At renewal, vendors raised prices by 20-37% while their own AI infrastructure costs fell 40-70%. Here's what's really going on, and how to prepare before your next contract arrives.
The renewal email lands, and the new figure is higher. Not slightly higher. High enough that someone on your team screenshots it and drops it in the group chat. The vendor's explanation is "AI enhancements included in your updated plan," with no list of what those enhancements are and no usage data showing you've benefited from any of them. All you get is a bigger invoice for a feature set you may never have opened.
That's how the AI tax typically shows up. It doesn't appear as a separate line item. It comes bundled into a renewal that costs 20 to 37 percent more than last year, with no opt-out, no tiered alternative, and the unspoken warning that your legacy pricing tier is about to be retired.
The price is only part of what's frustrating. The timing matters too. Over the same period, the AI infrastructure behind those "enhancements" (the compute, tokens, and model inference) got 40 to 70 percent cheaper for vendors. Processing data through the OpenAI API fell from $0.03 per 1,000 tokens in 2024 to $0.002 in 2026, and H100 GPU hours became noticeably cheaper on the open market. Vendors' costs went down while your bill went up.That gap isn't a coincidence. It's margin expansion.
Why per-seat pricing is falling apart
For two decades, enterprise software had one pricing unit: the seat. One user meant one license and one monthly fee. The math was simple, renewals were predictable, finance teams could forecast spend from headcount plans, and vendors grew ARR by growing seat counts.
AI agents undermine every assumption behind that logic. Agents don't occupy seats. They consume compute cycles, API calls, tokens, and workflow executions. An agent that resolves 500 support tickets has no named user attached to it, so the seat count stays flat while the value delivered climbs. As one analyst put it, charging per seat for an AI-augmented workflow is like billing per driver on a highway used mostly by autonomous vehicles.
AI-native software companies grasped this from day one. According to Flexprice's 2026 research, 83 percent of AI-native SaaS companies already offer usage-based pricing. The legacy incumbents (Salesforce, Microsoft, Google, Atlassian) built their businesses on the seat model, and they now face an awkward shift: their AI features demand a very different cost structure, yet their installed base expects predictable subscription billing.
Most chose not to redesign their pricing. They raised prices on the existing model and called it an AI upgrade.
Four tactics vendors use to raise your bill
Tropic's procurement team coined the term "AI Tax" and tracked it through real renewal data from its customers. The average AI-driven price increase at renewal runs 20 to 37 percent, compared with a typical annual SaaS uplift of 3 to 9 percent before AI. The difference between those two ranges is what's straining IT budgets right now.
Four recurring patterns explain most of it:
Forced SKU migration. The vendor retires your current pricing tier and moves you onto a new AI-inclusive package, so your legacy plan is gone at renewal. Slack, Google Workspace, and Salesforce have each run a version of this play in recent years.
Unbundling and rebundling. Features you already have get split into add-ons and sold back at a higher price. Gong moved from bundled features to an à la carte model in 2025, which cost roughly 40 percent more for a base product that includes less than before.
Credit obfuscation. AI usage is wrapped in a proprietary credit system that makes it hard to compare costs across vendors or see what you actually consumed. The design discourages you from working out whether the feature is worth the money.
Conditional discounts. A discount on your base product comes only if you buy an AI add-on with it. GitLab gives 5 percent off Premium if you decline GitLab Duo, but bundling Premium with Duo earns 12 percent off Premium plus 30 percent off Duo. The discount makes the AI add-on look like a saving rather than an extra cost.
Each of these works because switching friction is real. Migrating off a CRM or collaboration tool costs more in time and disruption than absorbing a 25 percent price increase, and vendors have built that knowledge into their renewal strategies.
What's taking the place of per-seat pricing
The market isn't settling on higher per-seat prices as a permanent state. It's heading toward three distinct models, and knowing which one you're negotiating for changes what a good deal looks like.
Usage-based pricing charges for consumption: tokens processed, API calls, queries run, documents handled. Snowflake and Twilio built this model for infrastructure, and it's now spreading into application software. The buyer's risk is bill shock. In July 2025, a single developer at a company using Cursor produced a $7,225 invoice from uncapped usage on an annually billed plan. For AI-heavy workflows, usage-based pricing is more rational than seats, but it needs metering, spend alerts, and consumption governance that most teams haven't built yet.
Outcome-based pricing charges per result: a resolved support ticket, a qualified lead, a reviewed contract. Intercom's AI agent Fin charges $0.99 per resolved customer interaction and is now the most cited live example of outcome pricing. Zendesk launched outcome billing in August 2024 at $1.50 per automated resolution, and Salesforce Agentforce charges about $2 per conversation or lead. The appeal is plain: you pay only when work gets done. The difficulty sits in the definitions. What counts as "resolved"? What happens when the AI resolves something incorrectly? Settle those questions in the contract before you sign, rather than discovering them in a dispute six months later.
Hybrid pricing pairs a fixed base subscription with variable consumption or outcome add-ons, and it's where most enterprise renewals in 2025 and 2026 are landing. Neither vendors nor buyers are ready for pure usage or outcome models, so hybrid gives both sides a predictable floor with flexibility above it.
Gartner predicts that at least 40 percent of enterprise SaaS spend will move to usage-, agent-, or outcome-based models by 2030, with the seat-based share of revenue falling from 21 percent to 15 percent. The transition window is open now, and buyers who recognize that have negotiating room that didn't exist two years ago.
The vendor-friendly math has a limited shelf life
One part of the SaaS pricing story gets less attention than it deserves.
According to recent procurement analysis, software vendor pricing is rising 12.2 percent a year, while general inflation across G7 countries runs at 2.5 percent. The 9.7-point gap between them is pure vendor margin expansion, and it's opening at the same moment that vendors' AI infrastructure costs are down 40 to 70 percent.
It holds up in the short term because buyers can't see what their peers pay for the same software. Vertice's analysis of more than $75 billion in spend data found that companies routinely pay close to three times what peers pay for comparable tools. The widest gaps were in sales tools at 2.99x, ERP at 2.92x, and marketing software at 2.87x. NPI's research puts it differently: companies overpay on 89 percent of IT purchases. Gartner puts the average overpayment at 25 percent or more.
That opacity is fading. Tropic, Vendr, Zylo, and similar platforms now collect real contract data at scale, and the information gap that let vendors charge double the market rate keeps shrinking. Bloomberg estimates that subscription-based SaaS pricing will fall from 60 percent to 30 percent of pricing models over the next decade, while outcome-based models rise from 10 percent to 60 percent. Any vendor whose valuation rests entirely on recurring per-seat revenue is sitting on structural risk.
For buyers, the distance between today's pricing and where the market is heading is a negotiating opportunity with an expiry date. The window is 2025 to 2026, and it closes as the new models standardize.
What to do before your next renewal
Tropic's renewal data shows that negotiation cuts initial AI uplift demands by about 55 percent on average. Companies that pushed back ended at roughly 12 percent, down from the 20 to 37 percent vendors first proposed. On a $200,000 annual contract, the difference between accepting a 30 percent increase and negotiating it down to 12 percent is $36,000 a year.
Here's the playbook that works in 2026:
Start six months before renewal, not six weeks. Vendors know they hold the leverage when renewal is close. Six months out, you have time to evaluate alternatives and make a credible case that you'd use them.
Ask what your current tier would cost without AI features. Don't assume a new SKU structure is mandatory. Asking "what would renewal look like at our current feature set?" often produces a number well below the AI-inclusive proposal.
Audit real AI feature usage before accepting any AI-based increase. Most businesses use one or two AI features regularly while paying for all of them across several tools. If you're paying an AI premium for features with single-digit adoption, that data is your strongest bargaining chip.
Benchmark against peer pricing. Opacity is on its way out. Procurement platforms now show what comparable companies pay for the same contracts, so if you're paying 2x the market rate, you can say so with data.
Push for hybrid contracts with consumption ceilings. If a vendor is moving you to usage-based pricing, negotiate a hard spend cap. Uncapped consumption becomes a budget exposure that compounds fast once AI agents are running at scale.
Use the structural shift as a framing tool. The most effective script in 2026 goes: "Analyst consensus says seat-based pricing is declining. We'd like a contract that reflects that trajectory, with transition terms written in now." It casts you as a strategic partner, not a difficult customer, and it's factually correct.
Vendors steer the pricing conversation when buyers don't understand the models shifting underneath them. Buyers who do understand the transition hold the stronger hand. That's new, and it won't last.
The bigger shift to keep an eye on
This transformation isn't only a procurement story. It points to something deeper about how software economics work.
When vendors price on outcomes instead of access, they carry delivery risk for the first time. Intercom's Fin charges only when a ticket is resolved, and Zendesk's pricing assumes the AI really does resolve the ticket. If the AI underperforms, the vendor earns less. That incentive structure differs sharply from selling seats whether or not anyone uses them, and it's healthier for buyers who care whether the software actually works.
Flexprice's 2026 research found that 78 percent of IT leaders report unexpected charges from consumption-based or AI pricing, and 90 percent of CIOs cite cost forecasting as their top challenge in AI deployment. The new models bring new governance needs: metering infrastructure, spend dashboards, usage alerts, and outcome measurement agreements. None is hard to build, but together they demand more operational sophistication from buying teams than the per-seat era ever did.
The companies that handle this well aren't the ones that reject the new pricing models. They're the ones that build the instrumentation to know what they're consuming before a vendor's invoice tells them.
The AI tax is real. So is the infrastructure cost deflation that vendors aren't passing along. And the negotiating window opened by the move from seats to usage and outcome models is real too, open now in a way it won't be once the models standardize.
Your next renewal email is on its way. Will you understand the pricing model well enough to negotiate from the right position, or will you accept the number because the justification sounded plausible and switching looked hard?
Most companies take the second path, which is why 89 percent of them are overpaying.
Enjoyed this read? Be sure to follow me for more stories that make people feel things. You can also connect with me on LinkedIn or Telegram @kawsarlog for more updates on my work.
I do freelance scraping projects, so if you need some data collected feel free to contact me on LinkedIn or email at hi@kawsarlog.com
Top comments (0)