The AI bubble looks like it is breaking

The Liquidity Map · Stock market report

The AI bubble looks like it is breaking

The name everyone knows, Nvidia, is flat on the year. The parabola was in memory and storage, and that is where it is breaking now. What a distribution after a parabola looks like, and where the money goes next.

July 30, 2026 · 9 min read

Key takeaways

The bubble was not in Nvidia. Nvidia is up 1% since January 1. The parabola was in memory and storage: Micron, Seagate and Intel rose between 258% and 285% into their June peaks.
From those peaks, Intel gave back 42%, Micron 39%, Seagate 30%. Even so, all of them remain far above where they started the year.
The signal that matters: SK Hynix reported record revenue, up 257% year over year, and the stock fell. When good news stops lifting the price, supply is bigger than demand.
The correction is concentrated, not broad. The semiconductor index is 29% off its high, while the S&P 500 sits just 4% below its own.
After a parabola comes distribution, then consolidation. Only then does the trapped capital look for somewhere to go, and crypto has been left behind by 28 to 36 percent this year.

Everyone looks at Nvidia when they talk about the AI bubble.

Nvidia is up one percent since January 1. Essentially flat for the whole year.

The bubble was somewhere else: in the companies that make the memory and the storage behind the data centres. That is where prices tripled in half a year, and that is where it is breaking now.

Definition

Distribution

The phase where large money exits a position but sells gradually, into strength, so it does not collapse the price on itself. It shows up through one simple sign: the news stays good, sometimes excellent, and the price stops rising on it. The opposite of accumulation, where the news is bad and the price refuses to fall further.

Where the bubble was

Not in Nvidia, but in memory

Nvidia closed at 190 dollars, 19% below its May high. On the year it is up one percent.

The name the press puts on the cover never went parabolic.

The real parabola was in the memory and storage suppliers. Micron rose 285% from January 1 into its June 25 peak. Seagate 280%, into June 22. Intel 258%, also in June.

These are old infrastructure companies. They moved slowly for years.

This year they were treated as growth stocks, because every data centre needs memory and drives.

When a hardware company almost quadruples in half a year, that is no longer a re-rating. It is a chase.

From the June peaks, the unwind was as fast as the climb. Intel gave back 42%, Micron 39%, Seagate 30%, AMD 26%.

In yesterday's session alone, Micron lost 10% in a single day, Intel 5%, AMD 5.5%.

And yet, even after those falls, Micron remains 134% above where it started the year, Seagate 166%, Intel 108%. The correction only looks violent if you measure from the top down.

+1%What Nvidia did since January 1, while Micron, Seagate and Intel climbed between 258% and 285% into their peaks. The bubble had a different address.

All indexed to January 1, 2026. Micron, Seagate and Intel go parabolic, Nvidia stays flat all year.

To see the shape of a complete parabola, look at a single name.

The shape

What a parabola looks like when it ends

Micron started the year at 315 dollars. On June 25 it was at 1,214.

Almost four times higher. In six months.

Within a few weeks it was back at 739. Not on any single piece of bad news, but because there was no one left to buy at those prices.

The climb took six months. Erasing a third of it took a few weeks.

This is the classic shape: a vertical acceleration into the end, a sharp peak with no consolidation before it, then an unwind faster than the advance.

In a healthy trend, price climbs in steps, rests, climbs again. In a parabola, the final months have no rest. That vertical part is exactly what gets erased first.

$1,214 → $739Micron, from its June 25 peak to now. A third erased in a few weeks, after almost quadrupling in half a year.

Micron, log scale, since last summer. The vertical climb from the start of the year, the June 25 peak, then the unwind.

But the clearest sign that this is distribution, not just a correction, came from somewhere else.

The signal

Record earnings, falling stock

This week SK Hynix, one of the largest memory makers in the world, reported record results. Revenue up 257% on the year, up 51% on the previous quarter.

The stock fell between 9% and 15%.

On record news.

The Korean market, home to both SK Hynix and Samsung, lost 12% in two sessions, its worst two-day move on record.

The news was excellent. The price fell anyway. That is the whole signal.

In a market that is still accumulating, news like that lifts the price. In one that is distributing, good news is exactly the liquidity someone large needed in order to get out.

It is not that AI demand disappeared. SK Hynix's numbers show the opposite. It is that the price had already run further than near-term revenue can justify.

Meanwhile the semiconductor index is 29% off its high for the year, while the S&P 500 sits just 4% below its own. This is not a market problem. It is one sector emptying out.

+257%The revenue growth SK Hynix reported year over year, a record result. The stock fell between 9% and 15% on that news.

The semiconductor index against the S&P 500, both since January 1. The correction is concentrated in one sector, not the whole market.

The question that matters for a portfolio is not how much further they fall. It is where the money goes next.

The rotation

What comes after the consolidation

After distribution you usually do not get a continuous slide. You get a sideways period, where price gets boring and volume dries up.

That is when the rotation shows. Money leaving an overheated sector does not sit in cash forever. It looks for something that has not already run.

For a whole year the money went one way. That leaves a lot of cheap things behind.

Look at what was left behind while capital sat in AI. Bitcoin is 28% lower than on January 1. Ethereum 36% lower. In the same year semiconductors are up 42%, even after the correction.

It does not mean the money moves tomorrow. Sector rotations take months, not days, and the sideways phase has to come first.

It means the list of assets left behind is long and cheap, exactly as the sector that absorbed everything starts releasing capital.

+42% vs -28%Semiconductors and Bitcoin since January 1, even after the chip correction. The gap shows how concentrated capital has been in one place.

Semiconductors, Bitcoin and Ethereum, all since January 1, 2026. A year of capital going one way.

Above all of this, the frame we read the correction in matters.

The frame

A sector correction, not the end of a cycle

What is happening now is a violent correction inside one sector, not a break in the market. The S&P 500 is 4% off its high and up 7% on the year.

On rates, my position has not changed since yesterday: yields are not falling, so no cuts, so no problem. Even a hike would not bring a cycle like 2022.

A parabola ending does not close the market. It moves the money elsewhere.

I stay with the underlying thesis. We are in a secular bull that runs toward 2032.

Inside it, entire sectors go parabolic, then give way, and the money moves on.

That is exactly what is happening to memory and storage now. It is not the first sector to do this in the cycle, and it will not be the last.

-4%How far the S&P 500 sits below its high, while the semiconductor index is 29% below its own. The correction has a precise address.

Which leaves the part that matters for you: what you do with this.

How we're positioned

That is the diagnosis. The part that matters for you is what you do with it.

We do not buy a name just because it is 40% off its high. A price 40% lower, after a 285% climb, is still 108% above where the year started.

We do not catch the falling knife. We wait for the sideways phase, the boring one, and only there do we build a position, in steps, not all at once.

On the assets left behind, the logic runs the other way: there we accumulate patiently, on the average price, not on any single day's low.

And on the frame, we do not mistake a sector correction for the end of a cycle. The broad market is 4% off its high, not in a crisis.

Conclusions

The AI bubble was not in the name on the cover. It was in memory and storage, where prices almost quadrupled in half a year.

That is where it is breaking now, with 30 to 42 percent erased from the June peaks, though all of them remain far above where the year started.

The signal confirming distribution rather than a simple correction: record results that no longer lift the price.

What usually follows is a sideways, boring phase. And after it, the rotation into what was left behind.

The broad market did not break. One overheated sector is emptying its positions, inside a cycle that still stands.

What to watch this week

The next strong report from a memory company: does the price rise on it, or ignore it again? That is where you see whether distribution is done.
The sideways phase: is volume drying up and the range tightening? That is the consolidation the rotation starts from.

The AI Bubble Looks Like It’s Bursting

Keep reading

2027 is THE YEAR

The consensus wants an October bottom. A green August says otherwise

The math for a rate cut