Sam Altman on X: "we are pausing new ChatGPT Plus sign-ups for a bit"
OpenAI’s decision to pause new ChatGPT Plus sign-ups is widely seen as a sign of hitting hard capacity limits, with commenters pointing to global GPU shortages, high inference costs, and recent performance slowdowns on GPT‑4. Many compare the scale and economics of large language models to Google search, arguing that current hardware, energy costs, and cloud capacity make mass access difficult and potentially unprofitable. Others criticize the requirement that all users of custom GPT bots must be Plus subscribers, noting that the signup freeze now prevents many promoted GPTs from being used at all.
Capacity, Outages, and the Plus Sign‑up Pause
- Many commenters think the pause reflects hitting hard capacity limits, especially after Dev Day publicity and new features (GPT‑4 Turbo, GPT builder).
- Others float alternative explanations: intentional scarcity/hype, fraud/chargeback concerns from mass sign‑ups with virtual or stolen cards, or simply unclear internal issues.
- Several note recent outages and much slower GPT‑4 responses, suggesting infrastructure strain.
Compute Scale and GPU Constraints
- LLM queries are seen as far more compute‑intensive than web search; multiple comments argue today’s hardware cannot support “Google‑scale” open access.
- Back‑of‑the‑envelope estimates suggest OpenAI may be consuming a significant fraction of all high‑end GPUs (e.g., H100s).
- Azure GPU capacity is reportedly so tight that even large enterprise customers can’t get firm commitments when they need the same hardware OpenAI uses.
- Some note GPUs wear with high thermal load, but data‑center hardware is generally designed for 24/7 use.
Economics, Profitability, and Nvidia
- There is debate whether OpenAI’s business is highly profitable or bleeding cash due to massive compute and energy costs.
- Some cite reports or speculation that similar products (e.g., Copilot) may be loss‑making; others argue users would pay more and premium tiers will emerge.
- Several see Nvidia as the main financial winner of the AI boom (“selling shovels in a gold rush”), while others worry about valuation bubbles and broader societal implications.
Product Quality and Usage Experience
- Users report GPT‑4 being slower, “broken,” or too eager to browse the web even for non‑time‑sensitive questions.
- There are claims that free or low‑tier API users are silently downgraded to weaker models under heavy load, making reproducibility difficult.
- Some prefer a simple, non‑browsing GPT‑4 option and note GPT‑3.5 is much faster.
Access Policy and Custom GPTs
- Multiple comments criticize the policy that both creators and users of custom GPTs must have ChatGPT Plus.
- With new Plus sign‑ups paused, creators who promoted their GPTs now find that many potential users cannot access them, despite there being no direct revenue share.