OpenAI Shared Limits Turn Seats Into Metered Capacity
Image design by Aragon, rendered by Canva AI.
By Adam Pease
OpenAI Shared Limits Turn Seats Into Metered Capacity
Starting Dec. 31, OpenAI will put ChatGPT and Codex usage on Plus, Pro and Business plans into one shared allowance. The news arrived within days of reports that the company had cut the target for its private funding round. The seat price stays flat, but what a seat can do now has a ceiling. This blog overviews the OpenAI shared usage limit announcement and offers our analysis.
Why Did OpenAI Announce a Shared ChatGPT and Codex Limit?
OpenAI says that from Dec. 31, conversations in ChatGPT and coding work in Codex will draw down a single remaining balance. Only chats in Instant mode, the company’s fast-response model, are exempt. The change is part of a broader move to bring the agent features of Work, OpenAI’s agent for multi-step professional tasks, into ChatGPT itself. Once that happens, a developer’s Codex session and a colleague’s research agent will be drawing on the same pool of capacity.
The timing tells the rest of the story. CNBC and the Financial Times reported that OpenAI cut its round target from $30 billion to $20 billion. The Wall Street Journal had reported that only about $11 billion was committed and that expected valuation had fallen from $1.4 trillion to around $1.25 trillion. Meanwhile, Axios, citing Ramp data, reported that token prices fell 35% from the end of June to about $0.59 per million tokens as of Oct. 20.
Costs are moving the other way. TrendForce expects the average selling price of high-bandwidth memory (HBM) to rise 121% next year. Samsung says its HBM supply for next year is sold out, and SK Hynix says its output is effectively committed through 2029. CEO Sam Altman met with Samsung and SK Hynix executives in South Korea over the weekend. OpenAI earns less per token while its costs rise, and usage limits are how it protects margin.
Analysis
At Aragon Research, we view the shared limit as metering, even though the plan still carries a flat seat price. The seat is no longer a fixed bundle of capability. It is a claim on a capped amount of compute that chat, coding and agent work now compete for. We have expected metering to spread across the AI stack, and the seat-based tier is where it has now arrived.
The bigger risk for buyers is that allowances keep moving. In September, OpenAI announced that Pro 200’s Work and Codex allowance would fall from 20x Plus to 10x. On Oct. 29, it cancelled that cut and doubled allowances on Pro 100 and Pro 500. Allowances have changed several times in a few weeks, and OpenAI does not publish them as fixed quotas. Fast mode uses included allowance at 2.5x the standard rate and Ultrafast at 8x, so the work a seat delivers depends on settings users pick themselves.
Other frontier providers face the same pressure. Anthropic’s revised S-1 shows $17 billion in third-quarter revenue and $895 billion in purchase commitments, and it faces the same capital and compute pressure. As long as capital and memory costs stay high, every frontier provider will keep adjusting what a seat includes.
The shared pool also collides with how agents are being deployed. Aragon predicts that by YE 2026, 50% of knowledge workers will be supported by at least one AI Assistant and three AI Agents. Those agents will compete for the same pools, and unmanaged agent use will burn through allowances fast. That is the kind of governance gap behind our forecast that by 2030, 65% of organizations will fail to achieve target business outcomes from Agentic Systems investments.
What Should Enterprises Do About This News?
Enterprises should stop buying AI per user and start treating it as a capacity contract. Four steps matter now:
- Measure before Dec. 31. Use Settings ? Usage and workspace admin data to see how much each team uses ChatGPT, Work and Codex. Find out which teams will drain the shared pool first; usually it is engineering teams running Codex alongside heavy agent users.
- Negotiate renewals around capacity. For Business and Enterprise renewals, ask for written allowance floors, advance notice of any allowance change, and fixed rates for extra usage. Remember that Business seat billing and API billing are separate.
- Set a usage policy and assign seats by need. OpenAI lists Business Premium at $100 per user per month on annual billing and Standard at $20. Give Premium only to people who run long agent tasks, keep reviewers on Standard, and limit Fast and Ultrafast modes to work where response time matters.
- Keep a second supplier ready. Multi-vendor strategies and data portability protect buyers. Test routing for coding and agent work now, so a mid-contract allowance change or a vendor’s funding trouble becomes a routing decision instead of a stoppage.
Bottom Line
OpenAI’s shared ChatGPT and Codex limit turns flat-rate seats into metered capacity, and capital and memory costs mean allowances will keep shifting. Enterprises that measure usage, contract for allowance floors and keep a tested second supplier will control their AI spend. Those that keep buying per user will find their capacity set by the vendor’s balance sheet.