Home > Article > Technology peripherals > Challenges and investment strategies faced by data centers in the AI era
Artificial intelligence applications must be supported by massive computing power, which means larger and richer data centers.
As the application of artificial intelligence increases, the data center market is growing rapidly to accommodate the surge in data caused by these technologies. Adding artificial intelligence to the already vast array of technologies available, including Internet of Things (IoT) devices, will generate even more customer data, causing data volumes to grow exponentially.
The bottom line is that all this data needs to reside somewhere, and organizations will move to data centers.
Kevin Shtofman, director of innovation at Cherre, explained that artificial intelligence will increase the demand for computing power, requiring investment in artificial intelligence-specific hardware, adopting new data center designs, and exploring emerging technologies such as edge computing.
“Artificial intelligence applications require a lot of computing power when training complex deep learning models,” Shtofman said. As artificial intelligence becomes more popular, more data centers will be needed to support the growing demand for computing power. ”
The adoption of artificial intelligence will also increase data storage requirements, as AI-driven applications require large amounts of data to train and improve models.
According to Shtofman, it needs to be stored and accessed quickly This data requires a large storage capacity. Therefore, data centers will need to expand their storage capabilities to meet the growing demand."
Shtofman added that as artificial intelligence applications become more and more widespread, There is also an increasing need for real-time processing and decision-making. This has given rise to the rise of edge computing, which involves processing data closer to the source rather than sending it to centralized data centers. Therefore, more data centers need to be built closer to the edge to support this trend.
Andy Cvengros, managing director of technology at Jones Lang LaSalle, pointed out that as the capabilities of artificial intelligence are integrated with daily technology functions, artificial intelligence at the consumer level is expected to Intelligence will explode. "As usage becomes more common, this will lead to a huge demand for data center computing power," Cvengros said. "Running and training these models requires a lot of computing power and a lot of resources, which limits the enterprises that can make breakthroughs." Quantity.”
The density of server computers required for AI also generates large amounts of heat, and to address this, innovations in liquid cooling are developing. To meet this growing demand, cloud computing companies are actively looking for development projects to acquire hundreds of megawatts of power in just a few years.
According to Cvengros: "The available power capacity in the primary data center market has been exhausted, and the secondary and tertiary markets can take this opportunity to expand.". ”
Cvengros pointed out that major cloud computing companies are adopting self-build and lease data center models. Hyperscale cloud users and colocation providers are scrambling to find high-performance land sites in almost all markets to support These are huge capacity needs.
Ten years ago, a data center requiring 10 megawatts was considered quite large, but by 2023, it is not uncommon to announce the construction of data centers exceeding 100 megawatts. Cvengros said: “When hyperscalers are unable to build data centers in a specific market due to land, power or supply chain constraints, they may rent the entire data center from a colocation provider, making it difficult for smaller businesses with smaller needs. Difficulty finding enough space. ”
Shtofman said that the main stakeholders in ensuring that data centers grow with the demand generated by artificial intelligence computing are data Center operators, cloud service providers, hardware manufacturers, government and regulatory agencies, as well as data scientists and artificial intelligence researchers.
Data center operators are responsible for managing and maintaining the physical infrastructure of supply-side data centers. The provider's AI applications can be supported by the computing resources and infrastructure provided on demand by the cloud service provider. To meet the demands generated by AI computing, including computing power, storage and network capabilities, they must ensure that they have sufficient capabilities.
At the same time, hardware manufacturers are responsible for designing and producing the specialized hardware required for artificial intelligence computing on the supply side, such as graphics processing units (GPUs) and tensor processing units (TPUs).
Shtofman "They must ensure adequate supply of these specialized components to support growing demand," said the company. This is a higher risk given recent issues with global supply chains. ”
Cvengros agreed that components needed to build and operate data centers have been delayed due to supply chain challenges and geopolitical tensions during the pandemic. This has delayed construction timelines, but as demand remains strong , users have shifted to pre-leases.
It is expected that the majority of new supply of vacant pipelines by the end of 2023 or 2024 will be pre-leases.
In Cvengros’ view, maintaining until safety needs are met Suppliers with large supply chain inventories will stand out in the competition to win hyperscale business.
Before expanding a data center, it is crucial to conduct in-depth research and analysis of the market and artificial intelligence computing needs, which is what Shtofman emphasized. "This will help demonstrate that the investment is aligned with market needs and that there is a clear path to return on investment. This market appears to be very prosperous, involving multiple transportation modes and role types, and therefore requires the use of edge computing technology."
It is recommended to develop a comprehensive strategy and update it frequently, because this market changes much faster than other cycles. "Data center assets require very specific infrastructure, design and compliance with local laws. For newbies, working with an experienced partner is best practice and this type of construction is not for them."
The above is the detailed content of Challenges and investment strategies faced by data centers in the AI era. For more information, please follow other related articles on the PHP Chinese website!