OpenAI · AI product · Entry · 45 to 60 minutes

How should ChatGPT handle demand spikes?

OpenAI is an AI company that builds models and products like ChatGPT, a conversational assistant for writing, coding, and answering questions. Answer this product management case and get it graded line by line against a weighted rubric.

Prioritization
Trade off thinking
Metrics
User empathy
Product judgment

The situation

Late 2022. You are a PM on ChatGPT, which has gone from research demo to mainstream product in just weeks. Usage is rising faster than the system can handle, and users are seeing slow replies, failed conversations, and capacity errors. Internally, engineers warn that every message creates meaningful inference cost and that the current stack was not built for this volume. Support tickets and social posts are starting to question whether ChatGPT is reliable. At the same time, many people are discovering real value in the product because it feels fast, clear, and broadly useful for writing, learning, design, and coding. Assume daily demand is growing so quickly that peak traffic could double again within a month, while compute supply stays tight. Your product lead asks you for a short recommendation before the next planning review: what product policy should OpenAI use over the next 8 weeks to keep ChatGPT usable during spikes in demand? You need to balance user trust, access fairness, cost, and long term growth, without relying on a major model breakthrough or a big infrastructure expansion arriving immediately.

What to produce

Write a 1 page product recommendation memo, 600 to 800 words.

Your answer must

State the single most important problem to solve now, and identify the main user groups affected.
Compare at least 3 product options for managing demand, such as usage limits, queueing or access prioritization, message length controls, or delaying non essential feature work.
Make 1 clear recommendation for the next 8 weeks and explain why it best balances reliability, cost, fairness, and trust.
Define 2 success metrics and 2 guardrail metrics, and say what direction you expect each to move if your plan works.
Name the top 3 risks, including at least 1 user trust risk, and propose a mitigation or communication plan for each.

Constraints

Do not assume a major increase in GPU capacity or a new model release within 8 weeks.
Safety and moderation must still operate in real time.
Focus on 1 clear policy decision for demand management, not a full product roadmap.
Your answer must include a product decision, not only an engineering request to optimize performance.

Format

Use sections for problem, options, recommendation, metrics, and risks.

How OpenAI actually played it

Discusses how ChatGPT's viral growth hit capacity limits, why inference is costly and where a PM fits.

Viral growth creates a scale-or-break moment
Inference cost is a product constraint
PMs shape access and quality tradeoffs

Built from Scale or Break: ChatGPT series part 4 by Sreerag Panat.