Frontier AI development relies on powerful AI supercomputers, yet analysis of these systems is limited. We create a dataset of 500 AI supercomputers from 2019 to 2025 and analyze key trends in performance, power needs, hardware cost, ownership, and global distribution. We find that the computational performance of AI supercomputers has doubled every nine months, while hardware acquisition cost and power needs both doubled every year. The leading system in March 2025, xAI's Colossus, used 200,000 AI chips, had a hardware cost of \$7B, and required 300 MW of power, as much as 250,000 households. As AI supercomputers evolved from tools for science to industrial machines, companies rapidly expanded their share of total AI supercomputer performance, while the share of governments and academia diminished. Globally, the United States accounts for about 75% of total performance in our dataset, with China in second place at 15%. If the observed trends continue, the leading AI supercomputer in 2030 will achieve $2\times10^{22}$ 16-bit FLOP/s, use two million AI chips, have a hardware cost of \$200 billion, and require 9 GW of power. Our analysis provides visibility into the AI supercomputer landscape, allowing policymakers to assess key AI trends like resource needs, ownership, and national competitiveness.

When I first saw that photo of Nvidia's new Project Digits mini PC unveiled at CES 2025, I couldn't help but notice the Apple influence -- minimalist, sleek, next to a monitor that looks like Apple's Studio Display. Apple's latest Mac Mini is revolutionary in many ways, delivering the company's impressive M4 silicon in an efficient, affordable package. Windows PCs haven't yet been able to reach the same level of design beauty and overall balanced unit. Could Nvidia's new Mac Mini-like small-form-factor AI supercomputer usher in a major disruption for Windows PCs? Let's dive into Project Digits, what it is, and what it brings to the table for the future. It's a little unfair to compare the Mac Mini and Nvidia's Project Digits, mainly because they target vastly different users and markets.

Everyone wants to be like Apple. The largest publicly traded company in the world, with a flagship product that prints money, and a cultural footprint that has reached world-historical importance: the 21st-century Ford. On a surface level, the companies that get slapped with that comparison are obvious enough. If you pump out well-made, slickly designed consumer electronics that arrive in a nice box, someone somewhere will compare you to the Cupertino giant. Dig a bit deeper, and there's more meaningful comparisons to be made.

The UK government has announced a $273 million investment to build its most powerful supercomputer yet, Isambard-AI, which will rank among the top AI supercomputers in the world when it's switched on. It'll pack thousands of NVIDIA superchips, allowing it to run more than 200 quadrillion calculations per second. Isambard-AI is expected to begin operations in summer 2024 and will be hosted by the University of Bristol. The supercomputer is being built by Hewlett Packard Enterprise and will use 5,448 of NVIDIA's GH200 Grace Hopper Superchips, NVIDIA said in its own announcement. It'll be able to achieve over 21 exaflops of AI performance, or over 21 quintillion floating point operations per second for AI applications, like training large language models.

NVIDIA CEO Jensen Hiang made a string of announcements during his Computex keynote, including details about the company's next DGX supercomputer. Given where the industry is clearly heading, it shouldn't come as a surprise that the DGX GH200 is largely about helping companies develop generative AI models. The supercomputer uses a new NVLink Switch System to enable 256 GH200 Grace Hopper superchips to act as a single GPU (each of the chips has an Arm-based Grace CPU and an H100 Tensor Core GPU). This, according to NVIDIA, allows the DGX GH200 to deliver 1 exaflop of performance and to have 144 terabytes of shared memory. The company says that's nearly 500 times as much memory as you'd find in a single DGX A100 system.

In the paper released earlier this week, Google explained how it connected over 4,000 TPUs to create a supercomputer. Google claims the supercomputers used for training its artificial intelligence (AI) models are faster and more energy-efficient than those employed by multinational technology firm Nvidia. Google researchers detailed how they created a supercomputer from more than 4,000 fourth-generation Tensor Processing Units (TPUs), as well as custom optical switches to link individual machines. The AI models are segmented across thousands of chips, which must collaboratively train the models for weeks or more. Google's Norm Jouppi and David Patterson explained, "Circuit switching makes it easy to route around failed components. This flexibility even allows us to change the topology of the supercomputer interconnect to accelerate the performance of an ML (machine learning) model."

Google recently announced they have developed a unique artificial intelligence (AI) supercomputer that is faster, more efficient, and more powerful than NVIDIA systems. Nvidia is the reigning champion of AI model training and deployment, dominating over 90% of the market, according to CNBC. The great AI race has been raging on for a while now in Big Tech, and Google has been developing AI chips called Tensor Processing Units (TPUs) since 2016. "Google has chosen a unique approach to AI development by creating its own'Tensor Processing Unit' (TPU) architecture, rather than relying on specialised GPUs [graphic processing units] from Nvidia," founder of Elo AI, Matt Falconer explains. "This decision allows Google to reduce their dependence on third-party vendors and achieve vertical integration across its entire AI stack," Falconer added.

When NVIDIA's founder and CEO Jensen Huang waxed poetic about artificial intelligence in the past, it mostly felt like marketing bluster, the sort of lofty rhetoric we've come to expect from an executive with a never-ending supply of leather jackets. But this year, following the hype around OpenAI's ChatGPT, Microsoft's revamped Bing and a slew of other competitors, NVIDIA's AI push finally seems to be leading somewhere. The company's GTC (GPU Technology Conference) has always been a platform to promote its hardware for the AI world--now it's practically a celebration of how well-positioned NVIDIA is to take advantage of this moment. "We are at the iPhone moment for AI," Huang said during his GTC keynote this morning. He was quick to point out NVIDIA's role at the start of this AI wave: he personally brought a DGX AI supercomputer to OpenAI in 2016, hardware that was ultimately used to build ChatGPT.

Jensen Huang getting carried away about an emerging technology is nothing new. This time last year, the charismatic and excitable co-founder and CEO of chip design giant Nvidia was telling anyone who'd listen about the potential of the metaverse (or the Omniverse, as Nvidia's marketing department prefers to call it). Since then, the metaverse bubble has suffered a slow puncture, and Huang is back to evangelising about one of his favourite topics: artificial intelligence. Describing the growth in power of generative AI systems like GPT-4 – the model that powers OpenAI's tools such as ChatGPT – as a "new era of computing", Huang told investors on his company's most recent earnings call that AI was at an "inflection point", stating that businesses have "an urgency to develop and deploy new AI strategies". However, Huang added that he believes many companies face "an insurmountable obstacle" in getting access to the resources and skills needed to make AI work, which is why, he says, Nvidia is getting into the services business.

New virtual machines for Microsoft Azure allow developers to create generative AI apps that can be scaled to work with thousands of Nvidia H100 GPUs. The ND H100 v5 VM series on Azure, which works in tandem with Quantum-2 InfiniBand networking, boosts the performance of large-scale deployments by companies such as OpenAI, creators of the much talked about ChatGPT, and Nvidia's chips. The new supercomputing system in the cloud provides the type of infrastructure required to handle the latest large-scale AI training models, according to Matt Vegas, principal product manager for Azure HPC and AI at Microsoft. "Generative AI applications are rapidly evolving and adding unique value across nearly every industry," Vegas wrote in a blog post this week. "From the newly released AI-powered Bing and Edge to AI-powered assistance in Microsoft Dynamics 365, AI is becoming a pervasive component of software and how we interact with it. We want to ensure that our AI infrastructure will be there to pave the way."