Archive · 27 Sep 2026
Startups Rethink Cloud Use as AI Bills and Costs Mount
Startups are paying much more for cloud computers than they did two years ago.
AI programs and powerful GPU chips are major reasons for the increase.
Some companies now spend tens of lakhs of rupees every month on cloud services.
This makes cloud bills an important business cost, not just a technology expense.
NoBroker lowered its AI costs by using some of its own computers and some cloud services.
Zippee saved money by removing unused computer capacity and changing how it stored data.
Other startups are hiring specialist companies to find waste in their cloud bills.
These companies are trying to make every computing resource work harder.
The goal is to spend less on infrastructure while continuing to build and run their products.
Growth-stage startups now typically spend Rs 18-45 lakh monthly on cloud, up from Rs 8-20 lakh two years ago.
Late-stage and enterprise companies’ monthly cloud spending has increased from Rs 50-80 lakh to Rs 80-150 lakh.
AI experimentation, complex workloads and GPU computing are driving higher infrastructure costs.
NoBroker cut AI usage costs by 80% by combining frontier and open-weight models across cloud and local infrastructure.
Zippee reduced cloud spending per shipment by 30-35% through right-sizing, spot capacity and stricter data-retention controls.
- Who
- Startups, including NoBroker and Zippee, along with cloud-optimization providers such as CloudKeeper and Utho Cloud.
- What
- Startups are reducing or restructuring cloud usage as AI and GPU-related costs drive sharply higher infrastructure bills.
- When
- Cloud spending comparisons cover the past two years; Zippee’s savings were measured over the past year.
- Why
- More complex workloads, AI experimentation and GPU computing have raised costs, while startups face pressure to control spending.
Self-Managed and Optimized Infrastructure
Provider-Hosted Cloud Services
How to control AI costs
Self-Managed and Optimized Infrastructure
Startups can use their own GPUs for heavy and predictable workloads, deploy open-weight models locally and eliminate idle or overprovisioned resources.
Provider-Hosted Cloud Services
Provider-hosted environments remain part of the mix and can be used when workloads require flexibility or when operating dedicated infrastructure is less efficient.
Who should manage cloud efficiency
Self-Managed and Optimized Infrastructure
Internal teams can reduce costs through right-sizing, spot capacity, workload ownership and tighter storage and data-retention policies.
Provider-Hosted Cloud Services
Specialist third-party providers can manage optimization; companies including MobiKwik, Zepto and Lenskart have used such vendors.
- Growth-stage spending
- Typically Rs 18-45 lakh per month, compared with Rs 8-20 lakh two years earlier.
- Late-stage spending
- Typically Rs 80-150 lakh per month, compared with Rs 50-80 lakh two years earlier.
- NoBroker savings
- AI usage costs fell 80% in less than 24 months.
- Zippee savings
- Cloud spending per shipment fell 30-35% over the past year.
- MobiKwik savings
- Overall Amazon Web Services costs fell 12% over the past year with third-party assistance.
- CloudKeeper estimate
- Customers achieved an average cloud-cost reduction of around 20%.
- Utho Cloud estimate
- Clients typically reduced costs by around 60%, with some reported savings of 75-80%.
- Operating expenditure share
- Cloud spending can rise from 8-10% at an early stage to 18%, 25% or 35% as companies scale, according to Utho Cloud.