Marvell’s $115 Billion Lumentum Deal Tests Wall Street’s Faith in the AI Boom as Chip Stocks Slide

Marvell Technology agreed on Monday to buy optical-components maker Lumentum in a $115 billion all-stock deal, one of the largest technology takeovers of the year and the boldest bet yet that the artificial intelligence building boom will keep paying for years to come. Wall Street’s first reaction was not applause. Marvell’s own shares fell nearly 13 per cent on the day, Lumentum plunged almost 19 per cent, and the wider AI trade slid with them, in a session that asked an uncomfortable question of the most expensive trade in the market: how much more can the AI boom cost before investors stop paying for it?

The deal, announced before the market opened on Monday, would unite Marvell’s custom chips and networking hardware with Lumentum’s lasers and optical parts, the components that move data at high speed between processors and servers inside the giant data centres where AI models are trained and run. Marvell said the combination would expand its total addressable market to $550 billion and generate more than $21 billion in annual revenue, according to Reuters. Both companies’ boards approved the transaction unanimously, and it is expected to close in the first half of 2028, subject to shareholder and regulatory approval.

Yet the market’s verdict on the day was blunt. Marvell shares fell 12.7 per cent and Lumentum shares fell 18.6 per cent, according to market data, while Micron dropped 5.8 per cent and Oracle fell 8.8 per cent. The S&P 500 slipped 0.5 per cent and the Nasdaq Composite fell 1.1 per cent, even as the Dow Jones Industrial Average rose 0.4 per cent to 51,989.19. Apple, reporting its first quarterly results under new chief executive John Ternus after the closing bell, beat Wall Street’s expectations on both sales and profit — and its shares still fell. Palantir, which reported revenue up 96 per cent, also fell in extended trading. It was, by any measure, a day when good numbers were not good enough.

The deal in detail

Under the terms of the agreement, Lumentum shareholders will receive 4.5 Marvell shares for each share they own, equivalent to about $1,132 per Lumentum share, according to Reuters. That represents a 23.9 per cent premium to Lumentum’s last unaffected closing price on 28 October, the day before Bloomberg News first reported that the two companies were in talks. The transaction gives Lumentum an equity value of $102.54 billion, with the $115 billion headline figure including debt. When the deal closes, Marvell shareholders will own about 67 per cent of the combined company and Lumentum shareholders about 33 per cent.

The price is steep by any conventional measure. The deal values Lumentum at about 33 times its projected core operating profitability over the next 12 months, more than double the median of about 16 times for industry peers, according to Reuters calculations based on LSEG data. It is also an equity value worth more than half of Marvell’s own market capitalisation, and more than ten times Marvell’s previous biggest purchase, the $10 billion acquisition of Inphi in 2021.

The agreement carries heavy break fees. Marvell would have to pay Lumentum $5.8 billion if its board withdraws support for the deal or walks away to accept a takeover offer for itself, according to a regulatory filing reported by Reuters. Lumentum would owe Marvell $2.9 billion in similar circumstances. Marvell would also pay Lumentum $3.2 billion if regulators block the transaction. Marvell said it expects the deal to close during the first half of 2028, but that it could extend to as late as August 2028 if regulatory approvals have not been obtained, according to Reuters Breakingviews. Qatalyst Partners is serving as the lead financial adviser to Marvell, while Citigroup and Evercore are advising Lumentum. Marvell chairman and chief executive Matt Murphy will continue to lead the combined company.

Investors had already been voting with their feet before the deal was formally announced. Since the Marvell-Lumentum talks were first reported late last month, Lumentum’s shares have fallen about 8 per cent and Marvell’s about 22 per cent, according to Reuters, as a broad-based sell-off in chip stocks accelerated following Intel’s weak forecast. Monday’s announcement did nothing to stop that slide; if anything, it gave it a focal point.

A brutal day for the buyers

The scale of Monday’s reaction is worth dwelling on, because it was not a protest against the idea of the deal so much as against its price and its timing. Marvell fell 12.7 per cent and Lumentum 18.6 per cent in a single session, according to market data. Lumentum’s fall was the crueller of the two: its shareholders are being paid in Marvell stock, so every fall in Marvell’s price cuts the value of what they will receive. The all-stock structure, chosen in part because it avoids the debt that has haunted other AI-era deals, leaves the sellers exposed to the buyer’s share price for the 18 months or more until the deal closes.

The damage spread well beyond the two deal partners. Micron fell 5.8 per cent, Oracle dropped 8.8 per cent after saying it plans to sell $10 billion of common stock as well as $10 billion of bonds to finance its AI buildout, according to Investor’s Business Daily. The Nasdaq Composite closed down 1.1 per cent at 25,164.32 and the S&P 500 fell 0.5 per cent to 7,541.40, while the Dow rose 0.4 per cent, lifted by consumer staples and other defensive names. Eight of the eleven S&P sectors finished higher, according to Stocktwits, but technology was the clear weak spot, with the technology sector fund falling 2.95 per cent and energy sliding 3.96 per cent as oil prices tumbled.

Oil was the other big mover of the day, and it told the same story of a market repricing risk. Brent crude settled on Monday at $104.77 a barrel, down 7.6 per cent, or $8.66, according to Reuters, as hopes of a diplomatic breakthrough between the United States and Iran grew. The fall in oil, like the fall in chip stocks, was a bet that the worst outcomes — a wider war in the Gulf, or a longer and costlier AI spending war — might be avoided. But for the AI trade, the message of the day was that investors are no longer willing to pay any price for growth, and that even a $115 billion deal struck at the top of the market can be greeted as a reason to sell.

The session capped a miserable stretch for technology shares. October was the Nasdaq’s worst month since the April 2025 “Liberation Day” rout, with the index falling almost 7 per cent and the Nasdaq-100 shedding more than $2.5 trillion in market value, according to Reuters. The final trading day of October had seen the Nasdaq fall 2.9 per cent, its worst day since June, while the Philadelphia semiconductor index dropped more than 7 per cent. Monday’s rebound at the open — the Dow rose more than 400 points in early trading — faded through the afternoon as the AI complex bled into the close.

Why Marvell wants the optics

The strategic logic of the deal is, on its own terms, easy to follow. The bottleneck in modern AI data centres is no longer only the chips that do the computing; it is moving data between thousands of those chips fast enough to keep them working. As AI models have grown, and as the agents and chatbots built on them have multiplied, the optical links connecting processors inside a data centre have become as important as the processors themselves. Lumentum makes the lasers and optical transceivers that carry that traffic. Marvell makes the custom chips and the networking equipment that sits around them. Put together, the two companies can sell a customer most of the plumbing of an AI data centre from a single supplier.

Marvell has been building towards this for some time. It closed a $3.5 billion purchase of Celestial AI in February, a deal aimed at the same problem: faster optical links between chips in AI data centres, according to Reuters. The Lumentum acquisition is the same bet, made at roughly thirty times the price. Marvell said the combination would expand its addressable market to $550 billion, a figure that reflects how much of the AI infrastructure budget now flows through connectivity rather than computation alone.

The deal also sharpens Marvell’s contest with Broadcom, its larger rival, which has long been Alphabet-owned Google’s main custom-chip partner and offers customers a broader menu of AI data centre hardware under one roof, according to Reuters. Marvell notched one of its biggest wins in August, signing a deal to supply chips that sit alongside Google’s AI processors, worth up to $120 billion in sales through fiscal 2033, with Google receiving the right to buy a $12.2 billion stake in Marvell. Adding Lumentum’s optics to that relationship gives Marvell more to sell to the same customers, and more reason for those customers to keep it close.

There is also a powerful backer in the background. Nvidia, the AI chip giant, invested $2 billion each in Marvell and Lumentum earlier this year, according to Reuters, pairing its Marvell stake with an artificial intelligence partnership. Jensen Huang, Nvidia’s chief executive, called Marvell the next “trillion-dollar company” in June, after which Marvell’s shares hit a record high. Just three years ago, Marvell was a relatively little-known chipmaker. Its market value has since surged to more than $170 billion, and its annual revenue has more than doubled from $3 billion in fiscal 2021 to $8.2 billion in fiscal 2026, according to Reuters. The Lumentum deal is an attempt to make that growth durable — to turn a company that sells parts of the AI machine into one that sells the machine’s nervous system.

Lumentum, for its part, brings a business that is smaller than Marvell’s — roughly half its size — but strategically placed. It competes with US-listed Coherent and China’s Zhongji Innolight in optical components, and it stands to benefit from a proposed US ban on new Chinese optical transceivers that could shift demand towards domestic suppliers, according to Reuters. That regulatory tailwind is real, but it is also double-edged: the same politics that might protect Lumentum from Chinese competition could also slow the deal itself, a point Reuters Breakingviews made in a note headlined “China adds latency to Marvell’s optical mega-deal”. Chinese approval is likely to be required, and Marvell has already allowed for the closing to slip to August 2028 if approvals are not obtained.

Apple’s first Ternus report: a beat that still fell

If the Marvell deal was the day’s biggest transaction, Apple provided its biggest earnings test. The iPhone maker reported after the market closed on Monday, in its first quarterly results under John Ternus, who took over as chief executive in September after Tim Cook stepped down. The numbers beat Wall Street’s expectations on both the top and bottom lines, according to Reuters. Fiscal fourth-quarter sales were $114.48 billion, above analyst estimates of $113.39 billion, according to LSEG data. Profit was $2.10 per share, above estimates of $1.99, and ahead even when excluding five cents per share from tariff refunds.

The detail was strong where it mattered most. IPhone sales, which included the new iPhone 18 Pro models but not the iPhone Duo, were $57.58 billion, above estimates of $56.54 billion, according to Reuters, rising 17 per cent year on year even though the quarter caught only eight days of iPhone 18 Pro sales. Services revenue reached an all-time high of $31.57 billion, in line with estimates. Mac revenue jumped 24 per cent to $10.79 billion, and revenue in Greater China grew 21 per cent to $17.54 billion, the fastest growth of any major region, according to CNBC — an answer, in hard numbers, to the quarter’s loudest worry about demand in China. Gross margins were 47.9 per cent, above estimates of 47.5 per cent, according to Reuters.

There were weak spots, and the market found them quickly. IPad revenue fell 10 per cent to $6.23 billion, the worst showing of any category, with Ternus blaming supply constraints and a tough comparison, according to CNBC. Wearables fell 8 per cent to $8.30 billion. And the margin story was mixed: Ternus told Reuters that Apple had seen higher memory costs compared with the June quarter, partly offset by inventory benefits and lower costs on some non-memory components, and that price increases on its products had provided some offset. With AI data centres absorbing the world’s supply of memory chips, component costs have soared across the industry, and Apple — which kept about 48 cents of every dollar it brought in after paying to make its products, down from about 50 cents the quarter before, according to Finimize — is not immune.

The new iPhone Duo, Apple’s foldable handset, also loomed over the report. The device did not reach shelves until after the quarter ended, and Ternus told Reuters that supply of the Duo is constrained and that “a very small percentage” of units had been returned for display issues, which Apple is replacing. Apple shares fell 1.8 per cent in regular trading on Monday and extended their losses after the results, as investors weighed a record iPhone quarter against shrinking iPad sales, rising memory costs and a holiday-quarter margin outlook that came in softer than hoped. It was, in miniature, the same judgement the market passed on Marvell: the business is performing, but the price of performing keeps rising.

Palantir’s record quarter, Oracle’s $20 billion ask

Palantir Technologies offered the day’s purest version of the same paradox. The data-analytics company reported third-quarter revenue of $2.313 billion, up 96 per cent year on year, according to market data, with US commercial revenue growing 148 per cent to $984 million and GAAP earnings per share of $0.44. The company also lifted its fourth-quarter revenue outlook. For the first time, Palantir’s commercial business passed its government business in size, according to Stocktwits — a milestone the company has been working towards for years, and proof that its AI software is finding buyers well beyond the defence and intelligence agencies that made its name.

Palantir’s shares still tumbled more than 3 per cent in extended trading on Monday, according to Investor’s Business Daily. The reason was not the quarter just reported but the valuation attached to it: after a run that made Palantir one of the market’s most richly valued stocks, even 96 per cent growth was not enough to hold the price. The same logic applied to Oracle, whose shares fell 8.8 per cent on Monday after the company said it plans to sell $10 billion of common stock as well as $10 billion of bonds to finance its AI buildout, according to Investor’s Business Daily. Oracle is raising $20 billion to fund data centres and computing capacity; the market’s response was to mark down the stock of a company that needs to keep raising money to keep growing.

The pattern is now familiar enough to name. The companies building the AI future are spending at a pace that frightens even their own shareholders. Goldman Sachs has estimated that hyperscaler capital spending will reach about $800 billion in 2026 and $1.2 trillion in 2027. Anthropic, the AI company behind the Claude chatbot, underlined the scale of the commitments in an amended IPO filing on Monday: third-quarter revenue of roughly $17 billion, up about 1,127 per cent year on year, but a net loss of $23.3 billion for the period, according to market reports, with compute commitments running past a trillion dollars. Anthropic is planning to take its stock public at a valuation as high as $2 trillion, according to Finimize. The revenue growth is extraordinary. The losses are extraordinary too. And the market, on Monday at least, was in no mood to look past the second fact to celebrate the first.

The wider AI spending question

Monday’s trading did not happen in a vacuum. It followed a month in which the AI trade, the engine of the stock market’s rise for three years, began to look like a liability. September’s inflation data, released during the session, showed headline prices rising 3.6 per cent and core prices 3.1 per cent, according to Stocktwits — numbers that keep alive the prospect of the Federal Reserve raising interest rates rather than cutting them, and that make the cost of financing the AI buildout higher still. The Fed is now seen as more likely to tighten in December than it was a month ago, and the yield on the ten-year Treasury has climbed back towards 5 per cent. Money, in short, is no longer cheap, and the companies spending the most of it are being asked harder questions about what it buys.

The questions are not abstract. The AI infrastructure boom is being financed, in large part, with debt — by the hyperscalers themselves, by the neocloud companies that rent out computing capacity, and by the chipmakers and their suppliers. Oracle’s $20 billion raise is only the latest example. In October, CoreWeave, the AI cloud provider, saw its shares crash 16 per cent in a single session as yields on its bonds hit 15 per cent and investors questioned whether its $52 billion of debt and lease obligations could be supported by its cash flows. The episode was a warning that the AI trade has a financing leg, and that the leg can wobble.

Nor is the spending slowing. Advanced Micro Devices, Shopify, Arista Networks and SharkNinja are all due to report this week, with AMD’s results on Tuesday night the most watched, according to Investor’s Business Daily. Each report will be read the same way Apple’s was: not for whether the numbers beat expectations, but for what they say about the cost of the next round of growth. The market’s new test is not growth but the price of growth — and on Monday, for Marvell, for Lumentum, for Apple and for Palantir alike, the price looked too high.

There is, of course, another reading of the day, and it is the one Marvell’s management will be making in the weeks ahead. The AI buildout is real: data centres are being built at a pace never seen before, the demand for optical links is outstripping supply, and the companies that control the connectivity between chips will be as important to the next decade of computing as the companies that make the chips. A $115 billion deal that looks expensive today may look cheap in five years if the AI boom continues on its current path. That is the bet Matt Murphy has made, with his shareholders’ money and with Lumentum’s. Monday’s market did not believe him. The next eighteen months, until the deal closes, will decide whether he was right.

Analysis: Why It Matters

The market judged the price, not the idea

The most important thing about Monday’s reaction is what it was not. Investors did not sell Marvell because they doubt that optical connectivity matters — nobody in the industry doubts that. They sold it because Marvell is paying 33 times Lumentum’s projected operating profit, more than double the peer median, for a company half its size, in stock, at a moment when its own shares have already fallen 22 per cent since the talks leaked. A deal struck at the top of a market, paid for with falling currency, is a deal that asks shareholders to trust the future twice over: once that the AI boom continues, and once that the combined company can earn back a premium that leaves no room for disappointment. The market, on Monday, declined to extend that trust. That is a judgement about valuation, not about technology — and it is a judgement that will follow every large AI deal from now on.

Optics is the new bottleneck

The strategic case for the deal deserves to be taken seriously, because it points to where the AI industry’s next constraint lies. For years the story was about computing power: who had the fastest chips, the most of them, the best software to run them. That story is now, in part, a story about plumbing. A single AI training run can involve tens of thousands of chips working together, and the speed at which they exchange data determines how much of their computing power is actually used. Optical links — lasers and transceivers moving data as light — are the fastest way to connect them, and demand for those links is growing faster than supply. Marvell’s purchase of Celestial AI in February and now Lumentum is a recognition that the company which controls the connections between chips controls a choke point of the AI economy. If the bet is right, the price will be forgotten. If it is wrong, the price is all anyone will remember.

The Broadcom shadow

Marvell’s deal cannot be understood without Broadcom, the larger rival that has spent a decade building exactly the kind of one-stop AI hardware shop Marvell now wants to become. Broadcom has long been Google’s main custom-chip partner, and it offers customers a broad menu of AI data centre hardware under one roof. Marvell’s August deal with Google, worth up to $120 billion in sales through fiscal 2033, was its answer — and Lumentum is the next step, adding optics to a portfolio that already covers custom chips and networking. The logic is that customers prefer fewer suppliers, and that a company able to sell the chip, the network and the optics together will win more of the hyperscalers’ budgets than a company selling only one of the three. It is a sound theory. It is also a theory that Broadcom has already proved, at a scale Marvell has yet to reach, and the premium Marvell is paying reflects how far behind it believes it is.

The AI trade is splitting in two

Monday’s session made visible a split that has been forming for weeks. On one side are the companies whose AI spending is being rewarded — the hyperscalers with the cash flows to fund it, and the chipmakers whose order books are full. On the other are the companies being asked to pay for it: the suppliers raising debt, the software firms whose valuations assume endless growth, and now the acquirers paying top-of-market prices. Apple beat expectations and fell. Palantir grew 96 per cent and fell. Oracle announced a $20 billion raise and fell 9 per cent. Anthropic reported revenue up more than 1,100 per cent and a $23.3 billion quarterly loss in the same filing. The market is no longer treating “AI” as a single trade that lifts everything it touches. It is sorting the companies that can fund the buildout from the cash they generate from the companies that must borrow, dilute or overpay to stay in it. Marvell, after Monday, sits squarely in the second group — and that is a harder place to be.

The China clock

The deal’s longest shadow may be the one cast by Beijing. Reuters Breakingviews has already flagged that China “adds latency” to the transaction: Chinese regulatory approval is likely to be required, and Marvell has allowed for the closing to slip to August 2028 if approvals are not obtained. That is up to 21 months of waiting, during which Lumentum’s shareholders are paid in Marvell stock, Marvell’s management is distracted, and the AI market — which moves in quarters, not years — can change beyond recognition. There is also a political irony at work: Lumentum stands to benefit from a proposed US ban on new Chinese optical transceivers, which could shift demand to domestic suppliers, but the same great-power rivalry that creates that tailwind also gives Beijing leverage over the deal’s approval. A transaction designed to strengthen American AI infrastructure must now survive the geopolitics of the technology it depends on. That, more than the share price on any single day, is the risk that will determine whether this $115 billion bet pays off.

What to Watch Next

1. AMD’s results on Tuesday night: Advanced Micro Devices reports after the market closes on Tuesday, and its numbers — and its outlook for AI chip demand — will be read as the next verdict on the AI trade after Marvell’s deal and Apple’s earnings. Shopify, Arista Networks and SharkNinja also report this week.
2. Tuesday’s market session and the midterm results: Americans vote in the midterm elections on Tuesday, with polls suggesting the Democrats will retake the House. The market’s reaction on Wednesday, when the results are known, will show whether Monday’s AI sell-off was a one-day wobble or the start of a deeper repricing.
3. The Marvell and Lumentum shareholder votes and analyst calls: Both companies’ managements must now sell the deal to shareholders who watched both stocks fall on the announcement. Analyst downgrades, price-target cuts or a widening of the deal spread would signal that the market’s scepticism is hardening.
4. Oracle’s $20 billion raise: How the market absorbs Oracle’s $10 billion of new stock and $10 billion of new bonds will be a live test of the AI financing question. A weak reception would confirm that investors are losing patience with debt-funded AI spending; a strong one would suggest the appetite is still there.
5. China’s regulatory signals and the transceiver ban: Any sign of how Beijing will treat the Marvell-Lumentum deal — and of how Washington’s proposed ban on new Chinese optical transceivers develops — will shape both the deal’s timetable and Lumentum’s competitive position in the meantime.

Sources

• Reuters — Marvell to buy Lumentum in $115 billion deal in AI data-center bet
• Reuters — Apple results beat Wall Street expectations on strong iPhone sales
• Reuters Breakingviews — China adds latency to Marvell’s optical mega-deal
• Investor’s Business Daily — Dow Jones Futures: AI Trade Dives As Marvell, Lumentum Plunge; Apple Tumbles On Earnings
• Finimize — Your Daily Brief For November 3rd