Applied Materials filed an 8-K on Monday that says less than it does. The company signed an accelerated share purchase agreement for shares of BE Semiconductor Industries which, added to the 9% it bought in April 2025, will take it to about 20% of Besi’s outstanding stock. There is no price in the filing and no share count. Applied repeated the two lines it used eighteen months ago: it has no current intention to go past 20%, and it will not seek a seat on Besi’s board. Delivery of the shares waits on the Hart-Scott-Rodino waiting period.
That HSR condition is the first thing worth reading. The original 9% sat under the 10% ceiling of the exemption for voting securities held solely for investment. A 20% holder can’t use it, so Applied had to file, and the filing says more about the position than the line about the boardroom does.
The second is the number. Under US GAAP, 20% is where an investor is presumed to have significant influence over an investee (ASC 323), the point at which a stake normally stops being marked to market and starts contributing a share of the investee’s earnings. The presumption runs off the percentage. A board seat isn’t required. And 20% stops ten points short of 30%, the Dutch threshold that forces a holder to bid for the whole company. Applied has picked about the largest stake it can carry without being made to buy the rest.
The third is March. Besi traded to a record €199.85 on March 13 after Reuters reported takeover interest from both Lam Research and Applied, and Besi hired Morgan Stanley to evaluate the approaches. Lam was reported to still be interested afterwards. Dutch public offers normally carry high minimum acceptance conditions, commonly set at 95% and sometimes lowered toward 80%. With a fifth of the register in Applied’s hands, any rival bidder now needs Applied to tender. Whatever the intent, the effect is a blocking stake held by the one company Besi already works with every day.
Applied’s stated reason is hybrid bonding and what comes after it. The two companies have co-developed an integrated die-to-wafer hybrid bonding system since 2020, Besi joined Applied’s EPIC Center as an innovation partner earlier this month, and the partnership now extends beyond hybrid bonding into new interconnect architectures for AI chips. Applied could have built its own bonder. It chose to own more of the company that builds the most accurate one. That decision is the cleanest market signal on hybrid bonding this year, and it came from the world’s largest front-end equipment maker.
The rest of Monday’s filings line up behind it, layer by layer.
Amkor is the packaging layer, and its numbers show who carries the capital. Third-quarter net sales were $1.99 billion with net income of $191 million ($0.76 a share), helped by record Computing and Automotive & Industrial revenue. Fourth-quarter guidance is $1.9 to $2.0 billion of sales at an 18% to 19% gross margin, with net income of $160 to $185 million. Full-year 2026 capex is now about $2.5 billion. Add the four quarters and Amkor’s 2026 revenue lands near $7.5 billion, so roughly a third of revenue is going back into capacity at a gross margin in the high teens. For context, the whole company generated $308 million of free cash flow in 2025. The OSAT funds the factory. The toolmaker selling it the bonders doesn’t, and Applied bought the toolmaker.
Rambus is the interface layer, and it is the opposite case. Revenue was a record $218.3 million, up 22%, with product revenue of $114.3 million (up 15% sequentially), royalties of $77.0 million and record operating cash flow of $102.5 million. That’s nearly 47 cents of operating cash on every revenue dollar. Its chips (buffers, PMICs, now a PMIC for LPDDR6 SOCAMM2 modules) ride on the count and complexity of memory modules, and the memory price cycle is someone else’s problem. Rambus doesn’t care whose DRAM wins.
Celestica is the rack layer. Revenue rose 83% to $5.83 billion, adjusted EPS rose 109% to $3.30, and the adjusted operating margin hit a company record of 8.5%. The 2026 revenue outlook went to $21.0 billion from $20.5 billion. It also completed a $3.45 billion public offering. A record margin of 8.5% next to Rambus’s cash conversion is the whole argument in two numbers: integration scales revenue, and it needs equity to do it.
Cadence is the design layer, and it had the most interesting quarter of the group because the headline and the income statement disagree. Revenue grew 22% to $1.633 billion, backlog reached a record $8.2 billion, the full-year growth outlook moved to 21% with non-GAAP EPS of $8.40, and the board added $2.5 billion to the buyback after $400 million was spent in the quarter. GAAP net income was $289.7 million against $287.1 million a year ago. Flat. GAAP operating margin fell to 26.2% from 31.8% and non-GAAP to 44.1% from 47.6%, while cost of product and maintenance rose 46% to $173.9 million, which is consistent with a heavier hardware mix. Cash fell from $3.0 billion at the start of the year to $1.45 billion. The moat is intact (nobody tapes out a leading-edge chip without EDA), but this year’s growth is coming with less margin attached.
Intel’s $2.3 billion purchase of Classiq ($2.2 billion net of Classiq’s cash, in cash and stock, closing in the first half of 2027) belongs in the same list, and Intel said why. It compared Classiq to the arrival of EDA four decades ago: a synthesis layer that hides hardware complexity and turns design into an industry. That comparison carries a condition. Cadence is worth what it is worth because it serves every foundry and every chip designer. Classiq today runs on IBM, IonQ, Quantinuum and Rigetti machines and on Nvidia and Intel simulators, and its backers include AMD Ventures, Qualcomm Ventures and IonQ. Intel has promised to keep it hardware-agnostic after closing. It has to, because neutrality is the asset. A quantum EDA tool owned by one qubit maker is worth less to every other qubit maker’s customers, and Intel is paying about $15 million per employee for a company of roughly 150 people. Lip-Bu Tan, an early Classiq investor, recused himself from the talks.
Put the filings side by side and the money sorts itself. The companies that set a specification (the bonder, the interface standard, the design tool) keep high margins and attract strategic capital. The companies that build to the specification (the packager, the rack integrator) post bigger revenue growth, carry the capex and raise equity. Applied spent Monday buying more of the first kind.
The detail the 8-K left out will show up in Besi’s substantial-holdings notifications to the Dutch AFM, which trigger at 20% and again at 25%. If Applied’s notification lands at or above 20%, it wants the equity-method seat at Besi’s earnings. If it stops at 19-point-something, it doesn’t.
That filing is the one to watch.