By Amit & Animesh, Co-founders, Nuvika Technologies
Back in May we wrote about the 470 ways your cloud wastes money. That number didn’t hold still. Between then and now we kept scanning real customer environments, kept finding patterns our rule set didn’t cover yet, and kept closing the gaps. Fintropy is now at 497 rules - AWS 225, Azure 159, GCP 69, Kubernetes 13, On-Prem/VMware 24, Multi-Cloud 7.
Twenty-seven new rules is a lot to summarize one by one, so here’s the shape of where we expanded, not just the count.
1. AI/ML cost efficiency, now across all three clouds
The last round of AI/ML rules we shipped were mostly about picking the right model and the right pricing mode. This round is about whether the compute underneath is actually being used.
What’s new: low-GPU-utilization detection, overprovisioned GPU SKU flags, and dedicated-endpoint-to-serverless/batch conversion checks - and we built the same three checks for AWS, Azure, and GCP at the same time, not one cloud at a time.
Why it matters: a GPU-backed inference endpoint that’s idle 80% of the day looks identical to a busy one on the invoice. These rules compare actual utilization against the SKU you’re paying for, and flag the cheaper serving pattern (serverless, batch, or a smaller instance) when the workload doesn’t need dedicated always-on capacity.
2. Commitment drift gets deeper coverage
Reserved capacity was already one of our biggest categories - this round adds the parts that only show up after you’ve been running commitments for a while.
What’s new: Cosmos DB and SQL reserved-capacity gaps, expiring-reservation alerts, RI exchange opportunities, late-period Savings Plan purchase detection, and Convertible RI upgrade checks on AWS.
Why it matters: a reservation that was optimal six months ago can be wasting money today - the workload got resized, migrated, or decommissioned, and the commitment kept charging anyway. These rules watch for that drift instead of only checking coverage once at purchase time.
3. BigQuery gets real cost scrutiny
GCP’s rule count grew the most, proportionally - and a good chunk of that is BigQuery, which had been under-covered relative to how expensive it can get.
What’s new: full-table scans in ETL jobs, missing clustering keys, on-demand-vs-flat-rate pricing comparisons, and partition filters that aren’t actually enforced.
Why it matters: BigQuery bills by bytes scanned. A query missing a partition filter or a table missing a clustering key doesn’t fail - it just quietly scans far more data than it needs to, every time it runs. These rules catch the structural reasons a query is expensive, not just the fact that it is.
4. Azure’s waste surface got a lot wider
This is where most of the 27 landed. Azure went from 149 to 159 rules, and it’s concentrated in services that are easy to deploy and easy to forget: Bastion, Firewall, Synapse dedicated pools, Virtual Desktop, Redis, Event Hub, Event Grid, Data Explorer, Stream Analytics, and Data Factory all got new checks - idle detection, tier-mismatch detection, and autoscale-not-configured detection, depending on the service.
Why it matters: these are exactly the “deploy and forget” and “wrong pricing tier” patterns from our original 470 post, just extended into services we hadn’t fully covered yet. A Premium Redis cache with zero Premium features in use, or a Synapse pool billed 24/7 for business-hours queries, wasn’t a new discovery - it was a gap in our own coverage.
5. Picking the right compute architecture
What’s new: Graviton/ARM64 migration checks across Lambda, RDS, SageMaker training, and ECS Fargate on AWS, and an Intel-vs-AMD instance-family comparison on GCP.
Why it matters: ARM-based instances are frequently 20%+ cheaper for the same workload, and switching is often a configuration change, not a re-architecture. These rules flag workloads that are ARM-eligible today but still running on the more expensive architecture by default.
What this means if you’re already scanning with Fintropy
If you connected a cloud account before today, your next scan will run against all 497 rules automatically - no reconnection, no reconfiguration. You may see new findings appear that weren’t there last week. That’s not new waste appearing overnight; it’s coverage catching up to what was already there.
We’ll keep doing this. The honest version of “we scan for cloud waste” is that the rule set is never finished - cloud providers ship new services and new pricing models faster than any fixed checklist can track, so ours keeps moving too.
Fintropy scans your AWS, Azure, GCP, Kubernetes, and VMware environments against 497 deterministic cost rules. Every finding includes specific resources, specific actions, and specific savings amounts. Learn more at nuvikatech.com/Fintropy_Overview.html