Runway Doubled to $200M ARR. Its Rival Is at $700M.

TL;DR

On September 8, Runway co-founder and co-CEO Anastasis Germanidis said the company's annual recurring revenue had reached $200 million, roughly double the $100 million it was running at in April. Bloomberg reported the milestone, and a source told it Runway expects to cross $350 million by the end of the year. The ARR is the headline. The number that actually explains it is one Runway published three weeks earlier: net revenue retention over 300%. And the number nobody put in a headline is that Higgsfield, valued within $100 million of Runway, reported three and a half times the revenue last month.

The ARR is the lagging indicator

Runway has existed since 2018 and spent most of that time as the default answer to "which AI video tool do I use." Revenue did not track the fame. Sacra puts its 2024 recognized revenue at roughly $44 million against a $155 million EBITDA loss. In February it raised $315 million at a $5.3 billion post-money valuation with a team of about 140 people, led by General Atlantic with Nvidia, Adobe Ventures and AMD Ventures in the round.

Then the curve bent.

Bloomberg reports most of that money still comes from the video models, which is worth pausing on. Runway's own homepage now leads with "Real-World Intelligence" and splits the company into Creative, Dev and Robotics. The pitch is world models. The invoices are still for ads.

What net revenue retention over 300% actually looks like

On August 20, Runway CRO Sean Holcombe wrote: "Our business has more than doubled this year, and we've grown our NRR to over 300%." From the same post: "one Fortune 20 grew their use of Runway over 17x this year."

Net revenue retention measures what your existing customers spend now against what that same cohort spent a year ago, after upgrades, downgrades and churn. Anything above 120% is considered excellent in SaaS. Over 300% means the customers Runway already had are, in aggregate, paying it more than three times what they were, before a single new logo is counted. It is the difference between a gym that has to sell fresh memberships every January and a gym whose existing members keep signing up for more classes.

The mechanism is boring, and it is the whole story. A studio buys a handful of seats to find out whether generated video survives a client review. It does. The seats stop being an experiment and become a line item in the production pipeline for every campaign, which means every brief, every region, every aspect ratio, all year.

Holcombe's post lists what a company demands before it will allow that: uncapped IP indemnification including for third-party models, no training on customer data, full ownership of outputs, SOC 2, ISO 27001, GDPR, SSO, and model licensing for on-premises deployment. None of that is a model capability. All of it is what turns four seats into four hundred. If you sell to enterprises, that is the transferable lesson, and it is a great deal cheaper than a frontier training run: the expansion revenue was unlocked by paperwork.

Now the uncomfortable comparison

Three weeks before Germanidis posted, Higgsfield announced a $400 million Series B at a $5.4 billion valuation, led by DST Global, and put the number in the headline of its own press release: $700 million in annualized revenue, up from about $20 million a year earlier per the Financial Times. It claims 390 of the Fortune 500 as users.

Two AI video companies, valuations $100 million apart, and a 3.5x gap in stated revenue. Divide one by the other.

Read that chart with the caveats attached, because they matter. Runway's $5.3 billion is a February mark and its ARR is a September number, so the multiple is calculated against a valuation seven months stale, and a new round would presumably reprice it. Higgsfield's mark is three weeks old. And the revenue figures are not the same unit: Runway reported ARR, which is recurring by definition, while Higgsfield reported annualized revenue, which need not be. The honest version is that the gap in strictly recurring revenue is smaller than 3.5x, and nobody outside the two cap tables knows by how much.

What survives the caveats is direction. On last known marks, investors are paying roughly three and a half times more per dollar of Runway's revenue than per dollar of Higgsfield's.

What the premium is actually buying

Runway's answer is that it is not an AI video company. It released GWM-1, a general world model, in December 2025. On August 31 it shipped Solaris, built on its Gen-4.5 video model, which generates a working software interface frame by frame in response to your clicks at 720p with no DOM underneath. Press a button and nothing handles an event; the model paints what the next frame should look like. It is request-only, with no public API and no published pricing.

Over the summer Runway also acquired Paris-based Kinetix, whose seven-person team worked on 3D human motion and physically grounded video generation, to seed a robotics line. Germanidis told Bloomberg that robotics will be one of the biggest applications of world models.

That is what the multiple is pricing: the bet that a model good enough to predict the next frame of the world is a simulator, and simulators sell into robotics, gaming and software, not only into advertising. Higgsfield is priced as what it visibly is, a very fast-growing enterprise video business. Runway is priced as an option on being something else entirely.

The caveats, straight-faced

• Every figure here is company-reported and unaudited. The NRR and the Fortune 20 expansion come from Runway's own blog. The $200 million comes from an executive's LinkedIn post, relayed by Bloomberg. Higgsfield's $700 million comes from its own press release.
• ARR is not revenue, and annualized revenue is not ARR. A run rate is one good month multiplied by twelve; recognized revenue for the year will be lower for both companies.
• NRR over 300% cannot persist. Expansion that steep is a cohort crossing from pilot to deployment once. A seat that has already converted cannot convert again.
• The last public profitability data point is ugly. Sacra's $155 million EBITDA loss on roughly $44 million of 2024 revenue is two years old, and nothing published since suggests video inference got cheap.
• $350 million by December is a projection attributed to an anonymous source, not a company commitment.

Key Takeaways

• Runway's ARR doubled from about $100 million in April to $200 million in September, with a source telling Bloomberg it expects to pass $350 million by year end.
• Net revenue retention is over 300% per Runway's own August 20 post, and one Fortune 20 customer grew its usage more than 17x this year.
• Expansion beat acquisition. The unlock was indemnification, data controls and compliance, the things that let a pilot become a production pipeline, not a better model.
• Higgsfield reported $700 million annualized in August at a $5.4 billion valuation, pricing it near 7.7x revenue against Runway's roughly 26.5x on a February mark.
• The premium is a bet on world models, GWM-1, Solaris and the Kinetix robotics hire, rather than on video generation.
• Treat all of it as marketing. Two revenue definitions, two valuation dates, zero audits.

Sources: Runway, "The Next Phase of Enterprise Video Generation", Bloomberg, Techmeme, PYMNTS, Higgsfield press release, TechCrunch, Sacra, Runway Solaris, Runway GWM-1