SK Hynix Sets Q3 2029 for HBM4E Output in Indiana

SK Hynix said Thursday it plans to begin volume production of next-generation HBM4E memory chips at its Indiana facility in the third quarter of 2029, with the site eventually reaching annual capacity of hundreds of thousands of wafers.

SK Hynix has finally attached a quarter to its American manufacturing plans. The South Korean memory maker said Thursday that volume production of its next-generation HBM4E chips at its Indiana facility will begin in the third quarter of 2029, and that the site is eventually expected to reach annual capacity of hundreds of thousands of wafers.

That is a long runway by the standards of a market where allocation for high-bandwidth memory is currently negotiated a year or more in advance and sold out well before it is built. It is a short one by the standards of semiconductor construction, where a greenfield site involves land, power, water, ultra-clean space, tooling, qualification runs and a workforce that in most cases does not yet exist locally.

What HBM4E Actually Is, And Why Indiana Matters

High-bandwidth memory, or HBM, is DRAM stacked vertically and wired into an AI accelerator so that data can move between memory and logic far faster than conventional memory allows. It is the component that decides how much of an expensive GPU's compute actually gets used. HBM4E sits in the generation beyond the HBM4 parts now entering the market, meaning SK Hynix is committing US capacity not to today's product but to the one after next.

The Indiana plan, as the company describes it, is oriented toward advanced packaging — the stacking, bonding and testing steps that turn DRAM die into a finished HBM stack. That work has historically clustered in Asia, close to the foundries and the assembly chains. Moving a meaningful slice of it onto US soil is the part of the chip supply chain that Washington has struggled hardest to relocate, precisely because packaging is labor- and process-intensive rather than simply capital-intensive.

The capacity language is worth reading carefully. "Hundreds of thousands of wafers" a year is a plant of real consequence, but it is also an eventual figure rather than a day-one one, and the company has not published a phasing schedule. As BNN Bloomberg reported, the third-quarter 2029 date applies to the start of volume production, not to full-rate output.

The Customer On The Other End Of The Line

SK Hynix's position in HBM is unusual in the memory industry: it is the incumbent supplier to the dominant buyer of AI accelerators, Nvidia, in a product category where qualification is slow and switching is painful. Samsung and Micron are the two credible alternatives, and both have spent the current cycle trying to close the gap on yield and thermal performance rather than on headline specifications.

That competitive frame is what makes a 2029 start date a strategic statement rather than a logistics note. By committing to HBM4E in Indiana, SK Hynix is signaling that it expects to still hold the lead generation three product cycles out, and that it expects its largest customers to want at least part of that supply made in the United States. Hyperscale buyers have grown increasingly explicit about geographic diversification in their component sourcing, and a US-packaged HBM stack is a differentiator that no amount of price cutting from a competitor's Asian line replicates.

There is a second, quieter audience. Advanced packaging capacity in the US is a bottleneck that constrains everyone — including domestic logic manufacturers and the foundry capacity being built out on American soil. A large HBM packaging site changes the local ecosystem math for tool vendors, substrate suppliers and test houses that currently have little reason to staff up in the Midwest.

Three Years Of Execution Risk Between Here And There

The gap between announcement and output is where chip projects go wrong. The recurring failure modes are well documented across the current wave of US fab construction: construction cost inflation, contractor and skilled-trade shortages, permitting and utility interconnection delays, and the slow grind of qualifying a new site to the same yield as an established one. Memory adds a further wrinkle — it is the most brutally cyclical corner of semiconductors, and a facility greenlit during an AI capacity crunch may come online into a very different pricing environment.

Investors watching this project should treat the 2029 date as a milestone to be reconfirmed rather than a fixed point. The things that will tell you whether it holds are mundane and public: site preparation progress, utility agreements, tool orders, hiring postings, and whether the company reiterates the quarter at subsequent earnings calls. A slip of a quarter or two would be unremarkable in this industry. A slip of years, or a re-scoping of what the site produces, would say something about how SK Hynix reads the HBM demand curve at the end of the decade.

A Tape That Is Still Paying For AI Capacity

The announcement landed on a session that was friendly to the semiconductor complex. The Nasdaq 100, tracked by QQQ, closed at $721.11, up 1.37% on the day from a prior close of $711.37, and finished within a whisker of its intraday high of $721.35. The S&P 500 proxy SPY closed at $771.10, up 0.66%, while the Dow-tracking DIA rose 0.19% to $535.22. All figures are as of the last trade at 20:00 GMT on Thursday, 27 August 2026; markets are closed.

The spread between the tech-heavy index and the industrial one is the market's ongoing verdict on where earnings growth is coming from. Announcements like this one are the physical infrastructure underneath that verdict — capital being committed years ahead of the revenue it is meant to serve, on the assumption that demand for AI memory in 2029 will be at least as intense as it is now.

What To Watch Next

Three questions matter more than the headline date. First, phasing: how much of the "hundreds of thousands of wafers" arrives in the first year versus the fifth. Second, product mix: whether Indiana stays a pure HBM4E site or absorbs later generations as they arrive, which would tell you the company sees the location as permanent rather than a hedge. Third, customer commitment: any disclosure of long-term supply agreements tied specifically to US-made output would convert a construction plan into contracted revenue.

For now, SK Hynix has done the one thing the market had been asking for since the Indiana site was first floated — put a quarter and a capacity band on it. The next three years decide whether both survive contact with a construction schedule.

Key facts

• Volume production start: Q3 2029, SK Hynix Indiana facility
• Product: Next-generation HBM4E high-bandwidth memory
• Eventual capacity: Hundreds of thousands of wafers per year
• Nasdaq 100 (QQQ) last close: $721.11, +1.37%, as of 20:00 GMT Aug 27, 2026

Frequently asked questions

When will SK Hynix start producing chips in Indiana?

SK Hynix said on Thursday, August 27, 2026, that it plans to begin volume production of its next-generation HBM4E chips at the Indiana facility in the third quarter of 2029. That date marks the start of volume output, not full-rate production; the company has not published a detailed phasing schedule for how capacity ramps after that point.

How big will the Indiana plant be?

SK Hynix expects the site to eventually reach annual capacity of hundreds of thousands of wafers. The company framed this as an eventual figure rather than a day-one one, so the first year of volume production in the third quarter of 2029 would likely represent a fraction of that stated long-run capacity.

What is HBM4E?

HBM stands for high-bandwidth memory — DRAM chips stacked vertically and wired directly into an AI accelerator so data moves between memory and compute much faster than with conventional memory. HBM4E is a generation beyond the HBM4 parts now reaching the market, meaning SK Hynix is committing US capacity to a future product rather than today's.

Who competes with SK Hynix in high-bandwidth memory?

Samsung and Micron are the two other credible suppliers of high-bandwidth memory. SK Hynix has held the lead position in the current cycle as a supplier to Nvidia, the dominant buyer of AI accelerators. Qualification for these parts is slow and customer switching is difficult, which tends to entrench whoever is ahead on yield.

Why does US-based advanced packaging matter?

Advanced packaging — the stacking, bonding and testing that turns DRAM die into a finished HBM stack — has historically been concentrated in Asia and is the hardest part of the chip supply chain to relocate. A large US packaging site would give hyperscale buyers a geographically diversified source and would draw tool, substrate and test suppliers to the region.

How did chip-related indexes trade the day of the announcement?

As of the last trade at 20:00 GMT on August 27, 2026, the tech-heavy Nasdaq 100 proxy QQQ closed at $721.11, up 1.37% from a prior close of $711.37. The S&P 500 proxy SPY closed at $771.10, up 0.66%, and the Dow-tracking DIA closed at $535.22, up 0.19%. Markets were closed.

Sources

• SK Hynix to start AI chip volume production in Indiana in 2029 — BNN Bloomberg

Photo: EqualStock IN · Pexels Licence — source