enesptjazh-Hanszh-Hantfrdeitkoruarhiidtrnlplvithsvhemsukcselrodafinobn

Solomon says AI capital got pricier, not scarce, as Marvell pays $115 billion in stock

David Solomon says AI money is there at a higher price. Dan Niles lists pulled listings, and Marvell bid $115 billion for Lumentum in shares.

By Published November 3, 2026 7 min read Grounded in the CEOInterviews.AI corpus

On Monday, November 2, the first trading day after a Friday selloff that hit chip stocks hard, Matt Murphy announced that Marvell would buy Lumentum for an implied $115 billion, paid entirely in Marvell stock. The same morning in Hong Kong, Goldman Sachs CEO David Solomon told Bloomberg that the money for AI has not run out. It has only gotten more expensive. And on a podcast taped that day, fund manager Dan Niles went through, deal by deal, the AI financings that have stalled since September. All three are talking about the same thing. The question used to be whether anyone would fund the buildout. Now it is who can still afford the price.

Marvell Technology's operating headquarters in Santa Clara, California, seen from highway 237
Marvell's Santa Clara headquarters, May 2011. Photo by King of Hearts, CC BY-SA 3.0, via Wikimedia Commons.

Marvell announced into a selloff on purpose

The terms are in Marvell's release. Lumentum holders get 4.5 Marvell shares each, a 23.9% premium to Lumentum's October 28 close. They would own about 33% of the combined company, which expects more than $1 billion a year of cost savings and a close in the first half of 2028. The pitch is scale-up networking. Clusters now link thousands of GPUs, and Murphy argues the chips and the lasers that carry data between them should come from one supplier.

On CNBC, co-host David passed along a question from shareholders: is Lumentum's laser business a commodity that only looks valuable because it is sold out? Murphy's answer:

The reason that they're sold out is because the demand is exploding and we don't see that changing anytime soon. But through cycle, the differentiated and proprietary nature of these products is going to be lasting whether we're in a supply shortage or not.

Matt Murphy, November 2, 2026, CNBC Squawk on the Street

Later on Fox Business he was asked whether he had thought about waiting a week. He said the combined company was running at about $21 billion a year with operating margins over 40%, then gave the answer that tells you how he thinks about the stock:

So the idea is that you cannot time this stuff, and there's no way to time it perfectly. You just have to do it.

Matt Murphy, November 2, 2026, Fox Business

The market had its own answer. Marvell closed down 12.72% at $171.41 and Lumentum fell 18.59% to $684.40. At that close, 4.5 Marvell shares were worth about $771 against the roughly $1,132 the deal implied when it was priced. Lumentum traded below even the $771. Paying in stock means Marvell takes on no new debt. It does not make the deal free. The cost shows up as dilution and as a spread that says the market doubts a 2028 close.

Solomon's line between cost and availability

Solomon spoke at the HKMA's Global Financial Leaders' Investment Summit. He listed last week's big financings for SpaceX and Meta and the coming Anthropic IPO, and he said investors are still buying. The important part was a distinction most of the AI bear case runs together.

But I don't think it affects availability. I think it affects cost. And if cost goes up too much, it changes, you know, the returns in some of these projects.

David Solomon, November 2, 2026, Bloomberg Daybreak: Asia

His support for the availability half was the size of the pool. "Public equity market cap in the world is about 160 trillion," he said. I believe him about the pool. But the half he passed over quickly is the one that matters this month. A bank that earns fees on every one of these deals has good reason to say the money exists. Whether it exists at a rate that makes a given data center pay is a different question, and he conceded the point in the same breath. On debt investors, he said: "if they think the market's getting a little bit more challenging, they ask for higher returns."

Goldman Sachs CEO David Solomon at Goldman Sachs headquarters in New York in 2019
David Solomon at Goldman Sachs in New York, November 2019. U.S. Department of Defense photo by Lisa Ferdinando, CC BY 2.0, via Wikimedia Commons.

Pick sizes the bill and admits some of it fails

Morgan Stanley's Ted Pick came off the same stage and gave Bloomberg the bigger number. He expects $10 to $15 trillion of AI capex, deployed over seven to twelve years. He was blunter than Solomon about the misses.

But I think the kind of ranges we're seeing right now are about a trillion a year for the next several years, which gets us thinking we're somewhere in the second or third inning.

Ted Pick, November 2, 2026, Bloomberg Talks

Asked whether projects will all work, he said "the answer, of course, is no." Then he described how the losers get financed, which is the useful part for anyone underwriting this.

But this is an important piece of this: sometimes the technology is going to be faster than the project can possibly sustain, and sometimes the project is going to be building something that people are not yet home to receive, and some capital structures will be better than others.

Ted Pick, November 2, 2026, Bloomberg Talks

That is Solomon's argument in different words. Money stays available, but local opposition and completion risk each add a premium, and some projects will not clear it.

Niles counts the deals that did not get done

Niles, who runs the Satori Fund, spoke with Wilfred Frost on The Master Investor Podcast. He has the opposite read, and it rests on the same facts, seen from lower down the credit curve.

NScale, another company backed by Nvidia, that deal has gotten shoved out. AirTrunk, which is backed by, I believe, Blackstone Canada pension fund, that deal got shoved out.

Dan Niles, November 2, 2026, The Master Investor Podcast

His list holds up. Reuters reported on October 22 that Switch and Nscale had pushed back their listings, that Firmus and the AirTrunk REIT had pulled or paused theirs, and that Anthropic was the only large tech IPO still planned for the fall. Niles also said CoreWeave's bonds trade around 15%, which I have not checked. His bigger worry is crowding, with Meta, Oracle and OpenAI all raising money before Anthropic prices.

So all of these guys are competing for the same pool of money, and it's the same group of investors, and they're draining a lot of that capital, and investors are getting a little nervous about having all their eggs in one basket, which is the AI trade.

Dan Niles, November 2, 2026, The Master Investor Podcast

Solomon says cost, Niles says scarcity. For a hyperscaler with a free cash flow buffer, I think Solomon is right. For a neocloud listing in Singapore, the difference is academic. A price you cannot pay works the same as no money. Niles calls the neoclouds "toast" even if they rally, and he still expects enterprise agents to keep demand growing. So his view is not a call on the top. It says the people who get funded are about to change.

Why a selloff can mean more mergers

That brings it back to Marvell. John Chambers, who ran Cisco through the dot-com bust, told Yahoo Finance's Brian Sozzi he would "never bet against Nvidia" and then predicted a wave of deals that does not depend on the tape.

And if you fall behind, you can't catch up through organic movement. You're going to have to be creative.

John Chambers, November 2, 2026, Sozzi Unleashed

Solomon said much the same from the banker's chair, that "everybody recognizes the importance of scale in almost any industry" and that regulators are now allowing deals at that size. Murphy expects rivals to answer. He told Fox the next deal would be a number two or three combining with a number three or four. The logic follows. If outside money costs more, the companies with richly valued stock use it as currency and buy what the market has marked down. Marvell's $115 billion is the first test of that this cycle.

So who still gets funded?

The scaled companies do. Solomon and Pick are describing the deals their banks lead, for hyperscalers, SpaceX, Meta and Anthropic, and those are still getting done. Niles is describing everyone one rung down, and his evidence is a list of postponed filings. Both are telling the truth about their part of the market. The tell to watch is Anthropic's pricing. If it clears near the range, Solomon's cost-not-availability line holds for at least another quarter. If it prices well below, Niles's crowding argument reaches the top tier too, and a stock-funded deal like Marvell's starts to look like the cheapest money left.