Nebius Raises Compute Prices Oct 1 as OpenAI Signals a $500 Pro Max Tier

Two pricing signals arrived less than two weeks apart, and together they squeeze the middle of the stack. Nebius announced on September 17 that its AI compute prices rise on October 1, citing a shortage of GPU capacity. OpenAI hinted at a $500-per-month ChatGPT Pro Max tier in its code and interface text, which points to an ultra-premium seat above the current $200 Pro tier. Prices are rising at both the bottom and the top of the stack. The layer in the middle, which is most of B2B SaaS, has this quarter to decide whether to absorb the increase or pass it on.

Waiting until FY27 planning is the wrong response. Nebius's October 1 increase is a supply-side cost, and it reaches every vendor whose product pays an inference cost per query. That includes most AI-powered SaaS built on rented compute. The cost of a query goes up in the invoice cycle that clears next month, so the pricing response is a Q4 decision.

Passthrough is a customer conversation, absorption is a margin one

The two responses carry different costs. Passing the increase on means telling your customer the new number, defending it, and showing them how it compares with not raising prices at all. Absorbing the increase means giving up margin on the exact line item the board asked about last quarter. Most vendors will do some of each, and the ratio between them is the real decision.

Vendors that leaned toward "we absorb" in 2025 did it because customer churn from AI price shocks looked like the bigger risk, and compute costs were falling at the time. Both conditions have reversed. Compute is getting more expensive, and buyers have spent a year getting used to AI-enabled workflows costing more than seat licenses. Buyers now expect metered pricing on top of the base seat. What’s still open is which vendors are willing to charge the full difference.

OpenAI's $500 signal sets the anchor

The Pro Max price matters less as a consumer price point than as an anchor for enterprise conversation about how much a single seat of a frontier model is worth once the model can plan, execute, and hand off work across tools. If OpenAI ships Pro Max at $500, every SaaS vendor above it in the stack will be quoting to customers who see $500 as a normal price for serious AI work. That pulls mid-tier prices up and gives the vendor with the most AI built into their products room to price closer to $500.

There's a fair counterargument and it deserves airtime. Nebius is one provider sitting next to the hyperscalers. Its price rise will reach some buyer contracts, and not others, and enterprise volume discounts will absorb part of the increase before it reaches the invoice. OpenAI's Pro Max may not ship at $500, or it may ship at $500 with usage caps that limit exposure. Both of those readings mean the CFO conversation is smaller than the headline suggests, and the correct move right now is to model both scenarios and hold off on price change until the market is clearer.

Holding off is where that reading goes wrong. A CFO who doesn't model the pass-through this quarter will lose the pricing conversation to a peer who did, because the negotiating window is the renewal, and earnings come too late. Two vendor cycles from now, vendors that raised price with a clear AI cost story will have higher ARPU, and the ones that waited will be defending a margin they’ve already lost. The signal is early, and the decision can’t wait.

OpenAI settled the open question on September 29, when it launched the $500-a-month tier at DevDay under the name Pro 500, not Pro Max, and bundled the first Dot, its always-on agent, into Pro plans at no extra cost. Usage caps did arrive, as the counterargument allowed, but they landed on the $200 plan, where OpenAI's Pro tiers page says new subscribers get a smaller allowance than existing ones at the same price. That makes the middle seat more expensive per unit of work without a new sticker price. Our follow-up on how OpenAI priced Dots covers what the $500 seat buys a team that wants an agent.

What this means for SaaS CFOs and pricing leaders

First, model the pass-through at three levels: pass on 50%, pass on 100%, and absorb 100%. The model won’t make the decision for you but it gives you something concrete to bring to a board conversation that’s going to happen anyway.

Second, add an AI-cost line on the invoices of your 10 biggest accounts, framed as transparency rather than a price rise. Customer's procurement teams are already asking about AI cost pass-through in vendor reviews right now, and the vendor that shows the line first shapes the negotiation.

Third, decide before Q1 whether any product tier should switch to consumption pricing. Vendors that adopted consumption pricing in 2026 are seeing higher expansion revenue than those that stayed on seats, because consumption grows with usage while seat prices renew flat. Nebius just told the market that the costs are going up, and if the revenue side doesn't follow, the margin story will be a bad one.

This is a P&L question more than a market story, and it lands in the CFO's inbox on October 1 whether the vendor is ready or not.

Sources

• Nebius hikes AI cloud prices again, announced September 17, effective October 1 (Reuters)
• OpenAI signals a $500-per-month ChatGPT Pro Max plan

A CFO who doesn't model the pass-through this quarter will lose the pricing conversation to a peer who did, because the negotiating window is the renewal, and earnings come too late.

Model the pass-through at three levels, passing on 50%, passing on 100%, and absorbing 100%, before the renewals where the pricing conversation happens.

By Business of GTM