Tag
The 24 August 2026 launch says Ray on SageMaker HyperPod needs a cluster and the KubeRay operator. The dashboards, IDE attach, Grafana panels and KV cache each need a separate add-on.
Manu Shukla 
A decision guide to NVIDIA Nemotron-Labs Diffusion: one checkpoint, three serving modes, the real B200 and DGX Spark throughput numbers, and when tri-mode serving pays off versus autoregressive.
Manu Shukla 
Kimi K3's API costs $3/$15 per million tokens. Self-hosting the 2.8T open weights needs roughly 1.4 TB of VRAM. We size the GPUs, price the node against 2026 cloud rates, and work out the break-even.
Manu Shukla 
Kimi K3 (2.8T), DeepSeek V4 (1.6T) and GLM-5.2 (753B) are all open-weight and frontier-class. This decision matrix compares GPU footprint, API list price, and the monthly output-token volume where self-hosting each one
Manu Shukla 
A 2026 cost-per-token comparison of the NVIDIA B200 and H100 for LLM inference, with MLPerf v6.0 throughput, real cloud hourly rates, the arithmetic that makes the pricier chip cheaper per token, and the cases where
Manu Shukla 
Thinking Machines released Inkling on 15 July 2026 under Apache 2.0: 975B parameters, 41B active. We priced the GPUs against the Tinker API using AWS and vendor figures, and the break-even is brutal.
Manu Shukla 
In the 2026 State of FinOps, 98% of teams now manage AI spend and GPU cost has become the top concern. Here is what changed and how enterprises can regain control of the AI cloud bill.
Manu Shukla 
Companies waste about 27% of cloud spend, and AI is now the fastest-growing line. Six FinOps ways Indian teams cut AWS, Azure and GCP bills in 2026, from commitment discounts to cutting the GPU and token bill.
Manu Shukla 