Krutrim Cloud
Ola's Indian AI cloud, renting GPU capacity and serving mostly third-party open models through an AI Studio inference API, billed in rupees from Bengaluru and Hyderabad. Krutrim paused its own chip and foundation-model programmes during a 2025-26 restructuring to concentrate on the cloud business.
Pros
- Rupee billing with itemised invoices and Indian data residency across Bengaluru and Hyderabad, which is the entire point for teams with local compliance requirements. On-demand GPU is transparently priced at INR 189/hr for an A100 80GB and INR 213/hr for a single H100, scaling to INR 852/hr for four H100s, with Kubernetes-managed AI Pods from INR 24/hr and reserved discounts at monthly, six-month and one-year terms. AI Studio serves Llama 4 Scout and Maverick, Llama 3.3 70B, Gemma 3 27B, DeepSeek R1 and its distills, Whisper and Stable Diffusion 3, and signup is self-serve with no credit card required.
Cons
- The model side of the business has effectively been dismantled: chip design and foundation-model work were paused in the 2025-26 restructuring, headcount fell from over 550 in August 2025 to roughly 150 to 160 by March 2026, the Kruti consumer assistant was pulled from app stores in April 2026 without announcement, and Krutrim-spectre-V2 is the only in-house model left in the catalogue, making this an open-model reseller rather than a lab. Regions are Bengaluru and Hyderabad only, which is a latency problem from anywhere else, and the per-token rates carry an "introductory launch price applicable only for a month from launch date" caveat that makes the published figures unreliable for budgeting.
Cost
GPU on demand: INR 189/hr for an A100 80GB, INR 213/hr for 1x H100, INR 426/hr for 2x H100 and INR 852/hr for 4x H100, with reserved discounts at monthly, six-month and one-year commitments. AI Pods run from INR 24/hr up to INR 1,700/hr for 8x H100. Block storage INR 0.011/hr or INR 7.88 per GB per month, object storage INR 0.0023/hr or INR 1.66 per GB per month, floating IP INR 204/month, Kubernetes control plane INR 5,110/month, load balancers from INR 8.00/hr plus data transfer. AI Studio inference runs from around INR 3 per million tokens for DeepSeek-R1-Distill-Llama-8B up to INR 74.70 per million for Meta-Llama-3-70B, and INR 0.04 per minute of audio for Whisper. A free tier is offered with no credit card.
Verdict
Buy it for Indian data residency and rupee-denominated GPU capacity, not for the models. After the pivot it is an infrastructure provider reselling open weights, so anyone outside India will get better latency and a deeper catalogue elsewhere.
Visit Krutrim Cloud