As the year 2025 dawns, the world's attention has converged on artificial intelligence (AI).

Following China's DeepSeek's stunning debut that captivated the global tech community, it seems the entire world has abruptly embarked on a spree of launching large models and constructing data centers.

Elon Musk revealed that xAI will soon unveil its own model, potentially surpassing even DeepSeek; the French AI startup Mistral AI recently introduced Le Cha, heralded as the world's fastest "AI Super Assistant," outpacing ChatGPT by 13 times; and the Indian government has announced plans to develop a domestic large language model within the next 10 months to rival DeepSeek and OpenAI.

The proliferation of large models necessitates substantial computational power, prompting a global race to establish data centers. In January of this year, shortly after taking office, President Trump announced the "Stargate" plan: a $500 billion investment over the next four years to construct up to 20 ultra-large data centers.

Similarly, in January, British Prime Minister Keir Starmer announced an ambitious "AI Opportunity Action Plan" aimed at transforming the UK into an "AI Superpower." To this end, tech companies have pledged to invest £14 billion and create 13,250 new jobs, including significant data center projects in Wales and Liverpool.

France, a UK neighbor, hosted a two-day AI Action Summit recently. On the eve of the summit, French President Macron announced that France will invest over €109 billion (approximately $113 billion) to create a French version of "Stargate." Currently, 35 locations in France have been identified for data center construction. Macron hopes to establish Europe as a leader in the field of artificial intelligence (AI).

India is also heavily investing in AI. In January, the Reliance Group, led by India's richest man Mukesh Ambani, announced plans to build a super data center with the world's largest capacity in Jamnagar, India, with a total investment of about $20-30 billion.

At the ongoing LEAP exhibition in Saudi Arabia, Datavolt signed a $5 billion agreement with Neom to construct one of the world's largest AI data centers. AI development in Middle Eastern countries is also accelerating.

Will 2025, with its rapid advancements, truly mark the year that AI transforms the world?

DeepSeek's meteoric rise was unexpected. On the morning of January 27, DeepSeek simultaneously topped the free charts on the Apple App Store in both China and the US, surpassing the previously dominant ChatGPT. Just half a month earlier, the DeepSeek app had been launched on iOS and Android app markets.

This is the first time an app from China has achieved such a feat, sending shockwaves through the tech world.

The DeepSeek-R1 model rivals OpenAI's GPT-4 in various capabilities, yet its total training cost is only $5.576 million, compared to GPT-4's approximately $100 million training cost.

The world began to truly appreciate China's explosive potential in the realm of large AI models—even with limited resources, China can create its own large models by optimizing large model infrastructure.

Subsequently, multiple countries exhibited signs of "panic": Australian Treasurer Jim Chalmers urged caution when using DeepSeek and banned its use in government systems and devices; South Korea tightened regulations on DeepSeek-related apps, and relevant departments began reviewing companies using DeepSeek; France and Ireland inquired about DeepSeek citing data security concerns; Italy directly announced a ban on DeepSeek; and the US Congress introduced a new bill stipulating that downloading DeepSeek could result in a maximum sentence of 20 years in prison.

Beyond panic, there is a growing "awakening."

ChatGPT has long been the benchmark for global large models, but it has had to adapt in response to DeepSeek's popularity. On January 24, OpenAI made the ChatGPT O3 Mini version, which originally cost $200 per month to subscribe, free of charge. On February 1, OpenAI also launched its latest inference model, GPT-4 o3-mini. Similar to DeepSeek, GPT-4 o3-mini incorporates internet search functions and displays the thought process. Furthermore, ChatGPT has opened up the GPT-4 o3-mini model to all users for free, with ChatGPT Pro users enjoying unlimited access, and Plus and Team users having their message limit increased from 50 per day to 150.

After experiencing DeepSeek recently, Musk also stated, "You can expect China to do many great things, and DeepSeek is one of them." However, he seems most eager to highlight that xAI, which he founded, will soon release its own model, potentially even better than DeepSeek. This new model, named Grok-3, is said to have completed its pre-training using ten times more computing power than Grok-2. Greg Yang, a mathematician at xAI, also demonstrated a conversation with Grok-3 on the platform on January 19 this year. Various signs indicate that Musk is actively engaging in this large model battle.

On February 6 this year, French AI startup Mistral AI also released the iOS and Android versions of Le Cha, billed as the world's fastest "AI Super Assistant." It can access the internet in real-time and possesses multimodal capabilities. Not only is it free, but it is said to be 13 times faster than ChatGPT in terms of response speed. In just a few days, Mistral soared to the top of the French App Store. French President Macron has also mentioned the app's name, Le Cha, multiple times in a recent program, acting as a "salesman" and urging the French to "download" it.

Influenced by the current AI boom, on January 30 this year, the Indian government stated that it plans to develop a domestic large language model within the next 10 months to challenge DeepSeek and OpenAI. India's Minister of Information Technology, Ashwini Vaishnaw, said that with foundational work completed, the Indian government is currently focused on creating an AI system that meets India's unique needs. It is expected that by the end of this year, six major developers will launch foundational AI models. According to Indian media, the Indian government has approved 18 proposals aimed at accelerating the implementation of AI solutions in key areas such as agriculture and climate change. These support measures include providing computing power, data, and funding.

Earlier, on December 17, 2024, the Technology Innovation Institute (TII), supported by the UAE government, announced the launch of Falcon 3, the latest generation of open-source small language models (SLM) developed by the institute. Compared to large language models (LLM), SLMs have fewer parameters, a simpler design, and offer advantages such as high efficiency, low cost, and the ability to be deployed on devices with limited resources. In other words, they can efficiently run on lightweight infrastructure like laptops.

Of course, before this, the UK, Japan, South Korea, Russia, Canada, and Israel also had their own large models. Today, this competition has reached a new climax.

Muath Alduhishy, Executive Officer of Advanced Research, Development, and Innovation (RDI) at Saudi Information Technology Company (SITE), used the term "Sovereign AI" to summarize this development trend. This underscores that AI has now transcended the realms of technology and business, evolving into a form of "Sovereign AI".

In the current global AI landscape, what kind of structure do countries find themselves in?

Stanford University, a leading institution in the global AI field, released the "Global AI Power Rankings" on November 21, 2024, comprehensively assessing and ranking the AI activity of 36 countries through 42 indicators.

Global AI Vitality Rankings: Top 10 Countries in 2023

The United States ranks first in the total weighted index score. In terms of developing well-known machine learning models, the US reached 61 models in 2023, with notable examples being ChatGPT from OpenAI and Gemini from Google. Additionally, the US has been proactive in AI regulation, passing a total of 23 AI-related laws since 2017.

China ranks second, excelling in research and development, economy, and infrastructure. Benefiting from a vast market, robust network and computing power infrastructure, and data support from large-scale industrial production capacity, China has emerged with numerous excellent large models, including those from major companies like Baidu, Alibaba, Tencent, ByteDance, Huawei, and iFLYTEK, as well as models developed by startups like DeepSeek, Kimi, and MiniMax.

Furthermore, the report states that China shows strong growth momentum in AI innovation, particularly in patent applications, with the number of AI patent grants almost tripling that of the US. This is a crucial support for China's AI large models to swiftly transition from following to original innovation.

The UK ranks third in the score, with outstanding performance in research and development, education, policy, and governance. India ranks fourth, and the UAE ranks fifth. France ranks sixth, with strong performance in policy and governance, education, and infrastructure. South Korea ranks seventh; in October 2023, South Korean telecom company KT released the country's first AI model, Mi:dm, but South Korea's AI large models are mainly dominated by large chaebol companies, lacking an active startup ecosystem. Germany ranks eighth and is a significant contributor to AI research, ranking fourth in producing well-known machine learning models. South Korea, along with Japan and Singapore, ranks in the top ten, highlighting the growing importance of AI in Asian economies.

Stanford University stated that the performance of different regions in this ranking highlights the global nature of AI and the diverse strategies adopted by countries to promote the development and deployment of this field. Many national leaders have recognized the geopolitical significance of AI and are striving to enhance their country's AI development level.

Changing Competitive Dynamics of Global AI Vitality from 2017 to 2023

The above chart illustrates the evolving competitive dynamics of global AI vitality from 2017 to 2023. It reveals three distinct competitive tiers: The first tier is consistently dominated by the US and China; the second tier is relatively stable, including the UK and India; the third tier is more volatile, with countries like France, Germany, Japan, Singapore, and South Korea frequently swapping positions.

Undoubtedly, the US and China have become the epicenter of global AI competition, with a widening gap between them and latecomers.

France Invests €109 Billion to Create a French Version of "Stargate"

From February 10 to 11, the Global AI Action Summit was held in Paris, the capital of France. Representatives from about 100 countries attended the summit, including heads of state, business leaders, scientists, and civil society members, to discuss the application and global governance of AI.

At the conference, French President Macron announced that France will invest nearly €109 billion in AI development to create a French version of "Stargate." This includes a strategic framework agreement announced jointly by France and the UAE on February 6 to build a 1-gigawatt (GW) AI data center in France, with an estimated investment scale of $30-50 billion. Additionally, Canadian investment giant Brookfield Asset Management plans to invest €20 billion in French AI infrastructure by 2030 to help Europe catch up with China and the US in AI.

Macron believes that if France and Europe do not want to lag behind in global AI development, they need to accelerate their efforts. He hopes that Europe will become a leader in the field of artificial intelligence (AI), and the first battle is about "investment, investment, investment." Currently, 35 locations in France have been identified for data center construction. According to France's 2030 plan, the French government will invest an additional €400 million to support nine AI clusters, including Rennes, Saclay, Grenoble, and Toulouse. These clusters will undertake interdisciplinary research and provide additional training.

The United States Invests $500 Billion to Build the “StarGate”

The “StarGate” is a plan proposed by U.S. President Trump shortly after taking office in January this year, hailed as the “largest artificial intelligence infrastructure project in history.”

The plan entails OpenAI, SoftBank of Japan, and Oracle Corporation of the United States establishing a joint venture called “StarGate” and investing $500 billion in AI infrastructure in the United States over four years.

In fact, since the surge in popularity of ChatGPT, demand for high-performance computing and storage has soared, and the construction of data centers in the United States has continued to heat up, becoming the hottest commercial real estate in the country. Technology giants such as Google, Microsoft, and Meta have entered the market, investing billions of dollars in data center expansion.

China Started the “New Infrastructure” for Data Centers Four Years Ago

On the other hand, China proposed the “New Infrastructure” plan, which includes data centers, as early as 2020. Since then, Alibaba Cloud announced an investment of 200 billion yuan over three years to build data centers, Tencent invested 500 billion yuan over five years to lay out new infrastructure, and Baidu plans to increase the number of intelligent cloud servers to 5 million within 10 years.

To harmonize the distribution of energy, data, and other resources, the country has unveiled the "East Data, West Compute" initiative, which effectively merges eastern data hubs with western energy strengths, setting a pivotal strategic move for the transformation of the data center industry.

India Invests $30 Billion to Build Data Centers

As a co-host of the AI Action Summit in Paris, India has recently made significant strides in data center construction.

On January 31, India's Reliance Group announced plans to invest between $20 billion and $30 billion to establish a data center in Jamnagar with a total installed capacity of 3GW. If implemented as planned, this project will become the world's largest data center in terms of capacity.

This ambitious data center plan has garnered the attention of numerous international investors, including technology giants like Microsoft, Google, and Amazon, who have increased their investments in Indian data centers. For instance, Microsoft announced in early January 2025 that it would invest $3 billion in India over the next two years to build new data centers and propel the development of AI technology. Similarly, Amazon Web Services announced on January 23 that it would invest $8.3 billion in cloud infrastructure in the Indian state of Maharashtra as part of a broader $12.7 billion investment plan in India by 2030.

Previously, India lagged behind in the number of hyperscale data centers. As of 2023, India had only about 18 hyperscale data centers, compared to nearly 100 in China. However, India's data center market is expected to experience substantial growth in the coming years. According to data from India's Ministry of Electronics and Information Technology, by 2026, India's data center capacity is projected to increase from the current 819 megawatts (MW) to 1,800 MW.

With the intensifying competition for large models, the global race to build data centers has reached a fever pitch.

Saudi Arabia Welcomes Billions of Dollars in New Investments in Data Centers

While the AI Summit was underway in Paris, the fourth LEAP Exhibition was also being held in Riyadh, Saudi Arabia. At this year's LEAP Exhibition, artificial intelligence and the digital economy emerged as the hottest topics. During the conference, Cliff Chau, managing partner of IDG Capital, joined Hazman Hilmi Sallahuddin, chief investment officer of the Malaysian Civil Servants Retirement Fund (KWAP), and other guests to discuss "Asia-Middle East Cross-border Investment: Technology and Digital Infrastructure Collaboration." Notably, Saudi Cloud Computing Company (SCCC), a leading cloud service provider in Saudi Arabia, was jointly established by IDG Capital, Saudi Telecom, and Alibaba Cloud.

At the LEAP Exhibition, several global technology giants announced new plans in Saudi Arabia: Tencent Cloud announced that it would invest over $150 million (approximately 1.096 billion yuan) in the future for infrastructure, resources, and related investments and construction in the Middle East, with plans to build its first Middle East data center in Saudi Arabia. Equinix, a U.S. data center giant, announced an investment of $1 billion to construct a 100MW AI data center. Additionally, Datavolt signed a $5 billion agreement with Neom to establish one of the world's largest AI data centers.

During this period, Lenovo also held a groundbreaking ceremony for its new manufacturing base in Riyadh, which will produce one million computers and servers annually and commence official production in 2026.

The United Arab Emirates Deploys Data Centers in France and Colombia

Also on February 11, the World Government Summit 2025 commenced in Dubai, United Arab Emirates. Artificial intelligence and the digital economy were among the core topics discussed at the summit. During the conference, Baidu founder Robin Li engaged in a notable dialogue with Omar Sultan Al Olama, Minister of AI of the United Arab Emirates. Li emphasized the need for continuous investment in chips, data centers, and cloud infrastructure to create better and smarter next-generation models. In response, Al Olama expressed his anticipation for next year's conference, where he envisioned all vehicle fleets being supported by self-driving vehicles from Luobo Kuaipao.

Recently, the United Arab Emirates has also been intensively deploying data center construction. In addition to the AI data center recently announced with France, on February 5, the United Arab Emirates signed a cooperation agreement with Colombia to assist in establishing three strategic data centers in the northern city of Santa Marta.

The United Kingdom Lists Data Centers as “Critical National Infrastructure”

On January 13, 2025, British Prime Minister Starmer unveiled the UK government's artificial intelligence action plan, aiming to establish the UK as a "world leader" in this field. The government has committed to increasing the country's computing power by 20 times by 2030 and building new supercomputers to meet the computing demands of AI products.

Concurrently with the announcement of this plan, several large technology companies pledged a total investment of 14 billion pounds (approximately $17.4 billion) in the UK to construct the necessary AI infrastructure. For instance, Nscale, an AI infrastructure company headquartered in London, announced plans to invest $2.5 billion (2 billion pounds) in the UK data center industry over the next three years. Last October, four major U.S. technology companies - ServiceNow, CoreWeave, CyrusOne, and CloudHQ - announced a joint investment of 6.3 billion pounds (approximately $8.2 billion) in data center construction in the UK.

In September 2024, Peter Kyle, the UK's Minister of Science, Innovation, and Technology, announced that the government had classified UK data centers as "critical national infrastructure." The UK ranks third globally in terms of the number of data centers, with most concentrated in London and along the M4 motorway. However, these areas have significant energy demands, leading to delays or stoppages of projects due to insufficient power capacity.

The aforementioned global data center construction plans for several major countries alone will cost $700 billion, highlighting the fierce competition surrounding the AI leadership role.

Nonetheless, the greatest challenges lie ahead.

While the importance of data centers is evident, more and larger data centers do not necessarily equate to better outcomes.

Essentially, data centers sell electricity - in the current operating costs of data centers, IT electricity costs account for over 30%, and environmental electricity costs (cooling, lighting, etc.) constitute about 20%, amounting to a combined proportion exceeding 50%. In some older and smaller data centers, as the Power Usage Effectiveness (PUE) increases, electricity costs can surpass 60% of operating costs.

The upper limit of data center project supply in a region is contingent upon the supporting power facilities in that area, which is constrained by grid planning.

Taking the United States as an example, data centers in the country are predominantly located in northern Virginia. Currently, the state's high-voltage transmission line capacity has reached its limit, causing delays in new data center projects. To address this, Virginia has passed a bill to expedite the approval of a $60 million transmission project, a move supported by several industry giants, including Amazon.

According to Industrial Information Resources Inc. (IIR), it is estimated that by 2030, the power demand of data centers in the United States will more than double from 17GW in 2022 to 35GW.

To cope with this escalating power demand, data center operators are continuously advancing technology. For instance, by utilizing advanced cooling technology and renewable energy, data centers are progressively moving towards green and environmentally friendly development. Many data centers are also starting to harness wind and solar energy to reduce their reliance on traditional energy sources.

Furthermore, the rapid development of the data center industry has sparked debates regarding land resources and environmental protection. For example, Microsoft's new project is anticipated to result in the loss of nearly 8 acres of wetlands and over 1,600 feet of stream channels, a situation that has garnered the attention and discussion of environmental organizations. In the future, balancing environmental protection with meeting data center construction needs will become a crucial issue for industry development.

A few years ago, China's data centers also encountered growth bottlenecks due to energy and environmental concerns. However, since 2022, the country has coordinated resource planning and implemented the significant "East Data, West Compute" project, completing the overall layout design of the national integrated big data center system. This innovative approach combines eastern data with western energy to address these challenges.

In contrast, overseas countries currently have limited cases of coordinated data center planning. Especially in countries where power supply cannot be maintained at a large and stable level, blindly constructing large data centers may lead to an energy crisis and impede the healthy development of the data center industry.

Moreover, the paramount value of data centers lies in promoting the development of the digital economy. However, the overseas digital economy's development is not achieved overnight. Blindly constructing large data centers without considering the local level of digital economic development may result in issues such as vacant data center rooms and insufficient rack occupancy rates.

In fact, in recent years, domestic data center-related companies have accelerated their overseas expansion. Leading cloud service providers such as Tencent Cloud, Alibaba Cloud, and Huawei Cloud have been deploying data centers in markets like Southeast Asia and the Middle East for several years. Additionally, third-party IDC service providers like Qinhuai Data, 21Vianet, Sinnet, Global Data, and Runze Technology have also ventured abroad to invest, construct, and operate data centers overseas.

Compared to European and American data center giants, Chinese IDC enterprises are rapidly expanding in emerging markets and deeply integrating into the construction of local digital economies, thereby fostering their development.

It is foreseeable that in the global AI competition, China's new infrastructure experience will demonstrate even greater value.

References:

[1] Stanford University Release: "Global AI Power Rankings", Global Technology Map