Source: Data Ape

Image source: Generated by Wujie AI

In the era of rapid technological development, OpenAI's ChatGPT has undoubtedly become a shining star in the field of artificial intelligence. However, with the rapid increase in its user base, an undeniable problem has gradually emerged—shortage of computing power. This is not only a challenge for OpenAI alone, but also a difficult problem facing the entire artificial intelligence industry.

In this in-depth analysis article, we will explore the root causes of the computing power shortage and how it affects the development of large models and the future of the industry. At the same time, we will also discuss potential solutions to this problem, including the development of domestic GPUs and their potential impact on the global market.

Computing Power Shortage, a Persistent Challenge for OpenAI

The computing power shortage has always been a thorny issue that OpenAI has had to face during its rapid rise. Recently, this problem has become even more prominent due to a high-profile event. OpenAI held a large-scale showcase event, dubbed the "Tech Spring Festival," to demonstrate its latest technological achievements, attracting great attention worldwide. As a result, a tidal wave of users flocked to OpenAI's platform, especially its star product, ChatGPT.

However, behind this wave of enthusiasm lies a huge challenge. The explosive growth in the number of users quickly exceeded OpenAI's computing capacity. Just two days after the conference ended, a shocking fact emerged: the ChatGPT server crashed. Countless netizens reported that they were unable to use ChatGPT and the API provided by OpenAI.

Faced with this crisis, OpenAI had to make a shocking decision: to temporarily suspend new registrations for ChatGPT Plus users. The economic cost behind this decision is enormous. Calculated at a monthly subscription fee of $20, 100 million new users could bring in $2 billion in revenue for OpenAI each month. Such a huge potential income had to be given up due to the shortage of computing power, which is undoubtedly a very passive choice.

In fact, the computing power shortage problem has not only occurred recently. Since the launch of ChatGPT, computing power issues have always been a lingering shadow. For example, in April, the purchase of ChatGPT Plus's paid projects was also forced to be suspended. This situation has occurred from time to time, seemingly becoming a norm in OpenAI's growth path.

These events reveal an indisputable fact: in the current context of technological development, computing power has become a major bottleneck limiting the innovation of AI technology and the expansion of commercial applications. For OpenAI, this is not only a technical challenge, but also a strategic dilemma. Finding a balance between rapidly expanding market demand and limited computing resources has become a difficult problem facing OpenAI. This challenge not only concerns the company's short-term earnings, but also involves long-term market position and technological leadership.

Even the Landlord Has No Surplus Grain

OpenAI has repeatedly announced that there is not enough computing power.

It is worth noting that OpenAI is a star enterprise in large models, with huge financing and a large amount of computing resources. Moreover, it also has a "patron" like Microsoft, which provides comprehensive computing support. Microsoft owns the world's second-largest cloud computing resources.

From this perspective, OpenAI can be said to be a "landlord" in terms of computing power. However, the actual situation is that the landlord has no surplus grain. So, why would a company with huge financing and strong backing like Microsoft find itself in such a predicament?

We must realize that the demand for computing power by large models is unprecedented. These models are based on neural networks of tens of billions, and each computation is a huge test of computing power. Simply put, we are facing a new level of computing demand that cannot be found in the history of software development. Traditional computing resources such as CPUs are inadequate here, and GPUs, which are needed, are undoubtedly at the forefront of this technological revolution.

However, the problem with GPUs is that they are not only emerging technological products, but also face the dual challenges of design iteration and production capacity constraints. Despite the growing demand for GPUs in the tech industry, global chip manufacturing capacity has not kept pace with this demand. The existing semiconductor manufacturing and testing system is mainly designed around CPUs, and it is clear that they have not fully adapted to emerging GPUs. This means that there is still a long way to go in increasing GPU production capacity and adapting to new technological demands.

GPU technology continues to advance, with each generation of products striving to improve performance and efficiency, requiring continuous research and development investment and technological innovation. However, this continuous technological iteration also means increasing research and development costs, while also increasing the complexity of the manufacturing process.

In addition to production capacity issues, the cost of GPUs is also a significant concern. Building a GPU computing cluster capable of supporting large model operations requires not only technology, but also a huge capital investment. Even for tech giants like OpenAI, this is a significant burden. Finding a balance between cost and efficiency is undoubtedly a difficult choice.

If even OpenAI is struggling with the problem of computing power shortage, how will other companies fare? This is not only a challenge for OpenAI, but also for the entire artificial intelligence industry. What we are witnessing is a huge shift: from traditional computing to AI-driven computing. In this transition, computing power has become the most critical bottleneck.

We cannot ignore the fact that this shortage is not an overnight phenomenon, but rather the result of long-term mismatch between technological development and market demand. The production constraints, technological development, and cost issues of GPU chips are multifaceted, involving global supply chains, technological innovation, and economic models. The high computing power demand of large model applications presents unprecedented challenges to existing technological architectures, forcing the entire industry to rethink how to design, build, and optimize computing resources.

When B-End Applications Scale Up, the Computing Power Shortage Problem Will Become More Severe

There is another very important but easily overlooked issue.

When we discuss the computing power shortage problem, the focus is usually on the current C-end user experience. However, this is just the tip of the iceberg. A more severe but often overlooked problem lurks in the scaling of B-end applications. Currently, although large models like ChatGPT mainly serve C-end users, this is just the beginning. With the gradual growth and maturation of B-end applications, we will face an unprecedented surge in computing power demand.

In the Chinese market, this trend has already begun to emerge. Products such as Baidu's Wenxin Yiyuan and Alibaba's Tongyi Qianwen, although currently mainly serving C-end users, have already begun to explore B-end applications. Currently, these products are still in the product development stage, but once they enter the large-scale commercialization stage, the situation will be completely different. The complexity of B-end business far exceeds that of C-end. At the C-end, user interaction with the system is usually simple queries or command execution. However, at the B-end, each business process may involve more complex data processing, analysis, and decision-making processes. These processes not only require more computing resources, but also have higher requirements for the quality and stability of computing power.

Of greater concern is that the computing power consumption of B-end businesses is not only reflected in the complexity of individual interactions, but also in the frequency of calls. At the B-end, the application of large models is often continuous and high-frequency, in stark contrast to the occasional queries and usage at the C-end. For example, in industries such as finance, healthcare, and manufacturing, large models need to continuously process large amounts of data and provide real-time analysis and decision support. This high-frequency, high-load computing demand puts tremendous pressure on computing power.

It can be foreseen that as large models become popular in B-end applications, their demand for computing power will quickly surpass that of the C-end. This shift may be silent, but its impact will be profound. On the one hand, the increased demand for computing power will drive the development of related technologies, such as more efficient GPUs and optimized computing architectures. On the other hand, this will also have a significant impact on the allocation of resources, cost structure, and business models of the entire industry.

In this process, we may see some companies being forced to withdraw from the market due to their inability to bear the cost of computing power, while others may stand out with advanced computing power management and optimization technologies.

China Faces a Dual Computing Power Bottleneck

Globally, the shortage of computing power has become a major bottleneck for the development of artificial intelligence, and for China, this challenge appears even more daunting. Chinese large model enterprises not only have to deal with a global shortage of computing power ("natural disaster"), but also face unique market supply restrictions ("man-made disaster"), adding a dual pressure that makes the prospects for the development of large models in China complex and challenging.

We must recognize the limitations of Chinese large model enterprises in terms of computing power resources. Although companies such as Baidu, Alibaba, ByteDance, Tencent, and Huawei have made significant achievements in the development of large models, they are facing real and urgent challenges in computing power. Currently, due to the overall insufficient development of the global GPU industry, Chinese companies are facing major obstacles in obtaining sufficient computing power resources. This "natural disaster" problem is a necessary path for technological development and industrial upgrading, and it will take time and enormous investment to solve.

Even more challenging is the "man-made disaster" that Chinese large model enterprises face from the international market—especially the supply restrictions imposed by international giants such as NVIDIA on the Chinese market. These policy restrictions directly affect the ability of Chinese companies to obtain high-end GPU chips, thereby exacerbating the shortage of computing power resources. This dual restriction undoubtedly adds additional uncertainty and challenges to the development of Chinese large model enterprises.

Currently, although products like Baidu's Wenxin Yiyuan and Alibaba's Tongyi Qianwen have not yet reached the scale of hundreds of millions of users like ChatGPT, this does not mean that Chinese companies can easily cope with the existing computing power challenges. As these products develop and the market expands, especially when they begin to be widely used in the B-end market, the demand for computing power will increase sharply. At that time, the problem of computing power shortage will become more prominent and may severely restrict the development of the Chinese large model industry.

In the long run, if China cannot effectively address this dual computing power bottleneck, the development of its large model industry may be limited to a lower level. This will not only affect the competitiveness of the domestic market, but also limit China's influence in the field of artificial intelligence globally. Therefore, solving the problem of computing power shortage is crucial for the future development of China's large model industry. This is not only a technical issue, but also a strategic issue that concerns China's position and future in the global AI competition.

In the face of the dual computing power challenges in China, there have recently been some encouraging positive signals, especially in the development of domestic GPUs. Leading domestic technology companies such as Baidu, Alibaba, 360, and others have begun to collaborate with domestic GPU manufacturers such as Huawei.

The rise of domestic GPUs holds profound significance for addressing China's computing power shortage. If these domestic GPUs can match industry leader NVIDIA in performance and effectively resolve manufacturing bottlenecks, it will bring unprecedented opportunities for China's large model industry. Historically, once domestic technology matures, it can usually enter the market at a more competitive price. This means that if domestic GPUs are successful, they are likely to offer similar or even better performance at significantly lower prices than international brands.

This cost advantage can not only alleviate the current computing power shortage but also potentially change the market landscape completely. The high cost of GPUs has always been a significant factor limiting the widespread adoption and application of large model technology. If domestic GPUs can provide high-performance computing power at lower prices, it will greatly promote the application of large model technology across various industries and accelerate China's development in the field of artificial intelligence.

More importantly, this development may enable China to achieve a "comeback victory" in the global AI competition. In terms of large model computing power and applications, China may not only catch up with but even surpass leading countries such as the United States.

Of course, all of this is still in the early stages of development, and the success of domestic GPUs still needs to overcome technical challenges. However, the positive signs that have emerged indicate that China has taken solid steps towards achieving computing power independence. In the coming years, we can expect to witness the maturity and large-scale application of domestic GPU technology, and how it will drive the leapfrog development of China's large model industry.

In conclusion, in the journey of exploring the global challenge of computing power shortage, we have not only witnessed the continuous advancement of technological boundaries but also deeply experienced the complex challenges facing industry development. From the story of OpenAI to the dual predicament of Chinese large model enterprises, and to the rise of domestic GPU technology, all of this reveals a core truth: computing power has become a critical strategic resource in the future development of artificial intelligence. This is not only a competition at the technological level but also an investment and layout of global technological forces for the future.

Looking ahead, with the continuous evolution of technology and market demand, we have reason to believe that the problem of computing power shortage will eventually be resolved. In this process, innovation, collaboration, and strategic adjustments will be key issues that every participant must face. Ultimately, this challenge related to computing power will define the future form of artificial intelligence technology and shape our digital world.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。