AI Market Trends: Q3 2026 Review and Forecast

AI stocks carried the market in Q3 — but narrowly

In mid-July, China’s Moonshot AI released Kimi K3, a new AI model that revived fears that the huge sums US companies are spending on data centers may not pay off if powerful AI becomes cheap.

Chip and AI stocks sold off, while the war with Iran and fears of higher rates added to the pressure.

Second quarter earnings at the end of July deepened those worries.

Top cloud providers increased their capex budgets, leading to reduced free cashflow for Microsoft (NASDAQ:MSFT), Meta Platforms (NASDAQ:META), Amazon (NASDAQ:AMZN) and Oracle (NYSE:ORCL).

August brought a partial recovery. Results from Palantir (NASDAQ:PLTR) and NVIDIA (NASDAQ:NVDA) eased worries that AI spending isn't turning into revenue, and Salesforce's (NYSE:CRM) quarterly report sparked a rally in software stocks that had been sold off on fears AI would replace their products. By September, Meta’s launch of Muse, a personal AI agent that can book travel and make purchases for its users, helped draw buyers back into tech stocks, and as the month drew to a close the Nasdaq Composite closed at a record high for the first time since June.

However, gains rested on a handful of very large companies. Beneath the record highs, most stocks struggled; by the end of September, the majority of S&P 500 (INDEXSP:.INX) companies were trading well below their peaks, and the Nasdaq’s research team said major indexes were “mask(ing) meaningful correction for majority of stocks.”

The team also called rising treasury yields, which climbed through August and September and remain near their highs, the most important economic development of the quarter, since higher borrowing costs tend to weigh the hardest on smaller companies and those that depend on outside funding.

New listings followed a similar path. Renaissance Capital said Q3 initial public offering (IPO) activity “came in below expectations,” with SK Hynix's (KRX:000660,NASDAQ:SKHY) US listing in July accounting for the vast majority of Q3’s US$32.8 billion in proceeds. Brookfield-backed data center company Csquare (NYSE:CSQR) also came to market that month, while a handful of companies either postponed or withdrew their offerings.

Most notably within the sector, OpenAI said it plans to wait until 2027 to list. Outside the AI segment, smart ring maker Oura pulled its US listing on September 29 only hours before pricing.

Power is another issue. TD Economics wrote in July that “the most important bottleneck for the AI buildout is difficulty connecting data centers to energy infrastructure,” citing local approval as “another constraint on the AI buildout.”

Data centers have become an issue in US midterm campaigns as voters worry about higher power bills. Both major parties now say data centers should pay for their own grid upgrades, and a bipartisan group of senators introduced a bill on September 30 aimed at shifting more of those costs onto data centers.

Paying for AI buildout is the central question

Heading into Q4, the biggest open question for the AI buildout is whether it can be paid for. AI-related companies are now among the biggest issuers of US corporate bonds, according to Goldman Sachs (NYSE:GS).

In early October, the Financial Times reported that OpenAI told investors its annualized revenue was approaching US$50 billion, below figures that had circulated weeks earlier. CoreWeave (NASDAQ:CRWV), which rents computing power to OpenAI and other AI companies and has borrowed heavily against those contracts, saw its shares fall on the news. A week later, its bonds dropped to prices that signal investors are worried about being repaid.

On October 13, the International Monetary Fund's Global Financial Stability Report named the first risk to financial stability as “an AI-related market correction if expectations on investment or profits disappoint.” It also states that private credit funds face “growing liquidity pressures, which could restrict lending.”

Borrowing could get more expensive still. After September US inflation data came in higher than forecast, futures markets and most major banks now expect the Fed to raise rates again by December.

Investors are still putting money into AI infrastructure, but they have become pickier about where it goes.

“The conversation has shifted from growth at any cost to a much sharper focus on quality,” Harry Evans, a partner at law firm Freshfields, told Reuters in mid-October.

Are AI companies in the green yet?

Whether that money keeps flowing depends partly on whether AI starts earning it back.

Mark Rohrwasser, who leads the CXO Council at venture firm CI.VC, told the Investing News Network that the biggest AI labs and cloud companies are already seeing results.

“If you’re in the frontier labs or the hyperscalers, they’re making massive investments and they’re seeing kind of massive returns, almost unprecedented returns in some of the frontier labs and how quickly they’re growing,” he said, adding that ordinary companies are adopting AI more slowly, as they did with the internet and the cloud.

McKinsey’s State of AI survey backs up that slower pace. It shows that more companies say AI is spreading across their business, but its effect on earnings has not changed in a year.

“But I think most people at least are convinced that the return is there,” said Rohrwasser.

Lu Zhang, founder and managing partner of Fusion Fund, an early stage investor in enterprise and industrial AI, said that while it's still too early to see returns, Big Tech can’t afford to wait.

“They have to invest, regardless (of whether) they like it or not. That’s the reality," she said. Zhang expects new ways to pay for it: “We’re going to see more creative approaches of using different financial vehicles (to) try to fund the AI infrastructure, and also you will see more companies jumping into investment in AI infrastructure.”

Speaking about the banks, insurers and other financial companies she works with, she noted, “They also have a huge capital allocated as an AI budget, and the majority of the budget will be focused on buying GPUs,”

If returns stay uneven, the question is whether investors will keep paying at higher rates.

“I feel the market is losing patience a little bit more,” Zhang added.

Rohrwasser said companies will drop AI projects that no longer earn enough to cover the higher cost of borrowing, but put today’s rates in context: “Yes, interest rates are higher, but they’re also not historically high, right? You know, we’ve all gotten used to the last two decades of very, very low interest rates.”

The people closest to the buildout say rates alone won’t stop it.

Chris Miglino, CEO of Axe Compute (NASDAQ:AGPU), which owns and rents out AI computing equipment, said higher rates don’t change the economics of his company’s long-term contracts.

“Obviously, it’s more money, but it doesn’t change the whole thing. It’s not something that is catastrophic," he said.

Chip industry execs sound similarly calm. At a conference in Singapore on October 12, Broadcom (NASDAQ:AVGO) CEO Hock Tan said the cost of capital isn’t a problem as long as the biggest AI labs keep growing their revenue.

Anthropic, one of the biggest of those labs, gave an early read on that growth. On October 14, it told potential investors that its third quarter revenue rose sharply from the previous quarter, though more slowly than before, and that it turned a profit on an adjusted basis for the second quarter in a row, according to several news reports.

That slowdown has some potential investors discussing a lower valuation for the IPO than they had expected, Reuters reported, citing people familiar with the matter. Anthropic hasn’t published the figures or said what its adjusted measure leaves out, and the data hasn't been audited. The firm could start marketing its IPO as soon as November.

Some Wall Street analysts expect spending growth to slow anyway.

“We expect the market will start thinking about a potential peak year in 2027 soon, likely after 3Q EPS and midterms,” Ohsung Kwon of Wells Fargo (NYSE:WFC) wrote to clients in October.

The midterms matter for the buildout because of power, the other AI limit. Voters worried about rising power bills have made data centers a campaign issue, and the fight is now over who will pay for grid upgrades.

Several states have paused approvals or started reviews of how data centers affect local communities.

In Texas, Governor Greg Abbott suspended state approvals for new data center grid connections in August, citing grid reliability. Meanwhile, an industry coalition including Blackstone (NYSE:BX), OpenAI and SoftBank (TSE:9984,OTCPL:SFBQF) is campaigning against state bans. The pressure is pushing companies to find their own power, from natural gas turbines on site to small computing sites next to solar and wind farms.

Bruce Kahn, a fund manager who invests in power and grid equipment companies, said in August that demand for computing is already “well beyond what we can power and hydrate or cool,” and that more efficient chips won’t change that, since the problem is getting power from utilities to data centers in the first place.

“It’s not the chips. It’s the wires," he explained.

Miglino sees it differently. He said the bottleneck is more about permission than supply: “I don’t necessarily think it’s a power problem. I think it’s a regulation problem. There’s plenty of natural gas out there to run all of this stuff.” He expects building to pick up after the midterms, once states finish their reviews.

Investor takeaway

For investors, the next few months could show which of the two limits on AI — power or money — will weigh more on the buildout, and which companies can keep growing either way.

Heading into the fourth quarter of the year, the largest cloud companies are preparing to release their latest results, market participants await the Fed’s next rate decision and a handful of AI IPOs are still in the pipeline, including Nscale and SB Energy, Agility Robotics’ US listing and Anthropic.

How those listings are received could be an early test of how much investors are still willing to pay for AI growth.