Hyperscaler H100
~$0/mo
AWS, Azure, GCP on-demand pricing
Case Study — Scaleups & Startups
The Cloud Credits Trap
Hyperscalers lure scaleups with generous GPU credits. By the time they run out, your infrastructure is locked in and you're paying 6× more than you should. An H100 costs ~$6,000/month on AWS, Azure, or GCP. On alternative providers it can be under $1,000. stack8s makes switching trivial.
Alternative Provider H100
<$0/mo
Via stack8s multi-cloud fabric
Provider Switch Time
Minutes
Zero code changes on stack8s
One Control Plane. Zero Complexity.
stack8s handles billing, provisioning, and multi-cloud GPU pooling end to end from a single control plane. Your data stays exactly where it is — we bring compute to your data, not the other way around. No need to create or manage your own cloud accounts.
Unified Billing
One invoice for all GPU providers, tools, and services. No reconciling bills from five different clouds.
End-to-End Provisioning
From bare metal to running workload in minutes. stack8s provisions, configures, and manages the full stack.
Compute Comes to Your Data
Your data stays where it is. We orchestrate GPU compute from 15+ providers to run where your data lives.
No BYOC Accounts
No need to create AWS, Azure, or GCP accounts. stack8s handles all cloud provider relationships on your behalf.
Think of it this way: Instead of managing cloud accounts, wrestling with IAM policies, and reconciling invoices from multiple providers — you get a single platform where you deploy workloads and stack8s takes care of everything underneath. One login. One bill. Every GPU provider.
How the Cloud Credits Trap Works
It starts with a generous email. “Here's $200K in cloud credits to get you started.” Click through the timeline to see how the trap unfolds.
Month 1
The Generous Offer
You receive $100K–$350K in free cloud credits. Everything feels free. You spin up H100s without a second thought.
H100 Pricing Across Providers
Hover over any bar to see the markup compared to the lowest-cost provider. Hyperscalers charge 5–7× more than GPU-focused cloud providers.
AWSHyperscaler$5,932/mo
AzureHyperscaler$5,877/mo
GCPHyperscaler$6,102/mo
CoreWeave$2,210/mo
Lambda Labs$1,990/mo
FluidStack$940/mo
Jarvis Labs$860/mo
Vast.ai$780/mo
Key insight: After cloud credits expire, scaleups on AWS/Azure/GCP can pay 5–7× more per H100 than they would on GPU-native providers. With 4 GPUs, that's over $240K/year in excess spend.
Interactive Cost Calculator
Drag the sliders to see how much you could save by switching away from hyperscaler GPU pricing.
4
116
12 months
3 mo36 mo
Hyperscaler Cost
$286,560
$23,880/month · ~$5,970/GPU
With stack8s
$40,800
$3,400/month · ~$850/GPU
Your Savings
$245,760
86% saved over 12 months
Credit Burndown Simulator
Watch $200K in cloud credits vanish in real-time. 4 H100s at hyperscaler rates.
Credits Remaining
$200,000
Month
0
100% remaining
$0
Spent
$24,000
Per Month
9 mo
Runway Left
Try Switching Providers
Click any provider below to simulate a workload migration. On stack8s, this takes minutes — not months.
Current Provider
AWS
H100: $5,932/mo
Switch To
It's Not Just GPUs — Your Entire Stack Costs More
Cloud lock-in inflates the price of every tool in your AI stack, not just compute. Click any tool below to see the price difference between managed cloud services and running on stack8s.
Supabase
Backend-as-a-service for your AI product
Managed Postgres with auth, real-time, and storage. On cloud you pay premium for compute add-ons. On stack8s, it runs natively on your cluster.
Supabase Cloud Pro + compute add-ons$2,000/mo
On stack8s$750/mo
$1,250/mo saved
63% less · $15,000/year
Full Stack Savings
Running all 6 tools on stack8s vs managed cloud:
$15,100
Cloud / mo
$4,280
stack8s / mo
$129,840
Saved / year
The hidden cost of “managed”: Cloud providers bundle high margins into every managed service. A Supabase Pro plan with compute add-ons can cost $2,000/month. On stack8s, the same Supabase instance runs on your cluster for ~$750/month — same features, same performance, 63% less cost.
Why Leaving a Hyperscaler Is So Hard
It's not just the credits. The entire hyperscaler model is designed to create switching costs that keep you paying premium prices forever.
Proprietary APIs & Services
Your storage, networking, IAM, and observability are all wired into proprietary services. Moving means re-architecting, not just re-deploying.
Months of Migration Work
Without a portability layer, migrating GPU workloads to another cloud takes 2–6 months of engineering effort — time scaleups don't have.
Data Egress Fees
Hyperscalers charge significant fees to move your data out. Large training datasets can cost thousands to egress alone.
Price Increases After Lock-In
Once you're dependent, hyperscalers have no incentive to compete on price. Reserved instance contracts lock in high rates for years.
stack8s: The Escape Hatch
stack8s treats every GPU provider as interchangeable capacity. Your workloads run on Kubernetes-native infrastructure, so switching between any of our 15+ providers is as simple as changing a placement policy. No re-architecture. No pipeline rewrites.
Without stack8s
- Locked into one cloud after credits expire
- 2–6 months to migrate to a new provider
- Paying $6K+/mo per H100 indefinitely
- No price visibility across other providers
- Vendor controls your roadmap
With stack8s
- Portable across 15+ GPU providers from day one
- Switch providers in minutes, not months
- H100s from <$1K/mo via GPU-native clouds
- Real-time pricing intelligence across all providers
- You control the infrastructure, always
Built for Scaleups
Scaleups need to move fast, stay lean, and avoid burning runway on overpriced infrastructure. stack8s is designed for exactly that.
Deploy in Minutes
Spin up GPU-accelerated environments instantly. No infra team needed. Kubernetes-native from the start.
Switch Anytime
Move between providers as pricing changes. Stay on the cheapest available GPU without re-engineering.
Extend Your Runway
Pay <$1K/mo for an H100 instead of $6K. Save $60K+ per GPU per year and redirect it into product.
Production-Ready
Not just for experiments. Run production inference, training pipelines, and data workloads on enterprise-grade infrastructure.
No Lock-In, Ever
Your workloads are portable from day one. Use credits if you have them, but you'll never be trapped.
Scale With Confidence
As your GPU needs grow, stack8s scales across providers. Multi-cloud flexibility means costs stay competitive at every stage.
Scenario: Series A AI Startup
A Series A startup raises $8M. They accept $200K in AWS cloud credits to train their models. Here's how their first year compares with and without stack8s.
| Metric | Hyperscaler Path | stack8s Path |
|---|---|---|
| GPU Spend (Year 1) | $288,000 | $40,800 |
| Credits Used | $200,000 | $0 (not needed) |
| Actual Cash Burn | $88,000 | $40,800 |
| Provider Lock-In | Complete | None |
| Time to Switch Provider | 3–6 months | 5 minutes |
| Year 2 Projected Cost | $288,000 | $40,800 |
| 2-Year Total | $376,000 cash | $81,600 |
“We took the credits and it felt like a no-brainer. Twelve months later, we were spending $24K/month on GPUs with no easy way out. stack8s would have saved us nearly $300K in our first two years.”
Don't Let Credits Become Handcuffs
Your cloud credits will run out. When they do, you'll either be paying 6× market rate for GPUs, or you'll be on stack8s with the freedom to use any provider at the best price. The time to choose is before you're locked in.