The AI selloff has stopped rewarding every company with a chip ticker and a story about compute. If you are raising money on an AI premium, the public market just gave you a cleaner, colder benchmark.
The week ending June 26 was ugly because it was specific. This was not investors selling technology in one blind sweep. It was investors deciding which AI names still deserve the prices they were handed during the boom, and which ones were living off the glow from everyone else.
According to The Wall Street Journal, the PHLX Semiconductor Index fell 5.3% for the week, its sharpest weekly drop since April 2025. AP reported that SoftBank slid 12.5% in Tokyo after speculation that OpenAI may push its IPO into 2027, while Micron dropped 6.7% on Friday even after the company had just delivered the kind of earnings growth that would usually calm the market. ON Semiconductor fell 23.7% after announcing a $7 billion all-stock deal for Synaptics. Sandisk, Seagate and Western Digital were hit too. When memory, chips, AI backers and acquisition stories all get marked down together, you should pay attention.
The blunt point is this: demand for AI infrastructure has not vanished. Micron is still selling into a memory shortage created by data centers, and MarketWatch noted that its quarterly sales were more than four times higher than a year earlier. But investors are no longer treating that demand as permission to ignore price. They are asking whether the next dollar of AI spending turns into earnings, cash flow, or just another quarter of higher capital costs.
That is a different market.
The Broadcom wobble earlier in June helped set the mood, but the bigger signal came from the way good news stopped being enough. Micron's results were strong. They still did not prevent the stock from falling into the end of the week. If you need a cleaner sign that the AI trade has moved from belief to inspection, there it is. The market is not rejecting AI. It is rejecting the idea that every AI exposure deserves the same multiple.
You can see that in the shift toward companies with clearer cash flows. Goldman Sachs analysts have argued, as Business Insider reported in May, that the next stage of the AI trade may favor hyperscalers such as Amazon and Microsoft over the chipmakers that dominated the first leg of the rally. That does not mean chipmakers are finished. It means investors want to see who captures the value after the servers are bought, powered and filled with models.
The public market is now auditing the AI story
OpenAI is the name that makes this feel less like a normal chip correction. Investors.com reported that OpenAI is considering delaying its IPO until 2027 after filing confidentially with the SEC in early June. The same report pointed to SpaceX's volatile debut, including a slide from a peak above $225 to roughly $153, as one reason advisers were urging patience.
That matters for founders because OpenAI has become the private market's loudest proof point. If even that company has to think carefully about public timing, a late-stage startup with weaker revenue quality should not assume the old AI premium is still waiting at the door. Frankly, it probably isn't.
The private market has not cracked yet. Crunchbase data, cited by the Times of India, put global venture funding at $300 billion in the first quarter of 2026, the highest quarterly total on record. AI deals carried much of that total. Seed-stage AI companies are still getting valuation premiums because investors do not want to miss the next model company, infrastructure layer or workflow tool that actually sticks.
But public multiples travel back into private rounds. They always do, just with a delay. A founder who raised a Series C at a revenue multiple justified by chip stocks trading at euphoric public valuations may find that the next term sheet uses a different comparison set. If public investors decide 40 times forward earnings is rich enough for exposed AI names, private investors will not keep underwriting 2025-style prices forever.
The fundraising line that worked last year was simple: we are an AI company, so we deserve the premium. Do not bother leading with that now. You need to show which budget you replace, which customer is paying, what gross margin looks like after compute costs, and whether usage grows without discounts doing all the work. The market is asking for receipts.
This is healthier than the market it replaces. The past 18 months rewarded founders who could attach themselves to the AI build-out. The next year should reward the ones who can prove they are part of it. That is harder, but it is also clearer. The week of June 26 did not end the AI trade. It made the easy version of it look stale.
Also read: Regulators are finally building the AI they need to police the markets they oversee • Trump's 100% tariff threat over digital services taxes reshapes the cost calculus for every US tech company operating in Europe • Framework Ventures raises $400 million to chase AI and robotics beyond its crypto roots
