' ecosystem policies have become clearer, an increasing number of ISVs have begun embracing the cloud ecosystem, rapidly joining its pool. At present, for cloud vendors, large models are not only a new lever to drive the growth of their cloud computing market share but also a catalyst for accelerating their ecological strategies based on 'openness'.
From 'Being Integrated' to MaaS,
The Ecological Baton of Cloud Giants
2023 marks the fifth year of Alibaba Cloud's 'Being Integrated' strategy.
At the same time, this year has seen significant organizational changes within Alibaba Cloud. The most notable strategic shift is the abandonment of the previous 'industry + regionalization' approach in favor of announcing an 'AI-driven, public cloud-first' strategy, accompanied by sweeping reforms. Over the past year, the baton of Alibaba Cloud CEO has been passed from Zhang Jianfeng to Daniel Zhang and then to Wu Yongming. It's worth mentioning that Cai Yinghua, who was responsible for promoting government and enterprise business, also left Alibaba Cloud on December 13, 2023.
From Zhang Jianfeng stepping down to Wu Yongming taking office, and then Cai Yinghua's departure, a very clear signal is that in the era of large models, Alibaba Cloud will accelerate the improvement of its previously incomplete 'general integration' model. This is precisely the strategic route emphasized by Wu Yongming upon taking office, namely 'returning to public cloud and reducing offline projects.'
Over the past few years, domestic cloud vendors including Alibaba Cloud have not made smooth progress in customized projects. Now, objectively speaking, while Alibaba Cloud's public cloud-first strategy may have some short-term impact on government-related businesses, it will provide long-term benefits as a positive effect.
Similarly, Tencent Cloud announced its 'be integrated' strategy starting in 2022. Years earlier, Ma Huateng had said that Tencent would give half its life to partners. To date, Tencent Cloud's integrated revenue has grown by 500%.
In fact, 'rushing to the forefront as the big brother' was exactly the initial exploration model for all domestic cloud vendors. In the short term, acting as a general integrator allows cloud vendors to secure large contracts, explore specific scenarios in vertical industries, and accumulate industry know-how during their growth phase. However, this model does not favor profit margins. In the long run, being a general integrator not only affects profitability but also hinders the development of their core business models.
Therefore, over the past few years, whether it's lead partners, product development partners, or consulting and after-sales partners, all have been intensifying their efforts. They provide support through funding, cloud computing discounts, training, and other means to accelerate the growth of enterprises within their ecosystems.
Today, this integrated approach has found a new driver: MaaS (Model as a Service).
In the past, the core capabilities of domestic cloud vendors were largely limited to the three major components of IaaS—computing, storage, and networking—lagging behind overseas cloud vendors like AWS in PaaS-layer cloud services.
However, the true value of cloud vendors lies in their evolution toward PaaS platforms. "Leveraging the strengths of IaaS, PaaS, and SaaS to create composable capabilities, enabling secondary assembly on the platform, and ultimately driving innovation in both business and platform." This is how Gartner's 2023 Top 10 Trends report defines the industry cloud platform. The key term in this statement is "assembly," which implies that cloud vendors need to fully leverage their PaaS platform capabilities. This is precisely why domestic cloud giants have been advocating for "being integrated" in recent years.
However, the challenge lies in the fact that not all capabilities of cloud vendors have been fully PaaS-ified, meaning ecosystem partners don't possess complete service capabilities. Additionally, in certain core areas, cloud vendors still need to develop products themselves (such as databases) to maintain high profit margins, creating inherent conflicts between complete PaaS implementation and ecosystem development.
AI large models, or what we call MaaS (Model as a Service), are quietly transforming this model.
As of now, we can see that major cloud computing providers like Alibaba Cloud, Huawei Cloud, Tencent Cloud, Baidu Cloud, and JD Cloud have all launched MaaS services.
MaaS brings new growth opportunities for cloud vendors through two main approaches: First, API call services. In this scenario, cloud vendors can charge based on usage volume or time. Second, providing AI computing power services for training and running large models. Currently, with the continuous increase in domestic model parameters and the explosion of vertical domain models, demand for model training is rapidly emerging. For the ecosystem, the MaaS-based model can bring all ecosystem partners to a unified starting point for AI services. This not only avoids conflicts on the product side (by introducing a new service product) but also, under the MaaS ecosystem model, allows cloud providers to act as foundational players with strong profit incentives, offering more comprehensive and clearly defined service guarantees to ecosystem enterprises.
It can be said that compared to their previous focus on IaaS and PaaS, the emergence of MaaS provides cloud providers with new growth opportunities. Moreover, a greater value of MaaS for the ecosystem is that it brings together customers, developers, startups, and ISVs through data-trained models, collectively driving cost reduction and efficiency improvements.
From pure PaaS to MaaS, the ecosystem strategies of cloud providers are evolving.
How is the large model ecosystem of cloud providers developing?
In 2023, it is undeniable that domestic cloud giants are all in on AI. From the perspective of the IaaS+PaaS market, the year-on-year growth in the first half of 2023 was 15.9%, marking the lowest growth rate in nearly three years. This data comes from IDC's "China Public Cloud Services Market (First Half of 2023) Tracker" report. Consequently, large models have naturally become a new growth lever for cloud providers.
Among these, the AI ecosystem serves as a crucial component.
In fact, as early as the end of 2022, Alibaba Cloud proposed the MaaS (Model as a Service) concept, coinciding with the initial surge in popularity of ChatGPT overseas. At the 2023 Yunqi Conference, Jingren Zhou revealed that currently half of China's major large model companies—including Baichuan Intelligence, Zhipu AI, 01.AI, Kunlun Wanwei, vivo, and Fudan University—are running on Alibaba Cloud.
In terms of specific service offerings, Alibaba Cloud's open-source community ModelScope provides two main services: First, developers can directly access Alibaba Cloud's computing power and large model training/inference platform, which offers one-stop services for ecosystem partners. Second, the platform enables model fine-tuning, providing model training and data training services, along with optimization tailored to specific business scenarios.
Simply put, one approach offers direct API access, while the other assists enterprises in model fine-tuning by providing model training capabilities. Tencent Cloud is making rapid progress. At the Tencent Cloud Ecosystem Conference on January 18, 2024, Tencent Cloud announced its latest achievements: "Over the past two years, Tencent Cloud's revenue from 'being integrated' has grown by 200% year-over-year." Moreover, after launching its large model, Tencent Cloud now allocates 80%-90% of its business income to partners in the 'being integrated' business.
Additionally, Tencent Cloud has been accelerating its 'being integrated' business over the past two years. In fact, since proposing the 'being integrated' strategy, Tencent Cloud has stepped back from its role as a 'general integrator.' Today, with the MaaS model, Tencent Cloud's ecosystem approach has undergone a significant transformation. To some extent, Tencent Cloud's MaaS also helps the company accumulate industry know-how in vertical sectors.
Over the past year, as this model has deepened, Tencent Cloud has amassed substantial resources in ten industries, including healthcare, finance, culture and tourism, media, government affairs, and education.
Most of these vertical large models are completed in collaboration with ecosystem partners. In the construction of vertical models, the roles of all parties are clearly defined. Tencent Cloud provides PaaS-layer capabilities, large enterprises in vertical sectors supply data and requirements, and specialized vertical model construction teams finalize the industry-specific private deployments. Huawei Cloud has also taken swift action. It is reported that in terms of the large model ecosystem, Huawei Cloud adopts a cooperative approach with three types of partners: software partners, service partners, and consulting & system integration partners.
Not only does it support the invocation of the Pangu large model, but it also facilitates the deployment of various open-source and third-party commercial models. From large model development to customization, Huawei Cloud provides multiple engineering suites, including data engineering, model development, and application development, helping users build independent datasets and upgrade foundational models.
From a certain perspective, whether it's Alibaba Cloud, Tencent Cloud, or Huawei Cloud, their ecosystem development at the MaaS level has become more open, compatible, and decisive compared to before. Both in terms of computing power and data, there is ample space for ecosystem partners.
As of now, the MaaS ecosystem models of these companies have largely taken shape. With computing power as the foundation, the upper layers—such as large model collection, annotation, and pre-training—serve as key drivers, acting as ecosystem weapons to accelerate their own ecosystem construction and provide stronger support for the deployment of their large models.
In the era of large models, The Evolution of Cloud Providers' "Public Cloud"
In reality, the shift from IaaS+PaaS+SaaS to IaaS+PaaS+MaaS represents cloud providers' new monetization strategy in the era of large AI models. Moving from directly offering SaaS to exploring MaaS also signifies that cloud providers are fully embracing an "integrated" ecosystem model.
This ecosystem model inevitably aligns with the trend of "Public Cloud + AI." Objectively speaking, this approach is becoming the default choice for major players.
From a market demand perspective, public cloud significantly reduces the cost of AI models, especially amid GPU computing shortages. "We’re already seeing many enterprises and government agencies leaning toward public cloud," an industry insider told Industrial Insights. "The cost is much lower than on-premises deployment—often 3-4 times cheaper."
On the other hand, the public cloud’s AI large model approach can indeed drive growth for cloud providers to some extent. From the current business models of cloud providers, there are two main approaches: one is API calls, charging based on the number of tokens; the other is providing secondary development services for large enterprises. The former relies on the cloud provider's accumulated computing power, while the latter depends on partnerships within specific domains.
Whether it's Alibaba Cloud, Tencent Cloud, or Huawei Cloud, apart from multiple signals on the computing power side, the most common trend has been the construction of vertical domain-specific large models. This involves collaborating with leading enterprises in specific verticals to train domain-specific data representations, whether based on local deployment or public cloud models, ultimately resulting in models tailored for niche markets.
Additionally, through mutual agreements, some of these capabilities are integrated into public cloud models, further enhancing the foundational capabilities of the base models. These enhanced capabilities are then incorporated into their own products, such as Baidu's Wenku, Tencent's Tencent Meeting, and Alibaba's DingTalk.
In fact, overseas cloud providers have already set an example in exploring ecosystem models for large-scale AI. When ChatGPT was gaining popularity, Microsoft Azure did not choose to develop its own large model, but instead provided services directly based on GPT through deep integration. This approach does not mean abandoning large models. As seen in Microsoft's latest financial report, its revenue increased by 13% year-over-year, net profit grew by 27% year-over-year, reaching $22.2 billion.
A closer look at Microsoft's products reveals that its Windows, Office, Bing, and other software have all developed AI Copilot assistants. In this way, Microsoft has formed a business closed-loop from cloud to AI to software.
From strengthening private model capabilities to public cloud AI models and then to specific product implementations, domestic cloud vendors are also following this path from the second half of 2023 to early 2024.