{"version":"1.0","type":"rich","provider_name":"Acast","provider_url":"https://acast.com","height":250,"width":700,"html":"<iframe src=\"https://embed.acast.com/$/665dda1b3ce6480013459039/6a8737afa103c59e0286e8ab?\" frameBorder=\"0\" width=\"700\" height=\"250\"></iframe>","title":"Can Chip Matching Software Lower AI Compute Costs?","description":"<p>Bloomberg reported that Callosum raised $100 million to build software that matches AI workloads to specific chips. The company targets routing and scheduling so training and inference jobs run on hardware that fits their memory, bandwidth, and cost profiles. The raise comes as enterprises manage heterogeneous fleets across AWS, Microsoft Azure, Google Cloud, and specialized providers like CoreWeave and Lambda. The market includes established schedulers and recent deals such as Nvidia's 2024 agreement to acquire Run:ai and Databricks' 2023 purchase of MosaicML for $1.3 billion. Buyers will evaluate governance, cost allocation, and measurable utilization gains. Founders should pilot across multiple chip families and providers and track throughput per dollar, queue times, and job success rates.</p><p>Learn more on this news by visiting us at: https://greyjournal.net/news/</p><p><br></p><p><br></p>","author_name":"GREY Journal"}