Fabrinet Monday, Analog Devices Wednesday: the week the AI hardware trade gets audited from the supply side
Markets closed Friday with small caps the standout. The Russell 2000 finished at 3,068.42, up 0.51% and at a record close, the only one of the major indices to rise on the day, while the S&P 500 slipped 0.17% to 7,785.76, the Nasdaq Composite fell 0.28% to 26,729.16 and the Dow eased 0.20% to 53,732.41. Across the week the Russell added 1.11%, the S&P 0.36% and the Nasdaq 0.14%, while the Dow lost 0.56%. It was, in other words, a week in which the AI hardware complex kept moving without doing much to explain itself.
This desk spent Friday writing about exactly that problem: Applied Optoelectronics and Lumentum both rallied without a same-day catalyst anyone could point to. The coming week supplies the missing document. Between Monday's close and Thursday's open, the companies that actually build, test and supply the physical layer of the AI buildout report — and unlike a tape move, an earnings release has to put a number next to the claim.
The most consequential of them lands first. Fabrinet (NYSE: FN) will release results for the fourth quarter and fiscal year 2026, which ended June 26, 2026, after market close on Monday, August 17, with management holding a conference call and webcast at 5:00 p.m. EDT, according to the company's own announcement. That timing matters for how the week trades: Fabrinet's print will be the first hard datapoint the optics complex has had since Friday's unexplained move, and it arrives while the market is shut.
Fabrinet is worth more attention than its profile usually gets because it is not a brand — it is the factory. It is the contract manufacturer that assembles a large share of the industry's optical transceivers and photonic components, which makes its revenue a volume reading on what the optics names are actually shipping rather than what they are guiding. The concentration is extreme and the company discloses it plainly. In its fiscal 2025 annual report, Fabrinet stated that during fiscal year 2025 Nvidia and Cisco contributed 27.6% and 18.2% respectively of its revenues — 45.8% between two customers. The same filing shows the mix shifting rather than tightening: in fiscal 2024 the two contributed 35.1% and 13.4%, so Nvidia's share fell while Cisco's rose and the combined figure came down from 48.5%. Fiscal 2026, which ended in June, has not yet been filed. Even so, a Fabrinet quarter is, to a meaningful degree, an Nvidia optics supply-chain quarter reported by someone else.
The bar Monday tests is one Fabrinet set itself in May. For fiscal third quarter 2026, ended March 27, 2026, the company reported revenue of $1,214.3m against $871.8m a year earlier, growth of 39.3%. GAAP net income was $125.2m and GAAP diluted earnings per share $3.45; on a non-GAAP basis net income was $134.9m and diluted EPS $3.72. The distinction is not cosmetic — the roughly $0.27 per share gap between the two measures is the part of the result that depends on which convention you read.
For the fourth quarter the company guided explicitly. Fabrinet said it expects fourth quarter revenue to be in the range of $1.25 billion to $1.29 billion, GAAP net income per diluted share in the range of $3.48 to $3.63, and non-GAAP net income per diluted share in the range of $3.72 to $3.87, in each case based on approximately 36.3 million fully diluted shares. Set against the $1,214.3m it actually delivered in the third quarter, that revenue guide implies sequential growth of roughly 3% to 6% — a deceleration in pace, not in direction, and the specific thing a supply-side read should be judged on.
Chairman and chief executive Seamus Grady said in the May release that "Our third quarter was exceptionally strong, delivering record financial results that exceeded our guidance ranges." He went on to attribute the growth to ongoing and ramping programmes, and said the company expected several new customer agreements, particularly in datacom, to strengthen its trajectory moving into the fourth quarter. Whether those agreements have converted into shipped volume is the question the Monday call exists to answer.
On published consensus, Benzinga reported ahead of the print that analysts expect the company to report quarterly earnings of $3.81 per share, up from $2.65 per share in the year-ago period, on revenue of $1.28bn, attributing the figures to Benzinga Pro. Two cautions belong with that number. The sentence does not label the estimate GAAP or non-GAAP, though $3.81 sits inside Fabrinet's guided non-GAAP range of $3.72 to $3.87 and above its guided GAAP range, which suggests an adjusted basis. And $1.28bn sits in the upper half of the company's own revenue guide, meaning the estimate already assumes Fabrinet finishes near the top of what it promised.
Sell-side positioning going in is unusually dispersed, which is itself informative. Per Benzinga's compilation, Needham's Ryan Koontz cut his price target to $650 from $800 on July 22 while maintaining a Buy rating; Barclays' Tim Long raised to $702 from $548 on May 6 at Overweight; JP Morgan's Samik Chatterjee trimmed to $680 from $700 on May 5 at Neutral; Rosenblatt's Mike Genovese raised to $750 from $715 on May 5 at Buy; and, on the oldest entry in the compilation, B. Riley Securities' Dave Kang raised to $452 from $366 on February 3 at Neutral. A target cut of $150 that leaves the rating unchanged is a statement about multiple rather than about business, and the full spread — roughly $450 to $750, on notes struck as much as six months apart — is evidence of disagreement and of staleness, not guidance. This publication takes no view on any of these securities and none of these targets is our own.
One thing Monday will probably not resolve: the mix. Fabrinet's annual filing groups optical communications together without splitting telecom from datacom, so the disclosure does not permit a precise read on how much of the growth is AI datacentre interconnect versus carrier spending. What the filing does show is that optical communications fell as a share of the total, from 79.4% of revenue in fiscal 2024 to 76.6% in fiscal 2025, while automotive, industrial lasers and other markets rose from 20.6% to 23.4%. Anyone treating Fabrinet as a pure AI-optics instrument is reading roughly three-quarters of the company.
Wednesday brings a completely different instrument. Analog Devices (Nasdaq: ADI) said it will release third quarter fiscal 2026 results at 7:00 a.m. Eastern on Wednesday, August 19, and host a conference call at 10:00 a.m. Eastern the same day. Analog and mixed-signal semiconductors are a shorter-cycle, broader-exposure business than AI memory or compute — the parts go into factory automation, instrumentation, vehicles and base stations — and the sector has a long history as an early read on industrial demand rather than a late confirmation of it.
ADI's April-quarter result, for the period ended May 2, 2026, is the reason this one carries weight. Revenue was $3.62bn, up 37% year over year. On a GAAP basis, operating margin was 38.1% and diluted EPS $2.40; on an adjusted basis, operating margin was 49.0% and diluted EPS $3.09 — an eleven-point margin gap between the two conventions that no summary should collapse. The end-market detail is where the signal lives: industrial contributed $1.80bn, 50% of revenue, up 56% year over year; communications $555m, 15%, up 79%; consumer $398m, 11%, up 23%; and automotive $872m, 24%, up just 2%.
That table, from the quarter already reported, is the most interesting set-up going into next week. An industrial book growing 56% while automotive grows 2% is not a general semiconductor recovery — it is a sharply bifurcated one, with capital-equipment and factory demand inflecting hard while the auto channel sits flat. The 79% growth in communications also means ADI is not the clean non-AI control sample it is sometimes described as; it has its own datacentre-adjacent exposure. Wednesday tests whether the industrial inflection broadened, stalled, or was itself an inventory restock working through the channel.
For the third quarter ADI guided to revenue of $3.9bn plus or minus $100m, with reported operating margin of approximately 39.0% plus or minus 150 basis points and adjusted operating margin of approximately 49.0% plus or minus 100 basis points, reported EPS of $2.60 plus or minus $0.15 and adjusted EPS of $3.30 plus or minus $0.15. Against the $3.62bn just delivered, the midpoint implies about 7.7% sequential growth. Zacks pegs its consensus estimate for revenue at $3.92bn, suggesting growth of 36.3% from the year-ago quarter, and for earnings at $3.33 per share, indicating a rise of 62.4% — a figure Zacks notes has been revised downward by a penny in the past 30 days. Both sit fractionally above the company's own guidance midpoints — $3.9bn of revenue and $3.30 of adjusted EPS — and the comparison only holds on the adjusted line, which is the basis Zacks compiles. The company's reported-EPS midpoint of $2.60 is a different measure entirely, and the two sets of figures should not be blended.
Chief financial officer Richard Puccio framed the demand picture in May: "We continued to see growing demand in the second quarter with record bookings across our B2B markets of Industrial, Automotive, and Communications. These positive demand signals are reflected in our outlook for continued strong growth in the third quarter." Bookings that outrun revenue are the ordinary precondition for a guide like this one; the follow-through is what Wednesday scores. Chief executive and chair Vincent Roche separately said second-quarter revenue and earnings "were above the high end of our outlook, reflecting the combination of record demand and sharp operational discipline."
Wednesday is congested for a second reason. The minutes of the July 28-29 FOMC meeting are due at 2:00 p.m. Eastern that afternoon, four hours after ADI's call begins. Those minutes cover a meeting that held rates at 3.50%-3.75% on a 9-3 vote in which Hammack, Kashkari and Logan each dissented in favour of a quarter-point hike — a detail worth restating because the contested direction in this cycle is upward, not downward. An industrial-demand datapoint and the internal argument of a committee with three hawkish dissenters will be read against each other in the same session, with Jackson Hole following on August 27-29.
Three other reports fill in the week, each verified against the issuing company. Keysight Technologies (NYSE: KEYS) will report fiscal third quarter results, for the quarter ended July 31, 2026, after the close of the stock market on Tuesday, August 18, with a call at 1:30 p.m. Pacific, 4:30 p.m. Eastern. Test and measurement is a genuine leading indicator because instruments get bought before volume production, and Keysight's April quarter made the point loudly: revenue of $1.717bn, up 31%, against orders of $2.051bn, up 56% — orders running well ahead of revenue. GAAP diluted EPS was $2.02 and non-GAAP diluted EPS $2.87. The company guided third-quarter revenue to $1.730bn to $1.750bn, a midpoint representing roughly 29% year-over-year growth, with non-GAAP EPS of $2.43 to $2.49 on about 173 million shares. President and chief executive Satish Dhanasekaran said in May that "Keysight delivered the strongest quarter in the company's history, capping a record first half with all-time highs in orders, revenue, EPS, and free cash flow."
Wolfspeed (NYSE: WOLF) holds its fiscal fourth quarter call at 5:00 p.m. Eastern on Wednesday, August 19, and it is the week's necessary counter-example. Silicon carbide is in a downturn while AI optics booms. In the quarter ended March 29, 2026, Wolfspeed reported revenue of $150.2m against $185.4m a year earlier, a decline of about 19%, with a GAAP net loss of $119.9m, or $3.05 per diluted share, and a non-GAAP net loss of $128.2m, or $3.26 per diluted share. That is the unusual case in which the adjusted loss is the larger one, and the reason is worth stating: the GAAP figure is flattered by non-cash and one-off gains that the reconciliation strips out, principally a $28.7m gain on the change in fair value of liability-classified derivative contracts and a $10.0m gain on contingent cash. Non-GAAP gross margin was negative 21% and adjusted EBITDA negative $61.7m. The company guided fourth-quarter revenue to between $140m and $160m, with operating expenses approximately flat and gross margins expected to remain negative — a midpoint essentially flat on the quarter just reported. Its balance sheet reflects a restructuring completed on September 29, 2025, when Wolfspeed emerged from Chapter 11 having reduced total debt by approximately 70% and lowered annual cash interest expense by roughly 60%; it held $1.2bn in cash and short-term investments at quarter end against $922.2m of long-term debt and $798.3m of convertible notes. Notably, the company reported sequential growth in AI datacentre applications of approximately 30% — evidence that AI demand is real inside a business that is nonetheless shrinking, which is precisely why one datapoint should not be read as the whole cycle.
Tuesday morning carries the autonomy pair, both dual-listed and both reporting before the US open with calls at 8:00 a.m. Eastern: Hesai Group (Nasdaq: HSAI; HKEX: 2525), in lidar, and Pony AI (Nasdaq: PONY; HKEX: 2026), in robotaxis, which will publish second-quarter and interim 2026 figures. Daqo New Energy (NYSE: DQ) closes the week's frontier-adjacent slate with second-quarter results for the period ended June 30, 2026, before US markets open on Thursday, August 20, also with an 8:00 a.m. Eastern call. Polysilicon pricing is not an AI story, but it is a useful read on whether Chinese industrial overcapacity is still clearing.
What the week actually tests is narrower than the headlines will suggest. It does not test whether AI demand exists — Keysight's order book and Wolfspeed's datacentre line both say it does. It tests conversion: whether the demand showing up in bookings and in optics share prices is converting into shipped volume at Fabrinet, and whether the industrial inflection at Analog Devices is broadening beyond the half of its business that is already running hot while the automotive quarter stays flat. Those are answerable questions with numbers attached, arriving on a schedule. After a week when two optical names rallied and nobody could say why, that is a meaningful upgrade in the quality of available evidence.
