Sora之死：当AI公司开始做减法

Just under two years ago, Sora's release had sent shockwaves through the entire tech world.

A tool that could generate realistic video from text input seemed like pure magic at the time.

Now, OpenAI casually said, we need to say goodbye to Sora.

On the surface this looks like cutting a product line, but it actually reflects a deeper shift in the AI industry.

When Sora first launched, the results were genuinely stunning.

The generated videos had characters, depth of field, and dynamic lighting.

The September 2024 upgrade added sound and physics simulation.

Water flowing, objects falling, collisions — it all looked convincing.

The phone app hit number one on the download charts on its first day.

From a product perspective, Sora was a success for a while.

But success and sustainability are two different things.

Video generation is extremely demanding on computing power.

Every video rendered requires massive GPU resources.

And GPUs remain the scarcest resource in the AI industry.

OpenAI had already been forced to limit users' video generation quota.

Not having enough chips isn't a problem you can solve just by throwing money at it.

The thing is, if you put the same computing power into code generation, logical reasoning, and text processing, the return on investment is much higher.

This is a straightforward resource allocation problem.

At the same time, the competitive landscape has been shifting.

Anthropic's Claude series of models has been rising rapidly in the enterprise market.

Their strategy is crystal clear: no images, no video, all resources concentrated on text and code.

This focus has built them an extremely strong reputation among engineers.

If OpenAI kept spreading resources across video, it would be ceding ground on the most critical battlefield.

So cutting Sora is less about giving up and more about cutting losses.

Of course, Sora's closure also involves some thorny issues.

First is copyright.

Users generated huge volumes of videos featuring copyrighted characters.

Mario, Pikachu, Lara Croft — the rights holders never authorized this.

The deepfake risks go without saying.

Then there's Disney.

Last year Disney announced a billion-dollar investment to partner with OpenAI and bring Disney characters into Sora.

That deal has now fallen through completely.

Disney's official response was, we respect OpenAI's decision.

But a billion-dollar partnership collapsing like that — the backstory is probably not so calm.

From a broader perspective, the death of Sora reflects the AI industry's shift from doing everything to doing only what matters most.

Over the past two years, AI companies have been aggressively expanding their product lines.

Text, images, video, music, code — they wanted to do it all.

But now everyone's starting to realize: computing power has a ceiling, funding has a ceiling, and engineer bandwidth has a ceiling.

Trying to do everything means doing nothing well.

OpenAI recently completed a 110 billion dollar fundraise, with a valuation of 730 billion.

They don't lack money; what they lack is focus.

Shutting down Sora is an act of subtraction.

The signal this sends to the industry is clear.

The next phase of competition among AI companies won't be about who can build the most products, but who can be the best at what matters most.

Video generation is cool, but it's not the decisive battleground.

Code, reasoning, enterprise-grade text processing — those are the directions that actually make money right now.

Sora's story ends here.

A product that once amazed the world, shut down in less than two years.

This is the brutal reality of the tech industry: cool doesn't mean alive.