NVIDIA qualified all three HBM makers for Vera Rubin. The fight is now allocation.

The headline that crossed on June 5, 2026 reads like a clean win for memory: NVIDIA qualified all three HBM makers, SK hynix, Samsung, and Micron, to supply HBM4 for the Vera Rubin platform. $NVDA ticked up. The memory names got their green light.

Qualification was never the real question. Of course all three would qualify: NVIDIA is shipping an enormous platform and cannot single-source the tightest component in the stack. What the announcement does not settle is the only number that moves money, the allocation split, and on that the picture is unchanged: SK hynix keeps the lion's share. This piece walks through what actually got decided, where the volume goes, why $MU fell 7.7% on the very day it was qualified, and how the DRAM / HBM Memory bubble trades around news like this.

Why it matters now

Speaking in Seoul on June 5, Jensen Huang put it plainly: "All three vendors have been qualified. All three vendors are in production, and they're all racing to support Vera Rubin." It was the first public confirmation that all three memory makers cleared HBM4 qualification for the platform. He had already named Samsung, SK hynix, and Micron as Vera Rubin's HBM4 suppliers at GTC Taipei on June 1, with the platform now in full production and first shipments targeted for H2 2026.

The reason this matters is timing. HBM4 is the memory generation for NVIDIA's next accelerator cycle, and the qualification window is the moment the supplier roster locks. Lock the roster and the contest moves to the part the press release does not quote: how the volume gets divided.

Qualification is the floor, not the prize

There are two distinct events that retail coverage tends to collapse into one:

• Qualification. A vendor's HBM4 passes NVIDIA's technical and reliability bar and is cleared to ship for the platform. This is what was confirmed on June 5. It is necessary, but for a top-tier memory maker it is close to table stakes.
• Allocation. How much of NVIDIA's HBM4 demand each qualified vendor actually wins. This is contracted, it is where pricing and margin live, and NVIDIA has not disclosed it.

The market reads "all three qualified" as a leveling event, as if the three suppliers are now interchangeable. They are not. Qualification confirms each can ship; allocation decides who ships the most at the best terms. The June 5 news finalized the first and left the second exactly where it was.

Where the volume actually goes

Supply-chain analysts put the Vera Rubin HBM4 split at roughly:

| Supplier | Est. HBM4 share (Vera Rubin) | Access | | --- | --- | --- | | SK hynix | ~60-70% | KRX 000660.KS, pending US ADR (HXSCL) | | Samsung | ~25-30% | KRX 005930.KS | | Micron | the balance | $MU, NASDAQ |

The structure that has defined the HBM cycle holds: SK hynix is the leader, the first to ship HBM3E in volume and the front-runner on HBM4, and it carries the bulk of the platform's memory. Samsung is the recovering second source whose qualification matters most as relief for a supply-constrained NVIDIA. Micron is the third name, the one US investors can actually buy cleanly, capturing the remainder at improving but smaller share. Qualification of all three is the supply-side fix NVIDIA wanted; it does not reorder that hierarchy. For the per-name detail, see SK hynix (HXSCL) is heading to a US listing and Micron (MU) explained.

The tell: MU fell on the day it was qualified

Here is the part that should reframe how you read memory headlines. On June 5, the same session NVIDIA confirmed Micron's HBM4 qualification, $MU fell 7.7%. The decline tracked broad tech-sector pressure after employment data and Broadcom's results, not anything specific to Micron's memory.

That is the structural point the bull headlines miss. The memory bubble does not trade tick-for-tick on supplier press releases. It trades on the cycle (DRAM and NAND pricing, HBM allocation tightness) and on macro liquidity. A qualification win is a slow-burn fundamental input; a jobs print and a peer's guidance are same-day flow. When the two collide, flow wins the session. If you traded the qualification headline expecting a pop, the tape did the opposite.

This is why QA models memory as a bloc rather than as three independent single-stock stories. The names move together on cycle and flow, and the idiosyncratic divergence shows up later, on allocation disclosures and share shifts, not on the qualification date.

What HBM4 and Vera Rubin actually are

Briefly, for the evergreen frame:

• HBM4 is the next generation of high-bandwidth memory: stacked DRAM dies sitting beside the GPU, feeding it data fast enough to keep the compute fed. Each accelerator generation needs more HBM stacks and more bandwidth per stack, so HBM demand grows faster than unit GPU shipments.
• Vera Rubin is NVIDIA's next-generation accelerator platform, now in full production with first shipments targeted for the second half of 2026. It is the demand engine that the HBM4 qualification race is feeding.

The mechanics are the same ones laid out in HBM is the tightest bottleneck in the AI cycle: HBM is the choke point in scaling AI compute, the part of the stack with contracted pricing and the least spare capacity. Whoever owns the most qualified HBM4 capacity owns the best position in the cycle.

How to play the memory bloc

Micron ($MU, NASDAQ) is the only one of the three you can hold cleanly from a US account today. It is the default US-listed expression of the HBM thesis, the smallest HBM4 share but the easiest to own. Access mechanics for US-resident accounts are in /stack/ibkr.

SK hynix is the leader and carries the most HBM4 volume, but it lists on the Korea Exchange (000660.KS) with only a thin Frankfurt GDR and an illiquid US OTC ADR ($HXSCL) for now. That changes if its filed US ADR listing completes as targeted for end-2026, see SK hynix (HXSCL) is heading to a US listing. Track it live on /stocks/hxscl.

Samsung (005930.KS) is the broadest of the three, with memory as one segment of a much larger conglomerate, so it is the most diluted way to express a pure HBM4 view.

For most US investors the practical bloc trade is MU for liquidity now, with HXSCL on the watch list for the leader at a lower multiple once its listing is live. Both sit in QA's DRAM / HBM Memory bubble.

What to watch

• Allocation disclosures. Any concrete read on the HBM4 volume split for Vera Rubin and the generation after. That, not qualification, is the next real catalyst for relative performance within the bloc.
• Micron FY26 Q3 earnings (typically late June 2026). Watch HBM revenue and HBM4 share commentary against the SK hynix-dominant estimate.
• SK hynix US listing terms. Final size, the new-share component, and timing into end-2026.
• NVIDIA Vera Rubin shipment cadence. Each slip moves every supplier's HBM4 ramp window with it.
• Spot DRAM and NAND pricing plus macro flow. The bloc still trades on the cycle and liquidity, as the June 5 MU drop showed. Watch the flow, not just the fundamentals.

Live data on these tickers: /stocks/mu and /stocks/hxscl - price, ETF holdings, bubble correlation, bot positions.

Bubble context: /bubbles/memory - the DRAM / HBM Memory cluster these names belong to and how it's moving.

Adjacent reading: SK hynix (HXSCL) is heading to a US listing, Micron (MU) explained, and HBM is the tightest bottleneck in the AI cycle.

QuantAbundancia is educational research. Nothing here is investment advice. See /disclosures.