The market has more than doubled since 2020, with non-x86 server sales continuing to outpace those of x86.
Servers powered by AWS’ Trainium chip could democratize access for enterprises looking to experiment with AI, but who can’t get access to or afford Nvidia’s infrastructure.
The company connects providers with excess compute capacity with those who need it, and will host its second auction beginning Mar. 3.
The company plans to spend up to $65 billion on infrastructure for AI in 2025, and is planning a data center with a footprint almost as large as Manhattan.
The cloud giant has rolled out its Trainium2 chips for general availability and is already teasing Trainium3, but it’s still early in the game, and some analysts question whether the massive compute Trainium supports is even necessary — o
Some in the industry are scratching their heads about why such staff cuts are necessary as data center spending continues to accelerate to meet the needs of AI-intensive workloads.
In an outdoor trial, SoftBank’s AI-RAN infrastructure built on Nvidia AI Enterprise achieved carrier-grade 5G performance while using excess capacity to concurrently run AI inference workloads.
The hardware provider’s stock is down nearly 20%, its auditor quit last week, and it’s facing delisting. But this has all happened before, and enterprise customers aren’t likely to be fazed, analysts say.
Data centers orbiting the earth can ease terrestrial energy constraints, allow for rapid deployment and scalability and reduce operating expenses by ‘orders of magnitude,’ company says.
