At this critical moment when the world is entering the next round of technological competition, if we want to continue to maintain the second largest battlefield of artificial intelligence in the world, and even the largest battlefield of global AI application innovation, we obviously need to find solutions.
Contributing writer丨Boiling snow
Recently, Beijing clearly proposed to establish cooperation with cloud service vendors in the general artificial intelligence industry innovation partnership program to provide computing power for the industry.
Artificial intelligence has reactivated the global competition in science and technology, AI capabilities are giving birth to a large number of "new species", the speed of evolution is changing rapidly, and industrial ecology has become a key element to win the competition in digital technology.
Moving fast is a consensus. On May 5, the Central Financial and Economic Commission pointed out that "it is necessary to grasp the wave of new scientific and technological revolution in artificial intelligence". Recently, Beijing has successively announced two batches of industrial innovation partner lists, encouraging the acceleration of the development of artificial intelligence industry and building an industrial ecology from the way of industrial partnership.
Graph/Network
Among the above measures, there are two directions worth paying attention to, one is to clearly propose to strengthen cooperation with public cloud vendors and other market players, and the first batch of computing power partners includes public cloud vendors such as Alibaba Cloud; Second, the plan covers four types of partners: computing power, data, models, and applications, reflecting the four growth directions of the AI ecosystem.
Large models and intelligent applications are not breakthroughs in single-point technology, not only a set of algorithms or technologies, but a great challenge to the underlying computing infrastructure, but also a comprehensive innovation from the "public cloud + AI" technology system - the technological innovation ecology has never been as important as today.
01
Don't overlook the power of computing power
A new wave of artificial intelligence will bring huge demand for computing power. Computing power is considered to be the productivity of the era of big data, when data elements become an important factor to drive economic development, it is no exaggeration to say that computing power is one of the driving forces supporting the development of the national economy.
Since 2023, with the rapid development of the digital economy, especially the outbreak of artificial intelligence, the demand for computing power in the entire society has shown a rapid growth trend.
International Data Corporation (IDC) pointed out in the "2022 Global Computing Power Index Assessment Report" that the scale and diversity of China's computing power industry are currently continuing to grow rapidly, and from 2018 to 2022, the computing power scale of Beijing, Hangzhou and Shenzhen ranked among the top three in the country.
Expanding the scale of computing power is becoming an inevitable requirement for the economic development of regional center cities. In Beijing, for example, this layout has already been underway.
Recently, Beijing, which occupies an important position in the global field of artificial intelligence, issued three policy documents supporting and encouraging the development of artificial intelligence in just one month:
On April 27, Beijing issued the Implementation Plan for Accelerating the Construction of an AI Innovation Source with Global Influence (2023-2025) (Draft for Comments), which clearly proposes to make full use of the computing power supply capacity of cloud vendors, implement the computing power partnership program, integrate public cloud computing power leasing resources, open up to AI innovation subjects, and accelerate the construction of AI innovation sources with global influence;
On May 12, Beijing issued Several Measures to Promote the Innovation and Development of General Artificial Intelligence (2023-2025), proposing to strengthen cooperation with market entities such as leading public cloud vendors and implement the computing power partnership program;
On May 19, the Beijing Municipal Bureau of Economy and Information Technology, together with the Municipal Science and Technology Commission, the Zhongguancun Administrative Committee and the Municipal Development and Reform Commission, jointly launched the "Beijing General Artificial Intelligence Industry Innovation Partnership Program". The first batch of partners has a total of 39 members, including Alibaba Cloud Computing Co., Ltd. and Beijing Super Cloud Computing Center. On July 2, the Artificial Intelligence Summit of the Global Digital Economy Conference released the second batch of partners.
From these documents, we can see some keywords, public cloud, computing power partners, artificial intelligence innovation. In fact, this is not much good news in the face of possible lockdowns - the layout of computing power based on local public clouds is already accelerating.
Graph/Network
In fact, from the success stories of the world's best companies, the "flywheel effect" of coupling public cloud and applications to form a closed loop is becoming increasingly prominent. One of the best is AWS, the global leader in cloud computing.
Since AWS provided Amazon elastic cloud in August 2006, as Amazon's cloud computing sector, its existence is not isolated, but has opened up Amazon's business line, and made it have great priority opportunities in the iteration and application of new technologies.
We have turned our attention back to China, and since the public measurement of Alibaba Cloud in 2010, the development of China's cloud computing field has also made great progress. Today, as the third in the world and the first in China, Alibaba Cloud's innovation speed in the field of artificial intelligence is not inferior to foreign manufacturers.
In fact, behind the common meaning of Qianqian, based on the technology accumulated over the years, Alibaba Cloud has been able to build a "100,000-card" ultra-large-scale intelligent computing cluster on the public cloud based on intelligent computing Lingjun, and solved the core technical problems of key networks, data storage, and observable, scheduling, and load balancing of computing power, and built a full-stack AI service from cluster IaaS to PaaS to MaaS, which is convenient for large model manufacturers and enterprise customers in all walks of life.
This should also be the reason why Alibaba Cloud became the first batch of computing power partners in Beijing.
02
Focus on doing great things
If we want computing power to accelerate artificial intelligence, how can our layout improve efficiency? In fact, a clear answer is given in Beijing's "Innovation Partnership Program": accelerate the aggregation of existing computing power and provide diversified and high-quality inclusive computing power for market entities.
If you want to aggregate computing power, there are not many options left, and cloud computing is the most preferred option. This is because the cloud business model itself is centralized construction, operation and maintenance, elastic contraction, and providing computing power resources to the whole society - the most efficient, lowest-cost, and highest utilization rate.
In other words, for the whole society, the inefficiency and high energy consumption caused by the dispersion of computing power resources are more harmful than beneficial from any point of view.
Compared with private and hybrid clouds, public clouds are undoubtedly the most efficient. Through the elastic use of multi-tenancy, resource utilization can be further improved.
The data shows that the efficiency of public cloud CPU use can be roughly 5-10 times that of private cloud. It's like a "high-speed rail" at 350 kilometers per hour VS a "green car" at 60 kilometers per hour. The use efficiency is increased by 5-10 times, the server saves 300 billion yuan, and the electricity bill saves 80 billion kWh, which is equivalent to the power loss scale of one Three Gorges.
Even from the perspective of carbon neutrality, the public cloud is the best option. Taking the Fudan CFFF platform as an example, Alibaba Cloud's green data center technology combined with the natural climate advantages of Ulanqab can save more than 2,000 kilowatts of total electricity per year, save 5 million yuan in electricity costs, and save an average of 15 tons of carbon per year.
Figure/Figureworm Creative
However, from the current point of view, the more critical and core blocking point is that the dispersion of computing power resources may not actually be a problem of choosing a technical route.
It can be seen that driven by the artificial intelligence industry, the construction of new intelligent computing centers has become a trend, but many intelligent computing centers cannot respond to market demand and lack market-oriented operation mechanisms, resulting in serious waste of resources.
According to statistics, the CPU utilization rate of the non-public cloud intelligent computing center is much lower than that of the cloud data center, and the resources are idle for a long time, and even there will be a "digital rotten building".
In addition, many small and medium-sized traditional data centers and intelligent computing centers use hardware, software, model frameworks and other server technology systems, which are not compatible with international mainstream open source frameworks such as CUDA, do not have the ability of "one cloud and multiple cores", do not support AI open source models, and require high custom development costs to achieve compatibility.
Taking the most mainstream OpenStack open source framework of private deployment as an example, this is already a backward productivity eliminated in the United States, but it is widely popular in China's data centers, OpenStack can only pre-allocate computing resources, can not achieve elastic resource scheduling of public cloud, not only cause waste of resources, but also cannot undertake ultra-large-scale computing tasks that require unified scheduling.
The consequences of the "digital rotten building" are also obvious: not only will it cause a waste of computing resources, but also invest too much in new hardware infrastructure, but ignore the industrial and ecological construction of artificial intelligence.
03
The moment of truth, the way to break the game
When some domestic large model manufacturers take the gap of only a few months from chatGPT as a gimmick of large models, the problem behind this is that the time left for us may not be as much as we imagine.
The pace of artificial intelligence development is rapid and the time window for technology to catch up is constantly shrinking. It is no longer calculated in years and months, but may be calculated in days, or even minutes and seconds.
From the perspective of the choice of OpenAI, the parent company of chatGPT, it is bound to Microsoft, which holds massive computing power resources on the cloud, not only training ChatGPT, a large model that leads a new round of global AI waves, but also providing services based on Microsoft's public cloud, allowing developers to integrate customized AI experiences into their applications. The model and the public cloud form a flywheel effect of algorithms and computing power, and operate at high speed to consolidate their first-mover advantage.
Fortunately, we are not without companies with the same competitive advantage. Large-scale public cloud vendors such as Alibaba Cloud have been deeply involved in the research of large models, and recently released the Tongyi Qianwen large model, and also connected products such as DingTalk and Tmall Genie to the large model for testing.
▲Tongyi Qianwen Big Model (Figure/Network)
Public cloud + AI is a new artificial intelligence ecology, cloud computing is not only the training and inference base of AI large models, but also the provision mode of large model services and applications, just like Alibaba Cloud initiated the Chinese AI model open source community "magic match", forming a MaaS model, through cloud computing more than 300 models open source to researchers and teams.
In the past six months, China's large model market has blossomed, not only in the aspect of "refining large models", but also in the industrial landing, technology companies led by Alibaba Cloud are also trying to build industry large models.
Not long ago, Alibaba Cloud announced that it will join forces with industry partners to create exclusive models for multiple industries such as finance, transportation, communications, energy, and power. Achieving breakthroughs in the application of large models in the industry field may also be the key to our breakthrough in the global technology competition.
On April 28, the Politburo meeting of the Central Committee pointed out that "we must attach importance to the development of general artificial intelligence, create an innovative ecology, and attach importance to risk prevention"; On May 5, the first meeting of the 20th Central Finance and Economic Commission pointed out that "it is necessary to grasp the wave of new scientific and technological revolutions such as artificial intelligence".
At this year's Davos Forum, the Prime Minister also further emphasized the importance of AI technology. Since the release of ChatGPT at the end of 2022, the ability of artificial intelligence large models to "emerge" has spawned a large number of "new species", and industrial ecology has become a key element for China and the United States to win the competition in digital technology.
What can be seen is that since the beginning of this year, the "100-model war" of Chinese technology companies has set off huge waves around the world - from the arrival of Bill Gates and Musk, to Huang Jenxun, CEO of NVIDIA, who once rumored "landing", it is enough to prove that even the leaders of key areas of global artificial intelligence still attach great importance to the innovation and changes in the Chinese market.
At this critical moment when the world is entering the next round of technological competition, if we want to continue to maintain the second largest battlefield of artificial intelligence in the world, and even the largest battlefield of global AI application innovation, we obviously need to find solutions.
And the most realistic and most likely to break the game in front of us now is to empower those companies that are already the fastest in themselves.
Although those rushing technology companies may still face many problems at this stage, there is no doubt that their rapid response and rush once again prove a problem - as pointed out in recent media editorials, who is the main force of scientific and technological progress and economic growth, and the "key Mr." of the global science and technology race.
This is our established advantage and the key to our ability to continue to seize the opportunity in the next era of AI.