Runway started by helping filmmakers. Now it wants to beat Google at AI. - BERITAJA

AI video procreation startup Runway doesn’t person the emblematic Silicon Valley pedigree. No Stanford founders, nary ex-Google founders, nary nine-figure seed information that bought them clip to disregard revenue. Its 3 founders — 2 from Chile, 1 from Greece —  met astatine NYU’s Tisch School of the Arts and built the institution successful New York.

Runway besides could be, depending connected who you ask, 1 of the about consequential AI companies today. Not because of what it has built, but because of what it is trying to build next. 

For the past respective years, the AI manufacture has mostly operated connected the premise that intelligence lives successful language. Large connection models for illustration OpenAI’s ChatGPT and Anthropic’s Claude bespeak that bet. 

Runway, alongside different competitors, is making a different one. Its founders judge the adjacent shape of AI intelligence won’t beryllium built from text, but from video and world models that study really the world works, not conscionable really humans picture it. That favoritism sounds academic. Its implications are not. 

Runway co-Founder and co-CEO Anastasis Germanidis said training models straight connected observational information from the world is the adjacent frontier of AI. The companies that get location first, he argues, won’t beryllium the ones who perfected language.

“We’re fundamentally bound by our ain knowing of reality,” Germanidis told TechCrunch from Runway’s homey sunlight-filled office adjacent Union Square.  

“Language models are trained connected the full internet, connected connection boards and societal media, connected textbooks — distilling the existing quality knowledge,” Germanidis continued. “But to get beyond that, we request to leverage little biased data.”

Founded successful 2018, Runway built its estimation connected video-generation models — including its latest Gen-4.5 — and AI devices that fto group move matter prompts into editable, cinematic content.

Today, Runway’s exertion powers accumulation workflows for filmmakers and agencies, and the institution has signed deals pinch awesome media players for illustration Lionsgate and AMC Networks. Its devices person moreover been utilized successful films specified arsenic “Everything Everywhere All At Once.” 

Runway is now weighted astatine $5.3 billion and, according to 1 of its founders, added $40 cardinal successful yearly recurring revenue successful the 2nd 4th of 2026.

If Runway’s stake that video procreation is the way to world models pays off, the consequence will beryllium felt from Hollywood to supplier discovery. If it doesn’t, Runway risks being outpaced by competitors pinch acold deeper pockets — Google main among them.

Taking the leap

Within the past six months, the startup has put its scheme into action and expanded beyond video generation, launching its first world exemplary successful December, pinch plans to motorboat different this year. (World models are AI systems that simulate environments good capable to foretell really they’ll behave.) 

Runway isn’t unsocial successful its pursuit of turning physics-aware video models to world models, pinch near-term usage cases successful interactive entertainment, gaming, and robotics training. Startups Luma and World Labs are connected a similar trajectory, and Google has pointed its Genie world model successful the aforesaid direction.

Everyone is aft immoderate type of the aforesaid thing: AI that solves humanity’s hardest problems. That’s acold from Runway’s original product, but it’s the consequence of some emergent capabilities successful the exertion and founders who were predisposed to travel wherever it led.

For his part, Germanidis sees world models arsenic technological infrastructure. The much sensory information and observations you train a azygous exemplary on, the person you get to a moving integer copy of the beingness — 1 you could tally experiments connected faster than immoderate laboratory could. Much of the technological process is conscionable waiting connected results, he points out. If you could compress that waiting, you could compress advancement itself. 

“If we could build a amended intelligence than quality scientists, we could accelerate advancement successful really we understand the beingness and really we lick problems,” Germanidis said.

The moonshot

Runway streetwear merch astatine the company's AI Summit successful March 2026. Image Credits:Runway

Germanidis fell successful emotion pinch programming arsenic an 11-year-old successful Athens and came to the U.S. astatine 18 to study neuroscience and film. He turned backmost to machine science, moving astatine respective Silicon Valley tech firms earlier deciding he’d had capable of the culture. Co-CEO Cristóbal Valenzuela, calved and raised successful Santiago, studied economics arsenic an undergraduate earlier moving successful movie and past software. Another Santiago native, Chief Innovation Officer Alejandro Matamala-Ortiz studied advertizing and ran a creation firm. 

The 3 met successful 2016 while attending NYU’s ITP (Interactive Communications Program), a postgraduate programme that Valenzuela described arsenic an “art schoolhouse for engineers.” 

The co-founders had each aspired to beryllium filmmakers astatine definite points successful their lives, according to Matamala-Ortiz. So Runway started pinch a elemental mission: Can we usage AI to make everyone a filmmaker? 

After releasing their first video procreation exemplary successful February 2023 — which is staggeringly unimpressive compared to what Runway is putting retired coming — that ngo evolved into: Could we make everyone a great filmmaker, according to Matamala-Ortiz. 

It required increasing the squad to what it is today. The institution has 155 workers dispersed crossed offices successful New York, London, San Francisco, Seattle, Tel Aviv, and about recently, Tokyo. "But passim this process, we learned that these models could understand really the world works, and if you standard them, they could beryllium useful for galore different different things,” he added.

Things for illustration robotics, supplier discovery, and ambiance modeling — the kinds of problems that person stumped researchers for decades. Last year, Runway launched a robotics unit which Germanidis says has already resulted successful real-world testing and deployments.

Germanidis, like others, sees the section heading toward training a azygous exemplary connected galore different modalities — text, video, voice, and different sensors — and thinks the compounding effect is the point. 

His ain moonshot extremity for Runway’s technology, fixed capable clip and resources, is biologic world models and anti-aging research.

Whether Runway could transportation its video power into world models is acold from settled, and the title isn’t waiting around. Runway was among the first to AI video generation, but world models are a different title pinch deep-pocketed and well-respected competitors. Google, erstwhile Meta main intelligence Yann LeCun, AI’s ‘godmother’ Fei-Fei Li, and a increasing section of startups are each chasing the aforesaid goal. 

Kian Katanforoosh, CEO of AI skills benchmarking institution Workera and a teacher astatine Stanford, pointed retired that nary 1 has yet proven the jump betwixt video intelligence and generalized reasoning via world models, but that doesn’t mean it’s impossible. He said that if Runway wants to move its world exemplary stake into reality, it will request to proceed gathering resources — compute main among them.

Runway has deals pinch CoreWeave and Nvidia, but wouldn’t corroborate whether it has dedicated cluster entree — the benignant of guaranteed, large-scale compute that training frontier models requires. 

“How are you going to build a foundational exemplary without a cluster?” Katanforoosh asked. “I don’t deliberation anybody could do that.”

Runway has raised $860 cardinal to date, including a $315 cardinal round successful February from strategical partners for illustration AMD Ventures and Nvidia. That’s about successful statement pinch its about contiguous competitors, Luma AI and World Labs, which person raised $900 cardinal and $1.29 billion, respectively, according to PitchBook. 

But Runway is besides going up against incumbents for illustration OpenAI, which has raised about $175 cardinal per CEO Sam Altman, and tech behemoth Google, whose genitor institution Alphabet is worthy $4.86 trillion. Google is Runway’s biggest threat. The company’s Veo exemplary competes straight pinch Runway’s video procreation business, while its Genie world exemplary targets the aforesaid longer-term territory Runway is racing towards. 

Katanforoosh nodded astatine OpenAI, which shuttered its video level Sora successful March aft burning about $1 cardinal per day successful compute costs pinch hardly $2.1 cardinal successful gross according to immoderate estimates. His point: resources unsocial don’t guarantee survival. They don’t guarantee it for Runway either.

Katanforoosh isn’t penning Runway off. He pointed to AI audio startup ElevenLabs, which has outperformed OpenAI and Google connected their ain benchmarks, contempt lacking the resources and pedigree of either. Runway, he argues, could travel a akin playbook. 

The comparison isn’t mislaid connected Runway’s founders. Valenzuela says the startup’s deficiency of Bay Area “standardization” gives them an edge. Not only do they person diverseness of thought, he contends, but without Silicon Valley ties, they had to beryllium scrappier, lacking the warfare thorax galore of their peers person entree to that would person insulated them from the request to make gross early.

And according to Michelle Kwon, Runway’s main operating officer, the institution isn’t successful a unreserved to raise much funds, moreover arsenic compute demands summation pinch scale.  

“Their inheritance has led them to beryllium early, to beryllium correct much often than not, and to build a civilization that moves incredibly quickly,” early investor Michael Dempsey, managing partner astatine Compound, told TechCrunch. 

For Valenzuela, that civilization starts pinch really he sees the world successful the first place. He spends immoderate free clip he has — not much, arsenic a co-CEO and caller begetter — reference books, including the Chilean writer Nicanor Parra, whom he describes arsenic the antithesis of Pablo Neruda: little formal, little academic, holding a position that poesy belongs to the group alternatively than to rules. 

“Rules are conscionable rules they invented,” Valenzuela said. “That’s a driving unit of really we do things astatine Runway. They opportunity Silicon Valley is present and that’s wherever the startups are. Why? Those are conscionable made up rules. Scrub them each and commencement again.”