Analysis · Semiconductors · August 17, 2026
Intel Made $4.8 Billion Designing Chips and Lost $2.1 Billion Making Them
Intel returned to an operating profit in the second quarter — $1.80 billion, against a $3.18 billion operating loss in the same quarter last year. That is a swing of nearly $5 billion, and it is the headline the tape traded on. The segment table three pages into the 10-Q tells a more specific story: Intel’s product businesses earned $4.82 billion of operating income in the quarter, and Intel Foundry lost $2.09 billion. The company is running two businesses with opposite economics, and only one of them is currently a business in the ordinary sense — because 95% of what Foundry sold, it sold to Intel.
This matters because “Intel is profitable again” and “the foundry turnaround is working” are two different claims, and only the first one is supported by this filing. Decomposing the swing shows where the money actually came from, and it was mostly not the fabs.
1. Two businesses, opposite signs
Intel reports three operating segments — Client Computing and Physical AI (CCPG), Data Center and AI (DCAI), and Intel Foundry — plus an “All Other” category, corporate unallocated costs, and intersegment eliminations. Laid out side by side, the second quarter looks like this.
In millions of U.S. dollars, as reported. The two product segments earned $4.82bn between them; Foundry and unallocated corporate costs took $3.51bn back out.
The segment margins are as divergent as the signs. DCAI ran a 39.5% operating margin in the quarter and CCPG 26.4% — both genuinely strong. Intel Foundry ran at −36.2%. Put another way: the product businesses’ $4.82 billion of operating income arrives at $1.80 billion consolidated, a 63% reduction, once Foundry and unallocated corporate costs are subtracted.
2. Where the $5 billion swing came from
Year over year, consolidated operating income moved from $(3,176) million to $1,796 million — a $4,972 million swing. Because Intel discloses each segment for both periods, that swing can be attributed precisely rather than guessed at.
Change in each segment’s operating income, Q2 2026 versus Q2 2025, in millions. The six components sum to the full $4,972m swing.
Data Center and AI is the single largest contributor, at 37% of the swing, and it is the cleanest one: DCAI revenue grew 59% year over year and its operating income nearly quadrupled, from $633 million to $2,474 million. After several years of Intel being described as the company the AI build-out left behind, this is the quarter the data centre segment showed up in the numbers.
The second-largest contributor deserves a caveat rather than applause. Corporate unallocated costs improved by $1,339 million, and the income statement shows why: restructuring and other charges fell from $1,890 million in Q2 2025 to $170 million in Q2 2026. A smaller restructuring charge is a real reduction in reported cost, but it is not the same kind of improvement as selling more chips at a better margin. Roughly a quarter of the swing is of that character.
Foundry contributed 22% — a real $1,079 million narrowing of its loss. It is progress. It is not a turnaround.
3. What actually improved inside Foundry
The mechanism behind Foundry’s narrower loss is worth isolating, because it is mostly a volume story rather than a cost story.
In millions. Revenue grew 30.5% year over year while cost of sales and operating expenses grew 3.5% — which is the whole of the loss reduction. Costs did not fall.
Foundry revenue rose from $4,417 million to $5,765 million while its cost base rose from $7,585 million to $7,854 million. The loss narrowed because more volume went through a cost structure that barely moved — which is exactly what operating leverage in a capital-intensive fab business looks like, and is genuinely encouraging as far as it goes. But the cost line did not come down, and at current volumes the segment still needs roughly another $2.1 billion of quarterly revenue at similar incremental cost to reach breakeven.
4. The number I keep returning to
Intel Foundry booked $5.77 billion of revenue in the quarter. Intersegment eliminations for the same period were $(5.48) billion. Those eliminations exist because Intel Products buys manufacturing from Intel Foundry, and that internal sale has to be removed to avoid double-counting at the consolidated level.
The implication is that roughly 95% of Foundry’s reported revenue is Intel paying itself. The external foundry business — the part where Intel competes with TSMC and Samsung for third-party customers, which is the entire strategic premise of the segment existing as a reportable business — is on the order of a few hundred million dollars a quarter on these figures.
That is not hidden; it is arithmetic anyone can do from the same table. But it reframes what the segment currently is. Intel Foundry today is mostly Intel’s internal manufacturing arm, reported separately, running at a $2.1 billion quarterly loss. Whether it becomes a genuine merchant foundry is still a forward-looking question, and this quarter’s numbers do not answer it either way.
| Segment | Revenue Q2 26 | Op. inc. Q2 26 | Op. inc. Q2 25 | Change |
|---|---|---|---|---|
| Client Computing & Physical AI | 8,877 | 2,343 | 2,053 | +290 |
| Data Center & AI | 6,262 | 2,474 | 633 | +1,841 |
| Total Intel Products | 15,139 | 4,817 | 2,686 | +2,131 |
| Intel Foundry | 5,765 | (2,089) | (3,168) | +1,079 |
| All Other | 701 | 230 | 69 | +161 |
| Corporate unallocated | — | (1,416) | (2,755) | +1,339 |
| Intersegment eliminations | (5,477) | 254 | (8) | +262 |
| Consolidated | 16,128 | 1,796 | (3,176) | +4,972 |
5. The case for the other side
Internal is still real demand
A captive customer is a customer. Intel Products buying wafers from Intel Foundry is the same economic activity as buying them from TSMC, minus the margin paid to a third party. If the fabs are competitive on cost and yield, serving Intel first is a rational sequencing choice, not a failure of the strategy.
Operating leverage is showing up
Revenue up 30.5% against costs up 3.5% is the shape you want from a business that has already spent the capital. If that relationship holds for several more quarters, the loss closes without heroic assumptions about winning external customers.
DCAI changes the story materially
A 59% revenue increase and a near-quadrupling of operating income in data centre is the most important thing in this filing for anyone who owned Intel on a “they missed AI entirely” thesis. That thesis got weaker this quarter regardless of what the fabs did.
6. What this means for the book
Intel is a 4.13% position, held at a cost basis of $37.43 and currently up 140.98% — one of the larger winners in the book, and one I have written about before as a turnaround being judged in real time.
This quarter does not change the position. It changes which line I watch. I have been tracking Intel on consolidated profitability and on foundry process-node milestones. The segment table says the two most informative series are DCAI operating income — which is now doing the heavy lifting and is the reason the company is profitable at all — and Foundry’s cost base, which has not yet fallen and needs to, or needs to be spread over materially more volume.
There is also a portfolio-level observation. Intel at 4.13%, Micron at 5.38% and SanDisk at 7.94% put roughly 17.5% of the book in semiconductors before counting Broadcom, Arm or AMD. That is a deliberate thematic concentration, documented in The Concentration Premium, and it is the kind of clustering the Risk X-Ray exists to surface. Nothing in this filing changes that structure; it is context for reading a single good quarter at one of those names.
What I am not claiming
I am not claiming Intel Foundry will fail, and I am not claiming the segment is mismanaged. Building a merchant foundry is a decade-scale capital project and losses during the build phase are expected, disclosed and unsurprising. I have not modelled process-node yields, the capital-expenditure schedule, government incentive payments, or the terms of any external foundry agreement — all of which matter and none of which I have done the work on. The narrow claim is that this quarter’s return to operating profit was driven mainly by data centre and by a smaller restructuring charge, not by the fabs, and that the segment table says so directly.
What would change my read
WatchThree measurable things, in order
1. Foundry’s absolute cost base. It rose 3.5% year over year while revenue rose 30.5%. If costs start falling in absolute terms, the breakeven maths improves dramatically. If they resume growing with revenue, the loss becomes structural rather than transitional. 2. Whether DCAI holds its margin. A 39.5% operating margin on 59% revenue growth is exceptional and worth verifying across another quarter or two before treating it as the new baseline. 3. Any disclosure that shifts the internal/external revenue mix. The 95% internal share is the single number that would most change what Intel Foundry actually is. A material external customer showing up in the eliminations line is the tell.
Intel has spent several years being valued on a story about whether it can manufacture at the leading edge again. This quarter, the company made money — and it made it designing chips, selling them into the data centre, and booking a much smaller restructuring charge than a year ago. The fabs lost $2.1 billion, which was $1.1 billion better than last year. Both of those things are in the same filing, three pages apart, and only one of them made the headline.