Recent Episodes

Microsoft’s introduction of the Maia 200 adds to a growing list of hyperscaler-developed processors, alongside offerings from AWS and Google. These custom AI chips are largely designed to improve inference efficiency and optimize internal cost structures, though some platforms also support large-scale training. Google’s offering is currently the most mature, with a longer production…

OpenAI’s partnership with Cerebras has raised questions about the future of GPUs in inference workloads. Cerebras uses a wafer-scale architecture that places an entire cluster onto a single silicon chip. This design reduces communication overhead and is built to improve latency and throughput for large-scale inference. Mark Jackson, Senior Product Manager at QumulusAI, says…

NVIDIA’s Rubin GPUs are expected to deliver a substantial increase in inference performance in 2026. The company claims up to 5 times the performance of B200s and B300s systems. These gains signal a major step forward in raw inference capability. Mark Jackson, Senior Product Manager at QumulusAI, explains that this level of performance is…