👋 Hi, it’s Kyle Poyar and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups. Subscribe to join 83,000+ growth pros who get this newsletter delivered straight to their inbox.

Tech founders used to tell me their pricing was inspired by Salesforce or Slack. Now they tell me Intercom, HubSpot, OpenAI, and Clay (and, yes, some still say Salesforce).

Aside from Intercom, all have some form of credit-based pricing.

These iconic companies are far from alone. Credit-based pricing surged by 126% in 2025. Figma just joined the club, and their new AI monetization strategy sent the stock price surging (after previously falling by 80%). PostHog added AI credit pricing, too.

When these influencers change their pricing, I pay attention. And Clay did just that, announcing a big pricing overhaul just this morning.

Clay even open sourced their internal memo explaining why and how they made the decision (it’s worth a read). The team at Clay shared an advanced preview with me, although this post is not sponsored by Clay and nor was the post shared with the company ahead of publication.

I’ll unpack the new pricing changes from Figma, PostHog, and Clay. And, more importantly, I’ll explain what these signal about where AI pricing might be headed next.

Figma introduced an AI credit model in December 2025. But they didn’t actually enforce it. Aside from the delayed enforcement, Figma’s credit pricing looks pretty similar to other credit models from the likes of HubSpot, Salesforce, monday, and others.

The mechanics at Figma:

Figma’s decision to offer three months of free credits was smart for a few reasons. It encouraged users to try Figma’s AI offerings and, hopefully, get hooked on them. It armed Figma with consumption and cost data from users in production. And it allowed Figma to collect feedback to finetune the exact pricing mechanics before sending out a bill.

They introduced usage tracking for both individual users and for admins and saw a ‘power law’ distribution in AI consumption. Figma says that 75% of paid customers with $10,000+ in ARR are consuming AI credits on a weekly basis. A subset of these users are already exceeding the credit limits, although Figma didn’t disclose the exact number.

The new pricing seems designed to help Figma sell more seats rather than simply sell credits. A Figma Dev seat comes with 500 credits per month. If these users upgrade to a full Professional seat, they’ll unlock 3,000 credits, worth an extra $60 per month – a massive savings relative to the cost of the seat upgrade ($5 per month). This is likely to encourage more customers to adopt Figma wall-to-wall.

What I struggle with is that there’s an inherent tension between whether Figma credits are value-based (you pay for work) or cost-based (you pay to cover their AI bills). Just take a look at the rate table below.

Making a prototype is a flat 20 credits per use. Generating an image, on the other hand, could set you back anywhere from 5 credits (about 12 cents) to 25 credits (about 60 cents). The difference: which LLM you select for generating the image.

The onus is on the user to make that price-value trade-off. Is Gemini 3 Pro really 3x better than Gemini 2.5? Is this use case valuable enough to splurge or am I OK with something lower quality? How would I know if I made the right decision?

Meanwhile, it’s unclear whether users will benefit if (when?) LLM costs go down. This means they’re on the hook to constantly make cost optimization trade-offs without being fully in control of the tools to optimize their bill.

There’s an inherent tension between credits as cost-based or value-based. A cleaner approach would be dual-track monetization: delineate value (the platform) and cost (the tokens) into separate buckets.

My simple brain likes to think of this as paying for your car lease (platform) and then paying for fuel as you drive (tokens). Or paying for your Costco membership (platform) and then the goods you buy at Costco (tokens).

Infrastructure software products like Snowflake or Splunk have long navigated this. Snowflake solved it by charging for storage and data transfer (passing on costs) along with compute and cloud services (delivering value for customers). Vertical software products do this by charging differently for payment processing (passing on costs with a small margin) compared to software modules (higher margin).

Platform + tokens monetization is absolutely more complex for buyers, which may be its biggest downfall. But it does a few things that I really like:

PostHog is a great example of platform + tokens. Their new AI pricing is exceptionally simple: it’s a pass-through of AI costs with a 20% markup. And every customer gets $20 of free usage to try it out.

The company can do this because their value is elsewhere: PostHog has 10+ other products they can monetize. AI features can be used across all products and make these products even more valuable.

Clay’s new pricing model takes this a step further, formally splitting their pricing into two axes: data credits (tokens) and features, actions (platform).

Their legacy pricing was already ahead of its time. Clay was a pioneer in credit-based pricing in a space (GTM tech) full of seats. All Clay plans have long included unlimited users.

In the past, Clay credits could be spent on data points that Clay had to pay for access. A mobile phone number lookup, for example, might cost anywhere from 2 credits up to 25 credits depending on the data provider.

That data credit pricing made sense in the early days of Clay, which started largely as a data marketplace. It doesn’t make sense for an automation platform that can execute complex workflows. Customers who largely want data might say Clay feels expensive while customers who largely want platform capabilities might say the opposite.

The new pricing marks a clear separation of cost (data) and value (actions). They’ve also simplified packages, going from five down to four, with clearer messaging about who each plan is for.

As part of the move, Clay reduced the data credit cost by 50-90%. They’ll also be passing through the cost of sophisticated AI models with a 0% markup; customers just pay for actual token consumption. They’re also letting existing self-serve customers keep the legacy pricing.

This means Clay expects to see an immediate 10% drop in revenue from this pricing move. Their bet seems to be that they can make it up as platform adoption grows. Clay’s new flywheel: (a) open up access to the product, (b) make it affordable to get GTM data, (c) make money as customers run increasingly complex GTM workflows.

The biggest knock on Clay’s pricing change is probably the complexity involved for customers. Usage-based pricing is already hard to predict; usage-based pricing with two axes is even harder.

Clay seems to be solving for that in two ways. First, they’ve set fairly generous thresholds around actions. The company told me that 90% of existing customers wouldn’t hit the current action limits. Second, customers can opt to bring their own data via API. This effectively takes data credits off the table, simplifying Clay’s pricing.

There’s another tailwind that might actually shift complexity from being a bug to a feature, as I recently discussed with Scott Woody from Metronome: AI agents as the new buyer of AI credits. AI agents already recommend what products to buy (👋 AEO). It’s not much of a stretch to imagine taking this a step further, particularly in the context of agents like Claude Code.

AI agents might make purchase decisions in a way that’s far more rational than people. They don’t succumb to charm prices, decoy effects, price anchoring, or the zero price effect. And they have the computing capacity to ingest large amounts of information including the fine print most of us would ignore. This means AI agents would probably prefer more complex pricing as long as there’s transparent documentation, usage tracking, and the ability to set budget caps.

I’ll end with a thought exercise for anyone who’s made it this far. If your AI bills immediately went down to $0, what would be the fairest way to charge for your product?

Look for a model that (a) maps to how your customers realize value from the product, (b) allows you to tell the story of why your product is uniquely better than alternatives, and (c) overcomes friction in the buying process so customers can easily say yes. And remember, there’s no such thing as the perfect pricing model.

Recommended resources: