Nvidia Keeps Writing New Records

Nvidia reported fiscal Q2 2027 results (May–July 2026) after the close on 26 August 2026 local time. On the numbers alone it is a record beyond argument. Yet immediately after the announcement the stock dipped nearly 2% after hours before settling around 4% higher.

That temperature gap is the heart of these results. The market’s scoring shifted long ago from whether the results were good to how far they beat expectations.

The reported numbers

Consensus was around $92.3 billion in revenue and $2.08–2.09 in adjusted EPS. Revenue came in 4% above, EPS above too. Data centre alone beat analyst estimates ($86.3–87.3 billion).

Shareholder returns came with it. The company returned $26 billion in buybacks and dividends in Q2, and the next quarterly dividend is $0.25 per share, payable 1 October. Remaining buyback authorisation stands at $99 billion.

The real headline is the guidance

Q3 revenue guidance is $108 billion ±2%. With consensus around $104 billion, that is above too. It means entering the band where quarterly revenue passes $100 billion.

Do not miss the caveat attached. This guidance excludes China data centre compute revenue. In other words the number is built with the China variable set to zero. If regulation loosens there is upside; if it does not, the plan still holds.

Margins are coming down, and the reason matters

What actually shifted the mood in this announcement was CFO Colette Kress’s comments on margin.

• Q3 gross margin guidance: 74% ±0.5pp (below Q2’s 75%)
• Q4 (November–January): down to 71–72%, where it should bottom

There is one reason. Component prices, memory above all, have risen. Kress said “today’s memory shortage was in significant part brought about by the build-out of AI infrastructure itself.” The demand the company created has come back as its own cost.

For Nvidia that is bad news. But read from the side that sells memory, the same sentence means the opposite. More on that below.

“We are filling only half the demand”

Kress said she expects fiscal 2028 revenue to grow about 70%. The basis for it, though, was not optimism but constraint.

We expect roughly 70% growth because of supply constraints — followed by an explanation to the effect that customer demand points to twice that.

So 70% is a ceiling on what can be sold, not the size of demand. One more number supports it. According to Korea Economic Daily reporting, Nvidia’s purchase commitments for memory and other components reached $279 billion, up 134% in three months. It means locking up volume in advance — the most honest evidence that supply is tight.

So why didn’t the stock jump?

This is the fourth consecutive quarter of “sell the news”. The pattern should be treated as established. The reasons are simple.

1. Expectations rise faster than results. Beat consensus by 4%, and if 5% was already priced in over the preceding days, nothing is left
2. A margin decline lasts longer than a growth rate. Revenue gets re-scored every quarter, but falling margin reads as a multi-quarter story
3. The market’s question changed. From “how much did you make this time” to “how long does this spending continue”

CEO Jensen Huang said “AI has reached an inflection point. It is doing useful work, tokens are productive and profitable, and compute is now revenue.” That remark is less a boast about revenue than an answer to the bubble debate.

What to watch from Korea

The part of these results most directly connected to Korea is not revenue but cost.

• If what is pulling Nvidia’s margin down is memory prices, the parties receiving those prices are Samsung Electronics and SK Hynix
• Analysts read this as a phase where pricing power moves from buyer to supplier
• Korea Economic Daily reported that Nvidia announced a multi-year technology partnership with SK Hynix on next-generation memory solutions

It is early, though, to convert that benefit straight into earnings. How much of the memory price increase lands in which quarter’s results depends on contract structure and shipment mix, and that cannot be known before each company reports.

In short

1. FY2027 Q2 revenue of $96.2B (+106%), data centre $89B (+117%). A thirteenth straight record quarter
2. Q3 guidance of $108B is above consensus and is a number excluding China data centre revenue
3. Margins fall to 74% in Q3 and 71–72% in Q4. The cause is component prices, memory above all
4. The 70% growth outlook comes from a supply ceiling, not from demand. Purchase commitments rose 134% in three months
5. The stock did not jump because expectations had risen first, not because results were poor. Fourth quarter of the same pattern

Frequently asked questions

Why is fiscal Q2 2027 the May–July 2026 period?

Nvidia’s fiscal year ends in late January. So the year beginning in February 2026 on the calendar is fiscal 2027, and May–July is its second quarter. That is why press coverage mixes “Q2” and “FY27 Q2.”

Does a falling margin mean results are getting worse?

If revenue grows faster, profit itself keeps rising. With Q3 guidance at $108 billion and a 74% margin, gross profit in absolute terms is larger than Q2’s. Share prices tend to react to direction rather than absolute amounts, though, so a phase of compressing margin generally weighs on valuation.

What does “excluding China revenue” mean?

It means the guidance was calculated with revenue from China data centre compute products set to zero. Export control conditions change often, so the company took a conservative approach, and any easing leaves that much as additional upside.

If results are good and the stock falls, what should I look at?

Look at the share price movement in the weeks before the announcement and how far consensus was revised up, not just the announcement itself. Results are scored against expectations already priced in, so “beat estimates” on its own cannot explain the direction of the stock. The figures here are as of 27 August 2026 and are not a basis for investment decisions.