Two companies lost $314 billion in a weekend when a Beijing lab gave a frontier AI model away for free. The same repricing already hit your business. Here is the one thing they could not download.
Two of the most valuable companies on earth lost a combined $314 billion in a single weekend.
Nobody hacked them.
No product broke.
No scandal, no lawsuit, no bad earnings call.
And the thing that did it was a free download.
Here's what happened, in plain English.
A lab in Beijing called Moonshot built an AI system about as capable as the best ones OpenAI and Anthropic sell.
Then, on a Sunday night, they gave it away.
Not access to it. The actual thing.
They posted it online for anyone to download and run, free.
The phrase people use for this is "open weight."
Strip the jargon, and it just means the recipe is public now.
Not the finished dish you rent by the plate.
The recipe itself, yours to keep, yours to cook at home.
2.8 trillion parameters, billed as the largest ever built.
Ignore that for a minute, because the size isn't the point.
Most of that machine isn't even running at once.
The point is simpler, and worse for the incumbents.
Something that used to cost a fortune to access is now free, and it's close enough to the best that most people can't tell the difference.
The market saw it instantly.
OpenAI dropped too.
Add it up, and it's $314 billion erased.
Not because either company got worse.
But because the one thing they were selling stopped being rare.
That's the whole story.
And please don't get rocked to sleep thinking this is a "tech story".
It's a market looking at two companies whose entire worth rested on owning something, watching that something get handed out for free, and repricing them on the spot.
Now hold that thought.
Because the same thing already happened to your business.
You just haven't gotten the memo yet.
Let's start with the companies that lost the money.
Because their problem is about to sound familiar.
For two years...
The whole case for paying OpenAI or Anthropic came down to one sentence:
You pay more because we're better.
That Beijing model, called Kimi K3, doesn't erase that sentence.
It just weakens it.
It costs about a third of what the American systems charge.
It's slower in spots, and less reliable on some tasks.
But "close enough, for a third of the price" is all it takes.
Now the premium has to be defended instead of assumed.
The response told you everything.
Within days the American labs quietly loosened their own limits, and handed out more usage to keep people from walking.
That's a price war, starting in real time.
The rent on being the best dropped the same week the news hit.
And this isn't one lucky release.
A few days ago Forbes put it plainly...
A top-tier open model now runs at about nine cents per million words of output, the same price for a giant bank and a two-person shop in Ohio.
Free and near-free models from China have already passed the American ones in downloads.
Expensive labs are becoming the premium option.
Not the default everyone runs on.
Now let's bring it home.
Your strongest argument has been the exact one the labs made.
You pay more because I'm better.
Something cheaper just got close enough that "better" is a question your buyer asks out loud now, instead of an answer they assume.
Nobody fired you. Nobody canceled the contract.
The floor just moved under the price you charge.
That advantage was a rental.
And the lease just got cheaper for everyone.
Watch the feeds, and you'll see the next wave already cresting.
I can't knock it. A lot of these people are truly brilliant.
And I know firsthand the blood, sweat, and tears it takes to get this stuff dialed in.
Give them credit. Some of it works.
The repeatable execution that used to require a hire is getting cheaper by the month.
Pretending otherwise makes you look like the last person to notice.
But I want you to really listen to how those stories get told.
It's almost never.
The clients show up as a number to be doubled.
"He went from three clients to six."
In other words, the win is always the founder's math, never what any one client actually walked away with.
And even here, the execution isn't as finished or polished as the pitch suggests.
Earlier this year a benchmark from Mercor tested leading AI agents on real management, consulting tasks, the kind built from surveys of people at McKinsey, BCG, Deloitte, and EY.
Because the doing is getting cheap.
And it's still getting the answer wrong more than half the time on exactly the work your clients pay for.
That's what a layer looks like while it's falling.
The ability to get the grunt work done without you is turning into something everyone will have.
And the moment everyone has it, it stops setting you apart.
Here's the part the "vacation crowd" never follows to its conclusion.
The moment everyone can offload the busywork, freed-up time stops being an edge and becomes the new baseline.
Everyone has it.
An input everyone can buy at commodity prices is a lasting advantage to no one.
Please don't underestimate how far this goes.
Watch the loudest voices in tech. They’re floating government checks. Imagining a world where money barely matters.
That’s them conceding, out loud, that doing the work is losing its scarcity
And the economists pushing back aren't arguing the opposite.
They're saying people will still command wages exactly where the scarce input is human presence, judgment, and accountability.
So both sides agree on the part that matters to you: labor that's merely doing the work is being repriced toward zero.
The market is already sorting itself this way.
PwC, reading more than a billion job ads, found that jobs requiring AI skills pay a 62% premium, and grow roughly eight times faster than the market.
Freelance data shows the same split happening faster than anywhere else.
AI is compressing the value of execution and inflating the value of judgment.
And the people who turn AI output into an actual result are watching their rates climb while everyone else's fall.
So the vacation test was a doorway, not the destination.
Walk through it...
And the real question is staring you in the face waiting for you on the other side.
If you and your competitor can both make AI do the monotonous work, what do you do with the time that opens up?
I'm wrong about a lot sometimes, so I could be wrong about this too.
But, as far as I can tell...
Dedicated human attention pointed at a specific client's result.
Judgment applied to their problem.
The thing a government check can't provide, and an agent can't fake.
Because it requires caring enough to figure out what a particular person actually needs and then building it for them.
Every layer of leverage drops.
And it's the only asset that gets more valuable as everything above it gets cheaper.
When every competitor holds the same models, your edge is knowing what to ask for.
And knowing your client's business well enough to aim the tools at their result instead of the generic output the tools produce by default.
That last part has a name the whole series is going to earn.
I call it K.A.S.H Flow.
Forbes pointed out something else worth taking with you this week.
Companies won't win the next decade on access to cheap models any more than they won the last one on access to electricity.
They'll win on judgment, trust, and knowing their own business better than anyone else can.
I want to be straight about my own version of this, because I'm not standing outside it.
I spend a ridiculous amount of time thinking about how to make these tools serve my clients.
Borderline obsessive, actually.
Caring was never the gap.
If you feel the same way...
I need you to know the gap was the lag.
Because there's a stretch between the moment a new capability shows up, and the moment you have a repeatable way to point it at a specific client's result.
For a while I was closing that gap "by hand".
One client at a time.
Reacting each time something new dropped.
It worked, but it lived in my head.
And anything that lives only in your head doesn't scale and doesn't survive a busy week.
I didn't fix that until I built an actual model for it.
So the freed attention lands on client outcomes on purpose instead of whenever I happened to get to it.
That's the difference I'm going to walk you through.
1. When a new capability drops into your lap for free, how long does it sit before any client feels it?
2. When your competitor has the same models you do, what does your client get from you that they can't get from a stranger running the same stack?
3. And who on your side of the table is actually dedicated to spending the freed attention on your clients' results, instead of letting it evaporate into more meetings and more AI slop?
If you don't have an answer, you don't have a problem with AI.
You have a problem with the model you're using to deploy it.
That's fixable.
It's most of what I do.
Over the next few weeks I'm going to walk you through the model I use to redeploy freed attention toward client results.
The map first.
Then the method, phase by phase.
The actual system.
The one that turns "the tools got cheaper" into "my clients get something a leveled field can't give them."
It starts with a single picture.
One image that shows you exactly where your business is still defensible and where it's already a commodity.
And what it takes to move from one to the other.
Next week: the compass.
Stay surgical,
Colin Taylor
Creator of The Asset Alchemy Method™
Sources & Further Reading
Ready to see what you're sitting on?
Book Your Diagnostic Call