Microsoft has already contracted GPUs to stability prices | JP
Among the insights revealed during the latest quarterly earnings call from Microsoft is that it is de-risking its artificial intelligence (AI) acceleration expansion, with customers already contracted to buy graphics processor unit (GPU) capacity.
The company reported that Microsoft Cloud revenue surpassed $50bn for the first time, growing 26% year over year (YoY). Overall, it reported revenue of $81.3bn, up 17% YoY.
Microsoft said that it has expanded global datacentre capacity, adding nearly one gigawatt of total capacity this quarter and stating there has been a10x increase in growth of Microsoft 365 Copilot daily users. The company said it is making continued investment in AI infrastructure to meet the growing demand for Azure, M365 Copilot, and GitHub Copilot.
When asked about hardware capital expenditure, chief financial officer Amy Hood said: “The majority of the capital that we’re spending today, and a lot of the GPUs that we’re buying, are already contracted for most of their useful life.”
She said that the Microsoft 365 subscription has a faster payback when compared to Azure, which has a mix of compute usage and GPU. Looking at how Microsoft de-risks its GPU investments, she said GPU contracts are sold for the entire useful life of the GPU.
Microsoft’s AI infrastructure is being used to support heterogeneous and distributed workloads. The company’s stated goal is to optimise for “tokens per watt per dollar”, which involves increasing hardware utilisation and reducing total cost of ownership (TCO) through advancements in silicon, systems and software.
Microsoft said that it is seeing a six-year useful life for the servers and GPUs in its datacentre fleet. Expanding on Hood’s response, CEO Satya Nadella discussed the benefits of software optimisation in Microsoft datacentres in extending server life.
“We use software to continuously run even the latest models on the fleet,” he said. “That’s what gives us [longevity]. That’s why we think about ageing the fleet constantly. It’s not about buying gear one year – it’s about each year you ride Moore’s Law, you add and use software, and then you optimise across all of it.”
Following on from the announcement of Microsoft’s Maia 200 AI acceleration hardware, Nadella was asked if the company would be putting more focus on silicon. In response, Nadella spoke about the opportunity to optimise workloads: “When you have a new workload, you can start innovating end to end between the model and the silicon and the entire system. It’s just not even about just the silicon”
According to Nadella, innovation includes the way the networking works at rack scale, which needs to be optimised to speed up memory access needed by the workload.
Alongside the investment in silicon and a strategy to increase the longevity of server hardware, Microsoft has seen a shift in what customers are looking for from cloud providers.
“Sovereignty is increasingly top of mind for customers, and we are expanding our solutions and global footprint to match,” said Nadella. “We announced DC investments in seven countries this quarter alone, supporting local data residency needs. And we offer the most comprehensive set of sovereignty solutions across public, private and national partner clouds, so customers can choose the right approach for each workload, with the local control they require.”
Discussing the earnings call, senior analyst Dario Maisto said: “The earnings call brought us back to the fundamentals after months of hype. In the cloud space, customers are now much more cautious about what they buy and how they pay for it.
“As an effect of digital sovereignty concerns, clients are moving from a best of breed and hyperscaler-only approach to more selective and price-sensitive buying behaviour, and are now much more open to considering local players than they have in the past.”