Reflecting on my first few months at Runway and hundreds of customer conversations, there's a few emerging themes in AI that stand out. We’re still in the earliest stages of building real-world superintelligence. But our customers are already starting to shift the way they think about everything from cost optimization to model ownership.

Our business has more than doubled this year, and we’ve grown our NRR to over 300%. Many of our largest deployments continue to scale; one Fortune 20 grew their use of Runway over 17x this year. Our creative roots with companies like Lionsgate and Paramount will always be core to our ethos and design, but adoption from enterprises like Amazon, Microsoft, Allstate, Adobe, Robinhood and more has driven much of the recent growth.

We’re also seeing increased international adoption. Europe is our second-largest market, accounting for over 20% of our enterprise customer base, with subscription sales volume up 50% over the past twelve months. Japan has grown into our largest market in Asia, and one of our largest globally, driven almost entirely by teams that built Runway into their workflows before we had a commercial presence there. India and Brazil are two of our largest and fastest-growing self-serve user bases, and we’re increasingly seeing upmarket adoption as well. The pattern across every region is the same: enterprises arrive through a specific use case, then expand rapidly. That expansion is showing across every metric, from token usage to NRR.

A few specific themes I’m hearing consistently:

Models are converging. No single model-only provider holds a durable lead for long. As that gap closes, competition is shifting from which model is best to which product is best to build with and who can deliver it most efficiently.

Consumers and enterprises both demand an intuitive product experience, workflows that are easy to build and scale, unique and specific editing capabilities, collaboration tooling and integrations that make it easy to connect into the operating rhythm and systems of your business.

Data sovereignty is critical. Brand footage, product shots, campaign assets and character likenesses are created in Runway every day. Enterprises need to trust that a provider isn’t training on that content. This isn’t a given; it's amazing what some AI companies try to get away with if you read the fine print in their T&C's.

Token productivity and cost economics is top of mind for every CFO. Tokenmaxxing is well and truly dead, and productivity and ROI-maxxing is just getting started. Executives are just as eager to get AI working across their businesses, but they’re less willing to experiment, and more interested in targeting meaningful cost reductions. Our recent AI Media Report found production costs falling by two to three orders of magnitude across hundreds of enterprises; one financial services brand took a broadcast commercial that historically ran north of $5M, produced it for a few thousand dollars, and aired it on NFL Sundays.

Autonomous execution is what most companies actually want from AI: put these systems to work end-to-end, not just as a copilot for one step. Runway Agent has been key to removing the repetitive, expensive and slow parts of production so creators spend their time on the work that actually needs a person. Those time savings are allowing brands to 10x their creative volume (while still saving money), and many of our customers are using Runway to create campaigns and creative that would have been cost-prohibitive with traditional tools.

Ownership is becoming its own category of demand. A smaller but increasingly vocal group of the enterprises I talk to have started asking whether they can own a model outright. For a CTO or Head of AI sitting on proprietary IP, sensitive data or a compute bill measured in the tens of millions, it is worth asking: why route production-critical work through a shared model when you could hold the weights yourself?

Our Approach

Our teams are navigating this in a few key ways:

Owning Versus Renting: Model Licensing

Most of today's enterprise AI conversation is still about renting intelligence: call an API, pay per token, never touch what's underneath. But for certain types of companies, there’s another option: model licensing.

With a licensing agreement, we provide closed model weights and a proprietary training and inference framework directly to customers, who host the model inside their own environment, and fine-tune it on their own data. Four kinds of enterprises tend to end up here:

• Companies sitting on valuable IP - characters, catalogs or a body of creative work - who want a model trained on that IP and kept entirely in-house.
• Platforms with their own compute capacity and/or unique compute economics.
• Organizations with non-negotiable data requirements: government, regulated industries like healthcare and financial services or anyone required to run on their own infrastructure.
• Companies operating at a volume where a model tuned to their exact use case beats a general one on cost, quality and/or latency.

The Next Phase of Intelligence

We think that general world models are the fastest path to real-world superintelligence. As we scale our investment, team and research, we're committed to helping every organization see and think more clearly around the advancements in AI. The next phase of AI adoption won’t be determined by who has the best model, but rather who can make that intelligence trusted, usable, economically viable and operational at-scale.