NVIDIA Beat on Every Line With Record Data Center Revenue — So Why Didn't Memory Stocks Follow?

The Short Answer

NVIDIA beat on every line in FY27Q2, with data center revenue hitting a record $89.0B. Memory and storage stocks did not follow.

The interesting part is not a post-hoc story about "profit taking" or "overreaction." It is this: before the print, three structurally unrelated markets — US options, Hong Kong central clearing, and the Korea Exchange — had each already separated storage from NVIDIA. The earnings report did not create the divergence. It exposed one that was already there.

1. How Strong Was the Print?

From the company's 2026-08-26 post-close disclosure (US GAAP, USD).

• Year over year: total revenue +106%, data center +117%.
• Versus pre-print consensus (41–44 analysts): non-GAAP EPS beat by 6.14%, revenue beat by 4.39%.
• Next-quarter guidance: revenue $108.0B (+12.2% QoQ), gross margin guided to 74.0%.

This print directly addressed the question implied by the 08-24 semiconductor selloff — whether upstream AI demand was breaking. At the reported-financials level, it was not.

2. So Why Didn't Memory Follow?

2.1 The options market never priced this as a sector event

This is the most overlooked and most informative piece. As of the 2026-08-26 US close (settled):

Companion option premium percentiles (computed from public option chains): Micron 14.2, SanDisk 1.9.

The implication is unambiguous. Going into the print, the options market priced NVIDIA as the single most uncertain day of the year and priced storage as the calmest stretch of the year. The market never expected this print to transmit into memory. Storage staying flat afterward is therefore consistent with its own pre-event pricing, not an anomaly.

2.2 Hong Kong central clearing: rotation within storage (settled 08-26)

Same session, same class of leveraged instrument, opposite directions. This is cutting Hynix and adding Samsung — not exiting semiconductors.

2.3 Korea Exchange: a different participant set, the same direction (settled 08-27)

Hong Kong participant data and Korean investor-type data come from two separate clearing systems with two different participant pools. On the same day they point at the same thing: cut Hynix, add Samsung.

3. A Detail That Reverses If You Read Only One Leg

This deserves isolating, because it is a textbook definitional trap.

Foreign net selling of SK Hynix as a share of daily volume, across three sessions: 30.6% → 7.79% → 5.05%. Read that alone and the conclusion is "foreign de-risking is nearly finished."

Add domestic institutions, and the combined net selling share reads: 7.71% → 9.49% — the opposite direction. It is widening.

The cause sits in the 08-27 session: foreigners sold a net 159,385 shares, while domestic institutions flipped from net buying 2,000 shares to net selling 140,329.

On the foreign leg alone, the selling is done. On both legs combined, it is not — the seller simply changed hands.

Any capital-flow conclusion citing a single investor category needs a zero-sum check first.

4. Intraday Reading (⚠️ NOT settled — reference only)

The following US prices were pulled live at 2026-08-27 10:34 ET / 2026-08-27 22:34 CST. The 2026-08-27 US session is still open; these numbers are still moving and are not the basis for any conclusion above.

The intraday shape matches the structure the three settled markets already described: NVIDIA and the semiconductor index on one side, the three storage names on the other.

5. What This Actually Means

1. "Record AI capex therefore memory benefits" is a transmission hypothesis that requires verification, not an identity. On this particular datapoint, it did not hold.
2. The divergence predates the print. Hong Kong on 08-26, Korea on 08-27, US options on 08-26 — all three had already separated storage from NVIDIA before the release. The earnings report is the reveal, not the cause.
3. Storage is diverging internally too. Hynix reduced, Samsung added, confirmed independently in two markets. Reading "the memory sector" as one bloc discards this layer.
4. Flow conclusions need a zero-sum check. The same dataset yields opposite conclusions depending on whether you count one investor leg or two.

6. Verification Nodes (falsifiable; not investment advice)

• US 2026-08-27 settled close: Micron and SanDisk returns relative to SMH, and whether their IV Rank leaves the floor (currently 11.9 / 3.2). Rising = the options market concedes storage carries event risk after all. Still flat = divergence confirmed.
• Next Korean session: SK Hynix combined (foreign + domestic institution) net selling as a share of volume — 9.49% on 08-27. Two consecutive sessions below 5% = de-risking complete; back above 20% = it has restarted.
• Samsung foreign flow: whether a third consecutive session of net buying occurs, and whether the magnitude holds above one million shares (1.382M on 08-27, down from 2.518M on 08-26 — already decaying).

Sources and Settlement Basis

The three blocks belong to different trading days and are not compared side by side.

This article is a compilation and computation of public information. It contains no position, trade, or execution information, and is not investment advice.