Glossary / Compute

Compute

The processing power used to train and run AI models, supplied by chips such as GPUs in large data centers. Access to compute is one of the main limits on AI progress.

In the news

• Google launches Gemini Decisions API for multimodal decisions1h ago
• Anthropic starts IPO roadshow with shares priced at $70 to $781d ago
• Google announces TPU 8tx and TPU 8i Max systems2d ago
• Google launches AlloyDB Express in preview with fast setup and automatic pausingNov 4
• Anthropic reports $23.3 billion quarterly loss despite revenue surgeNov 2
• Google and Oracle plan to bring TPUs to Oracle Cloud in 2027Oct 27
• SpaceX secures roughly $40 billion to lease AI chips from ApolloOct 26
• IBM brings its Spyre accelerator into PyTorch workflowsOct 26
• Google introduces Agent Compute Engine and expands agent governanceOct 22
• Flapping Airplanes raises $1 billion Series A at reported $6 billion valuationOct 22
• Nvidia expands open AI models for business data, robotics and scienceOct 21
• Nvidia and Google Cloud announce A5X availability and confidential computing expansionOct 21