Runway's 'No-Code App' Demo Is Really a $5.3 Billion Bet on Simulation

Runway says its new model, Solaris, can generate an app interface without writing code. The more useful way to read that claim is the opposite: Solaris doesn't use code at all. Beneath the interface there is no HTML, no CSS, no JavaScript. The whole screen is generated live, frame by frame, the way a video is — and each click or drag becomes an instruction for what to draw next.

It's a genuinely different thing from the "no-code" apps you may have heard about, and that difference is the whole story.

What it actually is

Solaris is what Runway calls an "interface world model," built on its Gen-4.5 video model and the GWM-1 world model. World models are AI systems trained to simulate an environment rather than just produce words or images; Runway's core claim is that they understand how things behave — light shifts, momentum, the weight of an object — instead of treating a screen as a static page. Solaris turns that engine into a real-time renderer, producing frames in under half a second.

Runway's own study of 250 people making nearly 7,500 pairwise judgments preferred Solaris over a Claude-built coded interface 61% of the time for following instructions and 71% for behaving naturally. That last word is the point: the model renders what an apple would do if dragged into a bowl, not because a programmer coded it, but because that's how the world behaves.

The "no-code" label is the wrong label

Here is the tension worth noticing. The headline frames Solaris as the next "no-code" builder — the category occupied by Lovable, Vercel's v0, Bolt, and Cursor. But those tools don't remove code; they generate it invisibly. v0 writes React. Lovable produces full-stack code you can deploy, debug, wire to a database, and hand to a maintainer. That code, unglamorous as it is, is what makes software production-grade. It can be tested, audited, patched, and — critically — read by screen readers and accessibility tools.

Solaris discards that substrate by design. And Runway is honest about what it costs. It calls Solaris a research preview, not a product. Legible text is still unreliable, there's no underlying document structure for assistive software to read, the model can produce convincing-but-wrong answers, and keeping a long session coherent is unsolved. It isn't publicly available yet; the company is accepting early-access requests and hasn't announced pricing. So "generate a full app" on a headline means "a demo of a research direction," not a working competitor to the tools that actually ship deployable software.

That's the skeuomorphism in reverse: an old label — "no-code app" — hiding a genuinely new behavior. But the new behavior is precisely why it doesn't yet deliver the old label's promise. For the jobs that make software real — accessibility, accuracy, maintenance, accountability — code isn't overhead; it's the thing that lets software be trusted. Solaris has to buy those properties back another way before it's an app technology at all.

The game it's actually playing

Runway is private, so you can't buy its stock, which makes this a story about expectations and about how to read an AI claim rather than a trade. To see why this demo matters, identify the game. Runway doesn't need Solaris to be a product yet. It needs it to be evidence.

In February the company raised $315 million led by General Atlantic at a $5.3 billion valuation — nearly double the $3 billion it carried in April 2025 — and roughly $1.05 billion in total funding. The story that justifies that price is not "AI video." It's world models: AI that simulates environments, usable across robotics, gaming, and now software. Solaris is proof of concept that the same engine can generalize beyond clips and render interactive interfaces, lengthening the runway for that narrative. Its real competition for investor attention is Fei-Fei Li's World Labs and Google's Genie, not Lovable.

Reinforcing that game is a distribution strategy aimed at where the money already flows: Runway signed a multi-year partnership making Adobe its preferred API creativity partner, and struck custom-model licensing deals with Getty and Lionsgate; it seeded a $10 million Builders Fund to get startups hooked on its API. That's the "money flow as moat" play — capture a recurring share of model spend in media, advertising, film, and enterprise, not consumer no-code subscriptions.

The economics behind the headline

Now the gap that matters. Sacra estimates Runway hit roughly $90 million in annualized revenue by mid-2025, up from about $70 million at the end of 2024, on a consumer subscription base plus enterprise API and licensing. Management projected $265 million to $300 million in annualized revenue by the end of 2025 — a steep back-half acceleration. For scale, Runway recognized about $44 million of revenue in 2024 against around a $155 million EBITDA loss, most of it compute and model-training cost.

Put the price against the revenue: a $5.3 billion valuation on a roughly $90 million run rate is on the order of 60 times annualized revenue. That multiple is a bet on the forecast and on world-model infrastructure beating the ordinary economics — not on what's already being sold. A demo like Solaris widens the frontier. It does not, on its own, add a customer, a dollar of recurring revenue, or proof that anyone will depend on the interface generation rather than the ad.

What would change the reading

The useful test — for Solaris and the world-model story generally — is adoption residue, not spectacle. Do users return and pay for a repeated job after the novelty fades? Does any partner ship a generated interface that meets the accessibility, accuracy, and reliability standards that code provides essentially for free? That's the evidence that would let the valuation outgrow the demo.

For now the honest conclusion is narrower than the headline. Solaris is a striking demonstration that a simulation engine can be bent toward software. It is not evidence of product-market fit, and the company's real economics — heavy compute losses against a revenue multiple in the tens-of-times range — are what will decide whether the pledges every quarter eventually turn into the paid customers that make a roughly $5 billion price feel like a bargain or like froth.