Best of FinOps — 2025

1. 1
   Article

   Hacker News·28w

   Send this article to your friend who still thinks the cloud is a good idea

   A developer shares their experience moving projects from AWS to bare-metal servers with Hetzner, achieving 10x cost savings and 2x performance improvement. The piece argues that cloud services like AWS charge excessive markups (10x-100x) compared to renting or buying servers directly, and that most small-to-medium businesses don't need expensive managed cloud services. It challenges common fears about server management, suggesting that with modern tools like AI assistants, managing Linux servers is accessible and cost-effective for most developers.

   93

   18
2. 2
   Article

   n8n·50w

   Run N8N for 1$/month instead of 28$ (+ more features)

   Self-hosting n8n on a VPS costs only $1/month compared to $28/month for the SaaS plan, while providing unlimited workflows instead of the 5-workflow limit on the paid cloud plan. This approach offers significant cost savings and removes usage restrictions for automation workflows.

   80

   8
3. 3
   Article

   portkey·1y

   Task-Based LLM Routing: Optimizing LLM Performance for the Right Job

   Task-based LLM routing directs incoming AI requests to the most suitable large language model based on the task. This approach improves performance, reduces costs, and enhances scalability by matching tasks with models optimized for those specific needs. For instance, simpler tasks can be routed to lightweight models like GPT-3.5 to minimize costs, while complex tasks are handled by more powerful models like GPT-4. This method also enhances reliability and latency, and is useful in diverse applications like customer support, content creation, code-related tasks, and multilingual processing.

   69
4. 4
   Article

   Groww Engineering·31w

   A Framework for Cloud Cost Optimization: How We Saved 40% of our Cloud cost

   Groww Engineering reduced their cloud costs by 40% over three months through a systematic framework combining visibility, ownership, and architectural changes. They built an internal FinOps dashboard for granular cost tracking, standardized resource labeling across teams, shifted from fixed to elastic infrastructure, deprecated legacy services, migrated analytics to an in-house query engine, and established continuous optimization practices with team-level budgets and regular audits.

   54
5. 5
   Article

   Medium·34w

   Don’t buy GPUs for AI

   GPUs are becoming unnecessary for most AI applications as smaller language models like Mistral 7B and Phi-3 Mini deliver practical results on CPUs. Modern processors, edge devices with NPUs, and cloud rental options provide cost-effective alternatives to expensive GPU ownership. Specialized hardware like TPUs and software optimizations through quantization are making GPUs obsolete for all but the largest model training operations.

   53

   9
6. 6
   Article

   AWS Fundamentals·39w

   AWS Cost Optimization Key Principles for Beginners

   A comprehensive guide covering AWS cost optimization fundamentals for beginners. Explains how to understand AWS bills through cost allocation tags, Cost Explorer, and budgets. Provides practical strategies for optimizing storage (S3 lifecycle rules, EBS right-sizing), compute (instance sizing, Spot instances, Reserved Instances vs Savings Plans), and network costs. Covers common pitfalls like orphaned resources and mixing dev/production environments. Includes actionable tips like cleaning up unused resources, implementing proper tagging, and building cost-aware team culture through regular reviews and accountability.

   48
7. 7
   Article

   Hacker News·30w

   How Idealist.org Replaced a $3,000/mo Heroku Bill with a $55/mo Server

   Idealist.org reduced their staging environment costs from $3,000/month on Heroku to $55/month by migrating to a single Hetzner server running 6 environments. Using Disco for deployment automation, they maintained the git-push workflow and developer experience while sharing a single Postgres instance across environments. The migration required handling DNS/CDN configuration and accepting responsibility for server maintenance, but transformed staging environments from a scarce, expensive resource into an abundant commodity that developers could spin up freely.

   30

   1
8. 8
   Article

   Halodoc·22w

   Kubernetes OptimizationInPlace Pod Resizing,ZoneAware Routin

   Halodoc reduced Kubernetes infrastructure costs by implementing two optimization strategies: in-place pod resizing to dynamically adjust resources during low-traffic periods without restarts (achieving ~15% CPU and ~10% memory reduction), and zone-aware routing to minimize cross-AZ traffic (reducing data transfer costs by ~25% and latency by ~5%). The approach uses a custom scheduler for periodic resource patching and Kubernetes' trafficDistribution: PreferClose feature combined with topology spread constraints to keep traffic within availability zones.

   24
9. 9
   Article

   DevOps·51w

   Cloud Skills 2025: It’s not just about tools. It’s about knowing what really matters.

   Cloud skills are evolving, emphasizing the importance of understanding why tools matter rather than just knowing their names. Key areas of focus include mastering multi-cloud fluency, Infrastructure as Code (IaC), DevSecOps, observability, GitOps, FinOps, and scripting. Engineers should aim to build foundational skills that endure beyond fleeting trends.

   16
10. 10
    Video

    YouTube·1y

    How Duolingo Saved Millions in AWS Cloud Costs

    Duolingo, a popular language learning app, successfully reduced their AWS cloud costs by 20% through various methods. They utilized Cloud Zero for enhanced billing visibility, implemented lifecycle rules for S3 storage, established TTL rules for DynamoDB, optimized logging practices, rightsized resources, and addressed inefficient service interactions. Additionally, they leveraged AWS's in-built cost optimization features and better-managed reserved instances, leading to substantial savings.

    13
11. 11
    Article

    The Last Week in AWS·24w

    AWS Finally Lets You Find Your Idle NAT Gateways

    AWS Compute Optimizer now identifies idle NAT Gateways, helping users eliminate unnecessary costs. Each idle gateway costs approximately $35/month plus data processing fees. A NAT Gateway is considered idle when it has no active connections, no incoming packets from VPC clients or destinations for 32 days, and isn't associated with a route table. This feature addresses the low-end cost problem of forgotten resources, though high-volume data processing charges remain a separate concern.

    12
12. 12
    Article

    Community Picks·1y

    27 tips: How to reduce cloud costs

    This post offers strategies for reducing AWS cloud costs, focusing on EC2 instances, databases, and data transfer. It includes general tips such as using spot instances, automated scaling, and optimizing database and server configurations. AWS-specific recommendations like using Reserved Instances or newer EC2 instance types are also provided, along with Kubernetes cost optimization techniques such as consolidating pods and autoscaling cluster nodes.

    12

    4
13. 13
    Article

    Medium·1y

    Prompt chaining is dead. Long live prompt stuffing!

    Prompt chaining was once essential for working within limited context windows in LLMs, allowing the creation of complex and nested JSON objects. However, with the advent of modern LLMs like Gemini 2.0, which offer large context windows, prompt chaining has become obsolete. Instead, prompt stuffing—putting all necessary context into a single prompt—saves time and resources, significantly reducing API costs and complexity. This shift reflects the rapid evolution in the AI field, where practices can quickly become outdated.

    12
14. 14
    Article

    Jobs·28w

    The Day Our Database Bill Nearly Sank the Company

    A SaaS startup reduced their database costs by 60% through systematic optimization. The team identified missing indexes causing full-table scans on 50M+ row tables, removed unused indexes, archived old logs to S3, and refactored queries. Results included 5x faster queries, 40% lower CPU load, and 50% storage cost reduction. The key takeaway emphasizes database optimization as an ongoing practice requiring regular monitoring of queries, maintaining lean indexes, and proper data lifecycle management.

    11

    2
15. 15
    Article

    Cast AI·49w

    Kubernetes Pod Scheduling: Balancing Cost and Resilience

    Kubernetes pod scheduling requires balancing cost efficiency with resilience through strategic use of affinity rules, anti-affinity constraints, and topology spread configurations. The guide covers resource optimization techniques that can improve CPU utilization by 35-47% and memory utilization by 28-39%. Key strategies include using cascading constraints from strict at broad topology levels to flexible at narrow levels, implementing different patterns for various workload types (global services, stateful applications, microservices, cost-optimized services), and avoiding common pitfalls like overly strict anti-affinity rules and conflicting configurations. Proper implementation ensures high-performance applications remain available during infrastructure issues while optimizing resource usage.

    10