A Global AI Sell-Off Rocks Markets, the Iran Truce Frays, and Micron Posts a Record AI-Memory Quarter

The year's relentless AI rally finally cracked. On Tuesday, June 23, technology and chip stocks sold off around the world, with the Nasdaq falling 2.21% and the S&P 500 down 1.44%, while the Dow, lighter on tech, barely moved. The damage was worst in Asia: South Korea's Kospi plunged about 10%, tripping a circuit breaker that forced a 20-minute halt, as SK Hynix and Samsung Electronics, two of the world's leading memory makers, each fell more than 12%. Nvidia, Micron, and AMD led the retreat in the US.

There was no single trigger, which is part of what unnerved investors. The selling reflected a cocktail of fears: that the stunning surge in AI stocks had simply run too far, that the inflation stoked by the Iran war and the closure of the Strait of Hormuz could force interest rates higher, and that the Federal Reserve, which had turned hawkish a week earlier, was serious about keeping policy tight. Higher rates matter doubly for AI, because they raise the cost of the enormous debt now funding the data-center build-out, a vulnerability we explored in our look at how the AI boom is financed.

Then the data poured fuel on the fire. On June 25, the May PCE index, the Fed's preferred inflation gauge, came in at 4.1% year over year, the highest since April 2023, with the core reading at 3.4%. That kept a September rate hike firmly in play and gave the bears more ammunition. The slide never really reversed: the Nasdaq fell in every session and ended the week down about 4.6%, more than 6% below its early-June record, with Friday's losses finally shrinking to near zero. A late-week New York Times report that OpenAI was weighing pushing its own IPO into next year, amid the turmoil in AI shares, only deepened the gloom.

The episode is a vivid reminder that the AI trade and the macro picture are now joined at the hip. The same higher-for-longer rates the Fed is signaling lift the cost of the trillions in borrowing behind AI infrastructure, so an inflation surprise hits AI names twice over, once through their valuations and again through their financing. After a rally that had looked unstoppable, it was the sharpest setback yet, and a real test of whether the market's conviction in AI can survive a less friendly rate environment.

The Versailles truce frayed almost as fast as it was signed. On June 25, the Ever Lovely, a Singapore-flagged Evergreen container ship, was struck off the coast of Oman by a drone flown by Iran's Revolutionary Guard, the first attack on a cargo vessel since the ceasefire took effect. The ship was damaged but no one was hurt, and the strike came just after Oman and the International Maritime Organization had opened a new transit route through the Strait of Hormuz.

The United States responded within a day. On June 26, United States Central Command said it had delivered a "powerful response," with American aircraft striking Iranian missile and drone storage sites and coastal radar installations. President Donald Trump blamed Iran for firing "at least four" attack drones at ships passing through the strait, claiming three were knocked down and the fourth hit its target. The IMO promptly paused its plan to shepherd vessels through the waterway.

The messaging stayed contradictory. Iran's Revolutionary Guard warned against the new transit route and threatened to close the strait altogether over Israeli strikes in Lebanon, even as Washington insisted the waterway was open. Hormuz, through which a large share of the world's seaborne oil passes, has been largely blocked since the war began on February 28, and shipping had only just started to trickle back.

For markets, the timing could hardly have been worse. Oil and freight had begun to recover in the week after the truce, but the attack showed how fragile that rebound is, and it landed just as the May PCE print confirmed inflation at a three-year high. A chokepoint that can reopen and reclose on a single drone strike is precisely the kind of supply shock that keeps the Fed hawkish, raises the cost of financing energy and infrastructure deals, and, as the week's sell-off showed, ripples straight into the AI trade. Pricing that geopolitical risk is, once again, part of the job.

The day after the AI-fear sell-off, Micron answered the doubters. On June 24, the memory maker reported record fiscal third-quarter revenue of about $41.5 billion, up roughly 346% from a year earlier and well above its own $33.5 billion guidance, with a record gross margin near 85%. It then guided to roughly $50 billion for the current quarter, a sign it expects the AI-memory boom to keep accelerating. The driver is high-bandwidth memory (HBM), the specialized chips that feed AI accelerators.

The standout detail was scarcity. Micron's HBM is effectively sold out through 2026, and its next-generation HBM4 has begun shipping for Nvidia's Vera Rubin platform, ramping at roughly twice the pace of the prior generation. Memory, long treated as a boom-and-bust commodity that swung with PC and phone cycles, has become one of the most prized and supply-constrained links in the entire AI chain.

The contrast with the previous day was almost comic. Where the sell-off priced in fear that AI demand might finally disappoint, Micron's numbers said the opposite: demand for AI hardware is, if anything, outrunning the industry's ability to supply it.

That tension is the real story for investors and bankers. A sold-out order book and record margins are wonderful until capacity catches up, and memory has burned investors with its cyclicality before. The question hanging over Micron, and its Korean rivals SK Hynix and Samsung Electronics, is how long the shortage and the fat margins can last before the industry over-builds, as it always eventually has.

For now, though, Micron's blowout is a reminder that the picks-and-shovels of the AI era can be every bit as lucrative as the models themselves. The companies selling the memory, the networking, and the power into the build-out are capturing real, cash-generative demand today, even as the AI labs at the center of the story keep losing billions. In a week when the market wobbled on doubts about AI, the clearest profits came from the unglamorous business of making the chips that make AI run.

On June 23, Energy Fuels, a US producer of uranium and rare-earth materials, agreed to acquire Vacuumschmelze (VAC), a German maker of magnetic materials and rare-earth permanent magnets, from the private equity firm Ara Partners for about $1.9 billion in equity value, based on Energy Fuels' closing share price the prior day.

The strategic logic runs through one of the most politically charged supply chains in the world. Rare-earth permanent magnets are essential to electric vehicles, wind turbines, defense systems, and electronics, and China dominates their processing. Western governments and companies have been scrambling to build an alternative, and buying VAC gives Energy Fuels an integrated platform spanning mining, processing, and finished magnets, a rare end-to-end Western champion in a market the West has spent years trying to onshore.

The structure is instructive too. This is a cross-border transaction, a North American-listed buyer acquiring a German target, and it is a private equity exit, with Ara Partners realizing its investment. Critical-minerals deals like this carry complications that ordinary M&A does not, from national-security and regulatory review on both sides to currency exposure and the challenge of integrating operations across jurisdictions, the kind of issues our guide to cross-border M&A lays out.

The deal is a sign of where a growing slice of M&A is headed. Critical minerals have become a strategic battleground where transactions are driven as much by supply-chain security and government policy, including subsidies and protective measures, as by financial return. As Western states race to reduce dependence on China for the materials that power electrification, defense, and AI hardware, expect more resource and critical-minerals deals where geopolitics and dealmaking meet. For bankers, it is a niche that barely existed a few years ago and is now one of the more interesting corners of the market.

On June 24, Qualcomm, the company best known for the chips inside smartphones, agreed to acquire Modular, an AI software startup, in an all-stock deal valued at more than $3.9 billion. Qualcomm will issue up to 19.2 million of its shares to Modular's owners, and the companies expect to close in the second half of 2026.

The target is unusual, and so is the prize. Modular builds a 'silicon-agnostic' software layer that lets AI models run across chips from different vendors, Nvidia, AMD, and others, without developers having to rewrite their code for each one. Qualcomm framed the goal as delivering a 'compute layer across devices, edge and data centers,' part of its push beyond phones and into the AI data center.

What makes the deal strategically pointed is what it targets: Nvidia's software moat. Nvidia's dominance rests not only on its hardware but on CUDA, the programming platform that has locked a generation of developers into its chips. A vendor-neutral layer like Modular's is an attempt to loosen that grip, and buying it signals that Qualcomm and others intend to attack Nvidia's ecosystem, not just compete on silicon. It is a reminder that the AI battle is being fought in software as much as in chips.

The structure fits the week's pattern, too. This is another large all-stock technology acquisition, with the acquirer spending its richly valued shares rather than cash, and it lands even as AI stocks wobbled in the same week's sell-off. The scramble for AI capability keeps spilling out of funding rounds and into M&A, and the price of a credible foothold, here close to $4 billion for a pre-profit software startup, keeps climbing. For dealmakers, the AI land grab is now as much an M&A story as a venture-capital one.

Old economy met hot market this week. On June 24, Doncasters, a nearly 250-year-old British manufacturer of precision aerospace components such as turbine blades, vanes, and castings, priced its US initial public offering at $33 a share, above its $28 to $32 marketed range, raising about $919 million by selling 27.9 million shares. Listing on the New York Stock Exchange under the ticker DPC, the stock surged about 33% on its first day to close near $44.

The demand reflects a clear thesis. Investors are betting on rising commercial aircraft production, a wave of increased defense spending, and the energy infrastructure, including the industrial gas turbines, needed to power the AI data-center build-out. A centuries-old turbine-parts maker has, in effect, been repackaged as an aerospace, defense, and AI-power supply-chain play, and buyers loved it.

The context is just as telling as the deal. Doncasters priced in one of the busiest IPO weeks of the year, with around a dozen companies pricing on June 24 alone, and it underlines that the 2026 IPO window has reopened well beyond the headline technology and space names. Aerospace, industrials, and a range of other sectors are now finding eager public buyers, the breadth that distinguishes a genuine IPO market from a one-rocket wonder.

For equity capital markets desks, an oversubscribed, above-range pricing followed by a 33% pop is close to an ideal outcome, evidence that public investors will pay up for quality across the board, not just for AI. For a market that was effectively frozen two years ago, a 250-year-old maker of turbine blades popping on its debut is a vivid measure of how far risk appetite has swung back. The pipeline behind it just got a more encouraging read on what the market will pay.