SK Hynix Is Sold Out Through 2029 as Seagate Locks In Demand Into 2030

Listen to the audio version of this article (generated by AI).

Editor’s note: “SK Hynix Is Sold Out Through 2029 as Seagate Locks In Demand Into 2030” was previously published in October 2026 with the title, “Micron’s Latest Earnings Reveal Why the AI Memory Supercycle Could Run for Years.” It has since been updated to include the most relevant information available.

When Micron (MU) reported its quarterly earnings in early October, we told you Wall Street had developed a peculiar complaint: Business was booming, but perhaps it wasn’t booming quite spectacularly enough.

Revenue had surged. Earnings had exploded. Management’s forecast pointed to another record quarter. Yet, after the report, investors still found reasons to worry. Margins could be peaking. The beat wasn’t as big as previous beats. Memory is cyclical, after all.

Our answer then was that investors were drawing the wrong conclusion from the history. The question that actually matters is how much earning power these companies can sustain before supply catches up with demand – and Micron’s evidence suggested that catch-up could take years.

This week, the two other giants of memory and storage took the stand. And both testified for the bulls.

Seagate’s AI Storage Demand Now Stretches Into 2030

On Oct. 27, Seagate (STX) – the leading supplier of the high-capacity hard drives that store the world’s AI training data – reported fiscal first-quarter revenue of $4.23 billion, up 61% year-over-year. Earnings came in at $7.68 per share, versus $2.61 a year ago. CEO Dave Mosley called it “another quarter of record profitability and our highest ever free cash flow.”

More important than the record results was what Mosley said about visibility. The vast majority of Seagate’s near-term supply is already allocated under long-term agreements running through 2028, with some cloud customer agreements extending through calendar 2030. In his words: “Data creation, retention and reuse continue to compound storage demand with long term customer demand visibility reinforcing our view of a multiyear growth phase” – with that visibility “extending beyond 2030.”

And the market’s response to all this? The stock slipped, with analysts flagging – you guessed it – slowing margin growth.

Sound familiar?

SK Hynix Says DRAM and HBM Are Effectively Sold Out Through 2029

Then, the next day, SK Hynix (SKHY) – the high-bandwidth memory leader, newly listed on the Nasdaq this summer – delivered what may be the single most emphatic data point of this entire supercycle.

Quarterly revenue crossed 100 trillion won ($74.4 billion USD) for the first time in company history. Operating profit soared 632% year-over-year, blowing past Wall Street estimates on both lines. DRAM average selling prices rose by a high-20s percentage in a single quarter. Server DDR5 shipments jumped more than 40% sequentially. And HBM4 – the next-generation memory bound for Nvidia’s newest AI platforms – is set to surpass HBM3E in both shipments and revenue this quarter as production ramps.

But here’s the number we care most about: SK Hynix has secured long-term agreements with 14 major customers covering more than 50% of its production capacity for 2027 and ‘28 – and management considers its DRAM and HBM output effectively sold out through 2029, with demand expected to outpace supply through 2030.

The company is so confident in that trajectory that it announced plans to return roughly 100 trillion won to shareholders – at least 30 trillion in cash dividends, plus 60 trillion in buybacks and cancellations – while also significantly increasing investment next year.

Micron, SK Hynix and Seagate Point to a Multiyear Supply Squeeze

If customers expected plentiful supply and falling prices around the corner, they would have no reason to reserve capacity years in advance. They could simply wait and let suppliers compete for their business.

Instead, look at what the whole complex just told us, within a span of four weeks:

Micron: More than 75% of fiscal 2027 shipments already committed, allocation discussions moving into 2028, some agreements stretching beyond 2030 – and management still can’t identify when DRAM supply catches up with demand.
SK Hynix: Over half of 2027-28 capacity locked under long-term agreements, effectively sold out through 2029, demand outpacing supply through 2030.
Seagate: Near-term supply allocated through calendar 2028, cloud contracts into 2030, demand visibility beyond it.

That’s DRAM. That’s high-bandwidth memory. And now it’s hard drives, too.

Every tier of the data storage stack, sold out years in advance, confirmed by three different management teams with three different customer bases.

One supplier saying this could be company-specific enthusiasm. All three saying it, in the same month, is structural.

Now, as we’ve said before, volume commitments improve visibility, but they don’t make every future shipment irrevocable, and committed volumes don’t always carry fixed prices.

Still, when the entire supply chain is telling the same story, we give that story a lot more weight.

AI Agents Could Add Another Layer of Memory and Storage Demand

The story extends beyond the high-bandwidth memory attached to AI accelerators, too.

AI agents need the rest of the server to do useful work: CPUs to execute tasks, system memory to keep information available, and storage to retain and retrieve it. Micron specifically identified agents as a growing source of CPU demand, with implications for memory and storage across the system. Seagate’s results show the storage leg of that stool is already bearing weight.

And according to Micron management, customers are currently limiting memory per system just to stretch scarce supply across more deployments. That suggests today’s purchases may understate what customers would install if availability improved – a second wave of demand waiting on supply that, per this week’s reports, isn’t arriving anytime soon.

The Bottom Line: The AI Memory Supercycle Still Has Room to Run

Our bullish view doesn’t require memory and storage to stop being cyclical. It requires this expansion to last longer than investors currently expect. And the market keeps telling us expectations remain remarkably low.

Micron trades at roughly six to seven times forward earnings. Seagate just posted record profitability and its highest-ever free cash flow – and the stock fell anyway. These are skeptical valuations for companies whose customers are contractually locking up supply into the next decade.

We believe that disconnect – years of committed demand priced like a cycle about to roll over – remains the opportunity in the AI memory supercycle.

As demand grows, the companies supplying essential components gain both pricing power and a clearer view of future orders. Memory and storage are those constraints today. Power and physical infrastructure are next.

For investors willing to look beyond the most familiar AI names, those dependencies open up a much broader field of opportunities.

Explore our research on the infrastructure behind the AI buildout.