Beat, Raise, Down 18%: The Punishment Phase Is Back

The structural read behind today's Morning 10 — the software half of the night's damage, what the three different sell-offs have in common, and the memory test that is running in parallel.

What the pre-market is doing

As of late pre-market, Datadog trades near 232 against a 283.17 close — down about 18% — after beating on every line, raising its full-year outlook, growing revenue 36% year over year to $1.12 billion and printing $279 million of free cash flow. AppLovin is down about 19% at 338. HubSpot has deepened overnight’s fall to roughly −23% at 193. Around them: Snowflake −6.1%, Atlassian −5.8%, MongoDB −5.6%, Salesforce and Cloudflare −4.6%, ServiceNow −3.3%. The software ETF IGV is down 2.7%.

And the index is not participating in the drama: QQQ −0.5%, semis −0.3%. This is not a market event. It is a repricing of specific business models, name by name, while the tape around them stays quiet.

The Nokia pattern, software edition

The common feature across the casualties is the one investors know from legacy hardware — Nokia, IBM, the long list of businesses that had a real new thing growing inside a big old thing. The old core decelerates. The new piece grows fast but from a base too small to carry the whole. For a while the market accepts the blended number. Then one quarter it stops — and starts pricing the two speeds separately.

That is what a revenue guide cut plus an earnings guide raise means at HubSpot: cost coming out faster than the top line decelerates, customers up 14% but revenue per customer up only 4%. The seat engine — the old core of every front-office SaaS model — is stalling, and the agentic offset is not yet big enough to carry the number. Once one company shows that arithmetic in a filing, the market re-marks everyone with the same shape. Hence the gradient: seat-priced names down 3–6%, consumption and security names barely touched.

Three different courts, one sentence

AppLovin is the honest miss. Revenue of $1.92 billion came in below the company’s own guidance and below consensus, with a softer third-quarter outlook — management cited a lighter-than-normal pace of advertising-model improvements. Down 19% is what a genuine guide-down costs at a growth multiple.

HubSpot is the disruption print. It beat the quarter, cut the full-year revenue line, raised the full-year earnings line. That is not a bad quarter; it is a changed trajectory, and it is the first time this cycle the AI-disruption question has appeared as two numbers in a filing rather than as a thesis.

Datadog is neither. Nothing in the print decelerated — 36% growth, beat and raise, record large-customer adds. The stock went in 0.6% off its 52-week high carrying a premium multiple, and the market handed back 18% anyway. There is no whisper number high enough for a stock priced for perfection — the same mechanism that sold SanDisk at six times earnings works in reverse at forty times. And the one concrete overhang in this morning’s reports gave the premium a reason to compress: OpenAI, one of Datadog’s largest customers, is reportedly building its own in-house observability tooling.

That last item is bigger than one ticker. The frontier labs are the fastest-growing customers of the entire AI-opex complex — observability, edge, data warehousing — and they are also the organisations most capable of replacing those vendors from the inside. Their stated ambitions run toward infrastructure businesses, not applications; OpenAI building its own Datadog is the same motion as a lab building its own CDN or its own warehouse layer, and it puts a question under Cloudflare, Snowflake and every name whose growth story leans on AI-native accounts. The biggest customer and the likeliest future competitor are the same company — that is a new kind of concentration risk, and today is the first session the market has priced it.

The foil pair resolved in one session

Yesterday’s print-record cards set up exactly this test. Datadog: ten-for-ten on EPS, ten-for-ten on revenue, the cleanest record in the series, no reaction premium, entering at the highs. AAOI: five-for-ten and three-for-ten, the worst record on the board, the most bought into the print. One session later: the perfect-record name is down 18%, AAOI is down 0.8%.

The record did not protect the holder. The entry price did the damage. That is the whole argument of the series in a single trading day — the print record tells you about the company, the reaction tells you about the positioning, and positioning is what gets settled in the first session.

Memory’s test is running in parallel

Micron is down about 5% in the pre-market — the first cash-session test of this morning’s frame that the Korea leg was positioning, not pricing. The guided gross margins, read after the Morning 10 went out, support the frame: SanDisk guided 83–85% gross margin on revenue up 15–20% sequentially, with bits on allocation beyond calendar 2027; Western Digital guided 55.5% and above-consensus revenue and earnings. Nobody guided pricing down. The chart grids below show where Datadog’s print and the two memory prints left the tape.

If Micron’s decline stays contained in single digits and closes off its lows, the positioning read holds. If it breaks with the Korean names into the double digits without any new information, the unwind is larger than the frame allows and we will say so.

C · members block

Into the close

Five checks carry today, in order.

MU‘s close vs its −5% pre-market. Contained single digits, closing off the lows = positioning confirmed. A double-digit break on no new information = the frame is too small.

IGV −2.7% vs HubSpot −23%. As long as the ETF stays inside roughly 3% while the casualties sit at −18% to −23%, this is names being repriced, not software as a class. If IGV closes beyond −4%, the class is moving.

DDOG at 232. Whether the pre-market low holds through the first hour tells you whether the 18% was the full sentence or the first instalment. No opinion needed — just watch whether cash buyers show up where the overnight sellers stopped.

IEF at 93.31. Unchanged, still resting on the two-year trendline, still the veto over every bullish read on this board. Three sessions named, none resolved.

QQQ −0.5%. If the index stays this quiet while single names lose fifths of their value, today goes in the book as discrimination, not distribution. That distinction — which court is in session — is the only thing the close needs to answer.