Fly.io · Strategy · Mid · 45 to 60 minutes

How should Fly.io fix GPU demand mix?

Fly.io is a cloud platform that lets developers run applications and related infrastructure close to users around the world. Answer this product management case and get it graded line by line against a weighted rubric.

pricing strategy
analytical reasoning
customer segmentation
product judgment
communication

The situation

The head of compute product at Fly.io asks you for a recommendation on the company’s GPU portfolio for the next quarter. In recent months, demand has concentrated on an older A10 GPU offering, while a newer and more capable L40S offering is seeing weaker uptake than the team expected. This is creating a business problem because GPU capacity is expensive to hold, and underused premium inventory ties up capital and planning attention. The hard part is that there is no obvious single cause. Customers may be price sensitive, their real workloads may not need the extra capability, or Fly.io may have positioned the tiers around internal assumptions instead of actual usage. At the same time, changing pricing or packaging too quickly could hurt margins, blur the value of the lineup, or simply shift existing demand without growing overall usage. Leadership wants a practical strategy that can be implemented quickly, grounded in customer behavior rather than theory, and that helps the company decide how to price, position, and communicate its GPU offerings going forward.

What to produce

Write a 1 to 2 page product strategy memo.

Your answer must

Explain the core problem in terms of customer demand, workload fit, and inventory utilization, and list the top hypotheses you would test first.
Recommend a next quarter strategy for the A10 and L40S offerings that covers pricing, positioning, and which customer jobs each tier should serve.
Describe the customer segments or workload categories that matter most here, including where inference fits, and how their needs differ.
Define a validation plan with metrics, leading indicators, and 2 to 3 concrete experiments or decision checks for the next 30 to 60 days.
Address key risks, including cannibalization, margin pressure, and customer trust if Fly.io changes its stance or admits earlier assumptions were wrong.

Constraints

Use only the information in the prompt plus reasonable PM assumptions, do not assume access to new products, partnerships, or feature launches.
Assume the current GPU inventory mix cannot change quickly, so your plan should work mainly by improving demand allocation within existing capacity.
Your recommendation must be practical for the next quarter, not a long term research program.
Do not assume that the more capable GPU is automatically the better customer choice.

Format

Use sections: diagnosis, options considered, recommendation, risks, and validation plan.

How Fly.io actually played it

Fly.io lowers L40S GPU prices to 1.25 dollars an hour after finding that customers prefer the older, less capable A10 and that its GPU strategy guesses were wrong. The post is open about how a hardware provider formulates GPU strategy and what customers actually used. A candid pricing and inventory-strategy case.

Check what customers actually use, which can differ from what you expected.
Price the underused SKU to move demand where capacity exists.
Be open about wrong guesses to build credibility.
Match capability tiers to real workloads such as inference.

Built from We're Cutting L40S Prices In Half by Kurt Mackey, Fly.io.