SK hynix said on July 29, 2026 that it “began mass shipments of HBM4 in the second quarter and will ramp up production in the second half of the year,” in its second-quarter 2026 business results. HBM4 is the high-bandwidth memory generation that new AI accelerators, including AMD’s Instinct MI400 Series, are built around.
On this page
Second-quarter figures
SK hynix reported the following for the quarter, according to its release:
| Measure | Q2 2026 | Change vs Q2 2025 |
|---|---|---|
| Revenue | 79.3187 trillion won | +257% |
| Operating profit | 60.5426 trillion won | +557% |
| Operating margin | 76% | — |
Year over year, operating profit grew more than twice as fast as revenue, and about three quarters of the quarter’s revenue became operating profit.
What SK hynix said about HBM4
The company says its HBM4 “has demonstrated its differentiated technological edge by achieving customer-required operating speeds while delivering industry-leading power efficiency and cost competitiveness.” The speed and power-efficiency claims are SK hynix’s own.
The production plan has two steps: mass shipments started in the April–June quarter, and output ramps in the July–December half. HBM sits beside the GPU or accelerator die on the same package and feeds it data, so memory supply limits how many high-end AI accelerators can be built in a given period. AMD said on July 23, 2026 that its Instinct MI400 Series GPUs use HBM4.
Outlook
SK hynix said it will “reinforce both its production capacity and financial health by seamlessly preparing for mid-to-long-term growth opportunities while maintaining capital expenditure discipline.” It also cited long-term agreements finalized with about 10 key customers, alongside that commitment to expand production capacity.
What it changes
With HBM4 in volume shipment since the second quarter and a production ramp scheduled for the second half, accelerator makers that designed around the new memory have a supplier delivering it at scale. For buyers of HBM4-based accelerators, the second-half ramp is the next supply step. Accelerators that use HBM4 are listed with their memory specs in our AI accelerator catalog.
