Applied Materials Q3 FY2026 Results (13 August 2026): Free Cash Flow Snapped Back to $2.33bn from $210m — and China Fell by $42m, Not 7 Points

Applied Materials reported third-quarter fiscal 2026 results after the US close on Thursday 13 August. Revenue was a record $9.115 billion against a guide of $8.95 billion, non-GAAP EPS a record $3.50 against a $3.36 midpoint, and the fourth-quarter guide came in at $10.25 billion — roughly $710 million above consensus. But the line this post said to watch was free cash flow, which had collapsed 80% to $210 million in the second quarter. It came back to $2.330 billion. The mechanism behind that swing is visible in three balance-sheet lines, and it is a cleaner answer than the headline beat.

What actually happened

The release (primary source: Applied Materials' Q3 FY2026 press release) landed above the guide on every headline measure, for the quarter ended 26 July 2026.

Revenue landed $165 million above the midpoint of the company's own band; non-GAAP EPS $0.14 above it. Those are real beats but modest ones, and on their own they would have been the least interesting part of the release. Applied guides a tight range one quarter ahead and lands inside it consistently — as this post argued before the print, a "beat" against a company's own restated midpoint carries limited information.

The fourth-quarter guide carries considerably more: $10.25 billion ± $500 million of revenue and non-GAAP EPS of $4.02 ± $0.20. Reuters reported that consensus for the quarter had sat near $9.54 billion, making the guide roughly $710 million above expectation at the midpoint — a raise several times larger than the quarter's own beat.

The $210 million question, answered in three balance-sheet lines

This is the part worth slowing down on, because it is where a reader can check the thesis rather than take it on trust.

Free cash flow did not recover because profitability changed. Profitability was never the issue. It recovered because working capital stopped absorbing cash. The cash flow statement condenses that into one line — net change in operating assets and liabilities — and the two quarters read as follows.

The $1,764 million swing in that first line is almost exactly the $2,120 million improvement in free cash flow, once the $72 million of extra capital expenditure and the quarter's higher earnings are accounted for. Nothing else needed to happen.

Underneath, the composition is the actual evidence. Inventory rose only $221 million sequentially after building through the first half. Receivables rose $1,319 million. Those two facts together describe tools moving: stock that had been sitting part-built on the balance sheet shipped to customers, at which point it left inventory, was recognised as revenue, and appeared as an invoice awaiting payment. A receivable build is the signature of conversion, not a warning — it is the step immediately after the one that was in question.

And the reason a $1.3 billion receivable build did not itself crush cash flow is the third line: contract liabilities rose $701 million to $3.271 billion. That is money customers have already paid Applied for tools not yet delivered. Together with $558 million more in payables, it offset nearly the entire receivable increase.

China didn't shrink — the denominator grew

The most repeated line in coverage of this release is that China's revenue share fell to 28% from 35%, framed as a warning about a key market. The geographic table supports a narrower reading.

China's revenue fell by $42 million, or about 1.6% — effectively flat. The seven-point share decline is arithmetic on a total that grew $1.813 billion, and China contributed none of that growth. Meanwhile US revenue roughly doubled and European revenue roughly tripled, which is what onshoring fab construction looks like when it stops being an announcement and starts being a purchase order.

That distinction matters because the two conditions have opposite implications. A region whose revenue collapses is a demand or licensing problem. A region whose revenue is flat while other regions grow is a concentration change — and for a company that spent 2025 warning about exactly this exposure, a flat China alongside a doubling United States is a reduction in dependence, not evidence of one. The export-licensing risk itself has not gone away; it is simply not what these particular numbers show.

The mix broadened, which was not the expected result

Before the print, this post flagged the Semiconductor Systems customer mix as the third thing to watch, on the reasoning that DRAM rising with flash falling would confirm AI-memory concentration. The disclosed mix is more interesting than that, and honesty requires reporting it against both comparisons.

Year on year the expected pattern holds: DRAM up four points, flash down two. Sequentially it reverses — DRAM eased from 29% to 26% while flash recovered from 4% to 7%. On a segment that grew to $7.040 billion from $5.965 billion, a 26% DRAM share is still more DRAM revenue in dollars than a 29% share of the smaller base, so this is not a memory slowdown. It reads as flash returning from an unusually low quarter while DRAM grew slightly less quickly than the segment. A modest broadening, in other words, rather than the tightening concentration the year-on-year figures alone would suggest — which is the more durable of the two configurations, and the opposite of what a reader looking only at the annual comparison would conclude.

Segment profitability was the quieter achievement: Semiconductor Systems non-GAAP operating margin reached 38.0%, from 33.2% a year earlier, and Applied Global Services grew revenue 22% to $1.781 billion at a 30.1% operating margin. The site covered the supply side of the memory build when Samsung and SK hynix committed $38.1bn to new fabs, and the demand side when a DRAM shortage began showing up in PC makers' costs.

A beat, a raise, and a 5% fall

Reuters reported that shares fell more than 5% in extended trading following the release, attributing the move to "lofty investor expectations" and noting the stock had more than doubled during 2026 before the print. Brooks Idlet of CFRA told Reuters the results "were not enough to impress the Street, but the print was still solid."

The mechanism there is worth stating plainly, because it recurs across every heavily-owned name in this cycle. Published consensus is a visible number; the expectation embedded in a price that has already doubled is not. When a stock has run that far into a report, the release is measured against the second one, and no line in a press release can tell you where it sits. This is why a company can exceed its own guide, raise the next quarter roughly $710 million above consensus, and still see the shares fall within minutes — the results answer whether the business delivered, and the price move answers whether it was already owned. Those are separate questions with separate evidence, and conflating them is the most common error in reading an earnings reaction.

The same read-through logic ran through Cisco's report the day before, and through Coherent's the day before that, from different points in the same chain.

What it touches, and what it does not

Applied is a component of the S&P 500 and the Nasdaq 100, and not heavy enough in either to move an index much by itself. The transmission runs through inference. Equipment orders sit at the very front of the semiconductor production chain — before wafers, before chips, before the accelerators and datacentres that consume them. A fab that has announced construction has not yet spent anything at Applied; a fab that has taken delivery of tools has. The $684 million increase in US revenue is therefore one of the earliest hard measurements available of announced domestic fab capital expenditure actually being spent.

What it is not is a currency event. There is no plausible channel from an equipment maker's fiscal third quarter to the relative pricing of the majors, and forcing one would be worse than saying nothing. The one honest connection runs the other way, and it is mechanical: semiconductor capital equipment is overwhelmingly invoiced in dollars, while the fabs buying it earn revenue in won, Taiwan dollars and yen. A stronger US dollar raises the local-currency cost of an identical tool without changing its price. The dollar's own interest-rate, growth and risk factors are what the meter scores — the framework is explained on the about page — and the dollar reached this print having already been repriced by July CPI earlier in the week.

What would change the picture

Three things, now that the cash question has been answered.

First, whether the conversion sustains. One quarter of neutral working capital establishes that the pre-build shipped; it does not establish a run rate. With receivables at $7.691 billion against $5.185 billion at the October year-end, collection is now the line carrying the risk that inventory carried three months ago. The fourth quarter shows whether $10.25 billion of guided revenue arrives with cash attached.

Second, whether the $10.25 billion guide is compatible with the raised calendar-2026 framework. Dickerson said the company is "further raising our Semiconductor Systems revenue expectations for calendar 2026" and expects another strong growth year in 2027; the guide and that statement have to be arithmetically consistent, and the fiscal first-quarter guide in November is where any divergence appears first.

Third, the geography. If US and European revenue keep compounding at anything close to this quarter's rate while China stays flat, the concentration risk that dominated the 2025 discussion continues to shrink on its own — without any change in export policy. If US growth was a one-quarter delivery cluster, the share table reverts and the China question returns in its original form.

None of that requires a view on where the shares go. It requires knowing which line answers which question — and, this quarter, the line that answered it was one the headline never mentioned.

Educational macro context only — not investment advice.