SILICON VALLEY — The artificial intelligence boom is driving a surge in demand for cloud computing. According to data released in April by market research firm Synergy Research Group, off-premise data centers — a category combining self-built facilities, leased capacity and colocation, which provides only space and basic infrastructure — accounted for 68% of global data center capacity as of the end of last year. Hyperscalers such as Amazon Web Services, Microsoft Azure and Google Cloud held 48% of that total, while colocation data center operators such as Equinix accounted for 20%.
The share held by hyperscaler cloud computing data centers stood at just above 20% in 2018, but more than doubled in seven years, driven by the spread of remote work during the COVID-19 pandemic after 2020 and a surge in AI development demand following OpenAI's release of ChatGPT in November 2022. As the cost of building and operating data centers climbed and hyperscalers used their formidable capital to sweep up the world's most advanced graphics processing units, companies concluded they were better off using cloud computing data centers.
Over the same period, on-premise facilities ceded market leadership to cloud computing. On-premise capacity made up 56% of data center capacity in 2018 but shrank to 32% last year. By 2031, hyperscale operators are expected to hold 67% of total capacity, while on-premise is projected to fall to 19%. The on-premise share of data center capacity is forecast to decline by 2 percentage points a year.
Cloud data centers involve connecting to facilities installed outside a company over the internet, while on-premise refers to facilities a company builds internally. On-premise data centers are also called local or edge computing facilities. Companies once built and ran their own data centers, but as cost efficiency and specialized operational management have grown more important, they have increasingly relied on the infrastructure of dedicated cloud computing firms rather than running facilities themselves.
Yet the fact that cloud computing leads the data center market does not mean on-premise has become less important. Cloud growth has simply been so steep that it has overshadowed on-premise, which is also expanding. According to market research firm Fortune Business Insights, the global edge computing market is projected to grow from $18.64 billion last year to $25.63 billion this year, and to continue its steep climb to $267.42 billion by 2034. That is smaller than the cloud computing market, projected at $2 trillion to $3 trillion by 2030, but analysts say it is no slower in growth.
Companies are indeed recognizing the importance of on-premise. In a perception survey of companies conducted last year by the Uptime Institute, a data center research organization, 45% of IT workloads ran on company-owned infrastructure, while off-premise accounted for 55% — unchanged from a year earlier. That suggests companies are striking a balance in how they distribute workloads between cloud and on-premise even as the cloud computing market expands each year. Why are companies that had been switching from on-premise to cloud turning back to on-premise?
Cloud's Fatal Weakness: Vulnerable Security
The biggest reason on-premise is being reassessed is its strength in security. Because cloud customers connect to external data centers over the internet to use computing infrastructure, they are vulnerable to hacking and other outside attacks. AI companies such as OpenAI and Anthropic develop AI models and run applications in the cloud. The risks of cloud-based AI came into focus recently when an OpenAI AI agent was found to have escaped its developer's controlled environment, hacked external data and even accessed government and United Nations websites.
Edge computing, by contrast, means processing data at the edge of the network, as the term suggests. It can be handled on the device itself or on a local server without sending data outside. That reduces concerns about confidential information leaking during transmission or security incidents caused by hacking.
AI software companies are aggressively targeting demand from firms that prioritize cybersecurity. Splunk, which operates an AI data management platform, unveiled Cisco AI POD for Splunk at its .conf26 conference in Denver, Colorado, on the 14th. Built on the infrastructure of its parent company Cisco, the on-premise AI product allows fast processing without moving data outside. Splunk is working with Nvidia, the world's top AI accelerator company, for customers seeking large-scale, high-performance model inference without compromising data sovereignty. Internal infrastructure is strong on security but slower at AI processing than external clouds equipped with vast computing capacity. To address that, Splunk is adopting on-premise accelerated computing using Nvidia hardware and software.
Splunk operates an AI platform that collects, classifies and analyzes machine data generated automatically by software without manual human work. Its customers — public institutions, regulated companies and multinationals — have long stored data in on-premise facilities they built themselves for security reasons, because sending data from an internal network to an external cloud can expose it to hacking. Kamal Hathi, senior vice president at Splunk, said the product lets organizations run managed AI on sensitive machine data in on-premise and private cloud environments without sacrificing data sovereignty.
Lives Can Be Lost in an Instant: Physical AI Makes It Essential
As physical AI advances, on-premise is also growing in importance. Physical AI refers to the application of AI to physical reality, with self-driving cars, humanoid robots and robotic medical devices among the leading examples. Because physical AI operates in areas directly tied to human life or closely linked to safety accidents, development and operation based on on-premise infrastructure matter. No matter how fast a network is, the cloud requires time to travel to and from external data centers, so it cannot match on-premise on low latency.
Physical AI is a leading field where on-premise AI, or edge AI, is realized. Computing equipment or devices that serve as data centers are built as close to the end user as possible. Edge AI runs AI where data is generated and helps devices make decisions in milliseconds, or thousandths of a second, even without an internet connection. Local processing minimizes transmission delays, reduces bandwidth consumption, protects sensitive data and improves operational responsiveness.
Self-driving cars are a leading example of an on-premise-based field. Because a machine rather than a person is driving, it must make split-second decisions comparable to human instinct when a risk of an accident arises. Most processing, including navigation and obstacle detection, takes place inside the vehicle. At the stage of full autonomy with no human involvement, even a delay of milliseconds in the machine's judgment can lead to a serious accident, so processing must be handled quickly on an on-premise basis.
The same applies to medical robots and humanoids. In the smart analytics and robotic control functions increasingly used in operating rooms, for instance, low latency, network stability and maintaining bandwidth — the volume of data transmitted per unit of time — are critical. Because even the slightest error there can put a patient's life at risk, edge computing is essential.
The Era of Sovereign AI
Another reason on-premise is drawing attention is the emphasis companies and governments are placing on sovereign AI, meaning AI sovereignty. Sovereign AI refers to efforts to secure AI infrastructure and models independently so as not to depend on a handful of hyperscalers or AI model developers. As the AI market has come to center on the United States and China and on big tech firms such as Anthropic, OpenAI, Google and Nvidia, efforts to secure AI sovereignty have continued in South Korea and elsewhere.
Even hyperscalers are supporting on-premise environments to make up for the cloud's limits. In March, Microsoft announced an edge computing solution in partnership with Armada, an edge computing company. It is a strategic attempt by Microsoft to combine the strengths of cloud and edge computing to build a Sovereign Private Cloud. Douglas Phillips, chief technology officer at Microsoft, said in a press release that sovereign cloud capability has become a necessity rather than a choice as organizations accelerate digital transformation, adding that Azure Local will expand into the new domain of edge computing and provide the security governments and enterprises need to operate independently while meeting sovereignty and other core requirements.
