Netflix has just revealed one of the more revealing AI acquisitions of 2026.
In March, the company bought InterPositive, the AI filmmaking startup founded by Ben Affleck.
At the time, Netflix did not disclose the price
.
Now we know.
In a regulatory filing with the U.S. Securities and Exchange Commission, Netflix said it paid approximately $587 million in cash for an acquisition completed in March. The filing does not name InterPositive, but the timing and amount correspond to the Affleck startup acquisition announced that month.
InterPositive had a team of just 16 people.
That works out to roughly $36.7 million per employee.
But dividing $587 million by 16 misses the real story.
Netflix wasn't paying $36 million for a person.
And it wasn't simply buying another video-generation model.
It was buying something potentially much more valuable:
a way to make AI useful inside Hollywood without starting with a copyright and ownership nightmare.
Affleck's strange position on AI suddenly makes more sense
In January, Ben Affleck appeared on The Joe Rogan Experience with Matt Damon and was unusually skeptical about the AI boom.
He questioned the quality of AI-generated writing and argued that much of the industry's enormous AI valuation could be disconnected from what the technology could actually deliver.
He also made a distinction that now looks particularly important.
Affleck wasn't arguing that AI was useless.
He was arguing that AI's usefulness depended on what you asked it to do.
He viewed AI more plausibly as a production tool than as a replacement for human filmmakers.
Affleck said he considered it very unlikely that AI would simply write meaningful movies or make entire films “from whole cloth.”
Then the InterPositive story became public.
Suddenly, his skepticism looks less like an attack on AI itself and more like a distinction between badly designed AI products and properly designed production technology.
That distinction may be worth billions.
The problem with Hollywood's AI models
The obvious question is:
Why didn't Affleck simply use one of the increasingly powerful public video-generation models?
Because Hollywood has a problem that a normal AI startup doesn't.
Rights.
A model can produce spectacular footage and still be commercially dangerous if nobody can confidently answer:
Where did the model learn this?
Hollywood is built around intellectual property.
Films contain performances, scripts, cinematography, music, costumes, characters, sets and visual styles that belong to somebody.
Actors and writers have contracts.
Studios have distribution rights.
Producers have ownership agreements.
Unions negotiate protections.
Lawyers examine the chain of rights.
And then along comes an AI model trained on enormous quantities of material scraped from the internet.
Technically impressive.
Commercially useful.
Potentially complicated.
Affleck's response was to start somewhere different.
Build the training data yourself
According to Netflix, Affleck began developing InterPositive in 2022 after observing the early rise of AI in filmmaking.
He and a small team of engineers, researchers and creatives began filming their own proprietary dataset on a controlled soundstage designed to replicate the conditions of a real production.
The goal wasn't simply to teach a model what movies look like.
It was to teach the model how filmmaking works.
Cinematography.
Lighting.
Editorial consistency.
Visual continuity.
Missing shots.
Background replacement.
Production problems.
The strange inconsistencies that occur when thousands of individual decisions have to come together to produce a finished scene.
That is a fundamentally different philosophy from:
Give the model the entire internet and see what happens.
InterPositive was trying to give the model a controlled understanding of filmmaking.
This is where the dailies become important
The really interesting part of Netflix's strategy comes after that foundational training.
InterPositive's technology can train models using production dailies — the footage actually generated during the making of a particular film or television production.
Netflix says the tools can then be used for practical production and post-production problems, including missing shots, background replacements, reframing and lighting corrections.
Think about what that means.
A studio doesn't necessarily need a generic AI model that has absorbed thousands of unrelated movies.
It can have a system that understands the visual language of this production.
The actors.
The locations.
The lighting.
The lenses.
The camera positions.
The production's visual logic.
The continuity.
The footage that already exists.
Instead of asking:
“Generate me a scene that looks like a Hollywood movie.”
The filmmaker can effectively ask:
“Help me solve this problem inside the movie we are already making.”
That is a much more defensible commercial proposition.
The AI isn't replacing the movie. It's fixing the movie.
This may be the most important part of the acquisition.
InterPositive isn't primarily selling the fantasy of:
“Type a sentence and get a finished blockbuster.”
Its technology is aimed at the messy middle of actual production.
A shot is missing.
The background needs changing.
The lighting is wrong.
A frame needs to be extended.
A production problem has emerged after the cameras have stopped rolling.
Those are expensive problems.
And Hollywood encounters them constantly.
That makes InterPositive's technology much less glamorous than a text-to-video demo.
But potentially much more useful.
A two-minute AI-generated video clip may go viral.
A tool that saves a studio millions of dollars across hundreds of production problems is a business.
Netflix appears to understand the difference.
The real asset may be the data architecture
This is why the $587 million price tag becomes more interesting.
The model itself is not necessarily the scarce asset.
Models can be rebuilt.
Algorithms improve.
Open-source research spreads.
Computing becomes cheaper.
Competitors can hire engineers.
But legally usable, production-grade data connected to a controlled workflow is much harder to reproduce.
InterPositive began with its own proprietary dataset rather than relying exclusively on scraped material. Affleck specifically described the dataset as being filmed on a controlled soundstage with the characteristics of a real production.
The company also deliberately focused its foundational models on filmmaking techniques rather than performances.
That boundary matters.
Hollywood's most explosive AI disputes have often involved the question of whether technology is reproducing someone's identity, performance or creative work.
A system designed around production techniques rather than copying individual performances approaches the problem from another direction.
Netflix is already using generative AI
This isn't a speculative investment made by a company that has never used the technology.
Netflix has said that roughly 300 of its titles have already used generative AI.
That doesn't mean Netflix is generating 300 movies from prompts.
It means AI is already appearing in parts of its production pipeline.
Netflix executives have repeatedly framed the technology as a tool for filmmakers rather than a replacement for them.
Ted Sarandos has said that great artists are still necessary to make great work and that AI is intended to give them better tools.
InterPositive therefore fits into a much bigger strategy.
Netflix isn't necessarily betting that AI will make filmmakers obsolete.
It is betting that filmmakers who use better tools will eventually become more efficient.
That changes the economics
Imagine a production spends enormous sums creating a scene.
Then something goes wrong.
A shot is unusable.
A background has to be changed.
Lighting has to be corrected.
A scene needs to be reframed.
The traditional solution may involve reshoots, visual-effects work, additional crews, post-production time and scheduling complications.
If an AI system can solve some of those problems without compromising the creative intent of the production, the economics become obvious.
The value isn't:
“We made a cool AI video.”
The value is:
“We avoided an expensive production problem.”
That's a much easier business case for a studio CFO.
This may be the next phase of Hollywood AI
The first phase of generative AI was about spectacle.
Look what the model can create.
A spaceship.
A photorealistic human.
A fantasy city.
A movie trailer.
A talking animal.
The second phase may be much less spectacular.
Infrastructure.
Tools embedded inside professional workflows.
Models trained for specific industries.
Models trained on controlled datasets.
Models that solve specific production problems.
Models where ownership and provenance are easier to establish.
That's where InterPositive becomes interesting.
Netflix is not buying an AI company to compete with every general-purpose model in the world.
It is acquiring a system designed around its own industrial process.
And that could explain the price
The $587 million figure sounds enormous for a 16-person startup.
But acquisition prices aren't salaries.
Netflix isn't paying $587 million because it needs 16 employees.
It is buying:
- technology;
- intellectual property;
- proprietary research;
- a specialised team;
- filmmaking expertise;
- production workflows;
- training infrastructure;
- accumulated know-how;
- and potentially a strategic head start in AI-assisted filmmaking.
The 16 employees are simply the people who know how all those pieces fit together.
And that may be the most important asset of all.
The irony of Affleck's AI skepticism
There is a delicious irony here.
In January, Affleck sounded like one of Hollywood's AI skeptics.
Months later, Netflix paid $587 million for his AI company.
But those positions aren't necessarily contradictory.
Affleck wasn't saying:
AI is worthless.
He was essentially questioning whether generic AI systems could replace the human creative process.
InterPositive's philosophy is almost the inverse.
Don't replace the filmmaker. Build the machine around the filmmaker.
Don't train blindly on everybody else's creative work.
Build controlled datasets.
Don't ask AI to invent an entire movie.
Use it to solve production problems.
Don't remove the director from the process.
Give the director better tools.
That is a much more sophisticated position than either “AI will replace Hollywood” or “AI is useless.”
Netflix may have bought the legal moat
This is perhaps the biggest business lesson.
The next great AI companies may not necessarily be the ones with the biggest models.
They may be the ones with the cleanest data.
The most defensible rights.
The deepest workflow integration.
The strongest relationships with the industries whose data they need.
And the clearest answer when a lawyer asks:
“Where did this model learn that?”
That question is going to matter enormously in entertainment.
A generic model can demonstrate incredible capability.
But a production company needs something else:
permission.
Permission to train.
Permission to use.
Permission to modify.
Permission to distribute.
Permission to commercialise.
Permission to protect the people whose work made the model useful.
InterPositive's strategy is built around controlling that environment from the beginning.
The bigger AI lesson
There is a tendency to measure AI companies by their models.
How many parameters?
How fast?
How realistic?
How many GPUs?
How much compute?
But Netflix's $587 million acquisition suggests another metric deserves attention:
How usable is the model inside a real business?
A model that generates spectacular videos but creates unresolved rights questions may be less valuable to a major studio than a narrower model that solves a production problem safely.
The future of AI may therefore look less like one giant model replacing every profession and more like thousands of specialised systems embedded into existing industries.
Film.
Medicine.
Engineering.
Law.
Finance.
Manufacturing.
Advertising.
Each industry has its own data.
Its own vocabulary.
Its own workflow.
Its own liability.
Its own definition of quality.
And its own lawyers.
The companies that understand those differences may ultimately capture more value than the companies simply racing to make the biggest general-purpose model.
Netflix didn't buy the future of movies
At least, that's not what the evidence establishes.
It bought a 16-person team and a specialised filmmaking technology company for $587 million.
But the strategic message is much bigger.
Netflix appears to be betting that the valuable AI of the next phase won't necessarily be the AI that creates everything from nothing.
It will be the AI that understands exactly what professionals are already trying to do — and helps them do it faster, cheaper and with fewer production headaches.
That is a very different future from the one promised by the most spectacular AI demos.
And perhaps that explains the strange career arc of Ben Affleck's AI bet.
While everyone else was arguing about whether AI would replace Hollywood, he was quietly building an AI system designed to work inside Hollywood.
Netflix just paid $587 million to make that distinction official.
The model may be impressive.
The data may be valuable.
But the real asset is the workflow.
Comments
Post a Comment