Last updated: May 2026
Quick answers
Why are companies pulling back from AI in 2026?
Companies are pulling back from AI because most pilots aren’t producing measurable ROI, infrastructure costs are crushing margins, regulatory and reputational risk is climbing, and a populist backlash over energy costs is forcing public-facing rollbacks. Gartner’s 2026 data shows 80% of AI projects fail to deliver business value, while companies that cut staff to fund AI saw identical financial returns to those that didn’t.
Is the AI bubble bursting in 2026?
Not exactly. Global AI capex is still rising, with Goldman Sachs projecting $527 billion in 2026 alone. But the gap between spending and measurable returns is widening. Morgan Stanley found only 21% of S&P 500 companies can point to a concrete AI benefit. What’s happening looks more like a correction than a collapse: money is shifting from broad experimentation to narrow, proven use cases.
Which big companies are pulling back from AI?
Salesforce cut staff from its Agentforce AI team while four senior AI executives departed within three months. Fidelity cut 800-1,000 jobs during a tech restructuring, explicitly stating AI wasn’t part of the equation. Meanwhile, community opposition has blocked or delayed $98 billion in data center projects across the U.S., forcing companies like Meta and Microsoft to reconsider expansion timelines.
Why are companies pulling back from AI?
In January 2026, Salesforce CEO Marc Benioff stood on stage at Dreamforce and called Agentforce the company’s most important product launch in years. Three months later, four senior executives tied to the AI initiative had left the company. Then came the layoffs: nearly 1,000 roles cut, with the Agentforce team among those affected.
Companies are pulling back from AI in 2026 because pilot programs aren’t converting into measurable returns, infrastructure costs are ballooning faster than productivity gains, and a growing public backlash over energy consumption is making AI expansion politically risky.
Salesforce’s situation isn’t an outlier. It’s a pattern. Across industries, companies that went all-in on AI between 2023 and 2025 are hitting the same wall: the technology works in demos but stalls in production. The question founders need to answer right now isn’t whether AI matters. It does. The question is whether the way most companies adopted it was fundamentally wrong.
The ROI gap is worse than anyone expected
The numbers are stark. RAND Corporation data shows 80.3% of AI projects fail to deliver business value. That breaks down as 33.8% abandoned before reaching production, 28.4% completed but delivering no measurable value, and 18.1% unable to justify their costs. Only 19.7% of AI projects actually achieve their stated business objectives.
That failure rate isn’t a startup problem. It’s an enterprise problem. S&P Global reported that 42% of companies abandoned at least one AI initiative in 2025, more than double the 17% rate from 2024. IBM put the number of initiatives delivering expected ROI at just 25%.
The median spend tells the story even more clearly. Companies invested an average of $6.8 million per AI initiative but delivered only $1.9 million in value, according to aggregate industry data. That’s a negative 72% median ROI. For the projects that did work, median ROI hit 188%. The gap between winners and losers isn’t narrow. It’s a canyon.
A Gartner survey of 782 infrastructure and operations leaders published in April 2026 found that only 28% of AI use cases fully succeeded and met ROI expectations. Twenty percent failed outright. And here’s the finding that should make every founder pause: companies that reported AI-driven workforce reductions didn’t see higher returns. The businesses that cut headcount to fund AI saw the same financial results as those that kept their teams intact.
57% of I&O leaders who experienced at least one AI failure said they expected too much, too fast. The companies with the highest gains weren’t replacing workers. They were amplifying them.
The pattern repeats across sectors. Healthcare companies that bought AI diagnostic tools found them accurate in test environments but useless without clean, standardized patient data. Financial services firms deployed AI trading assistants that worked on historical data but couldn’t handle the volatility of real markets. Retail companies implemented AI demand forecasting that was no better than the spreadsheet models it replaced. The technology works. The organizational plumbing to support it doesn’t exist in most companies.
Gartner predicts that 60% of AI projects will be abandoned through 2026 specifically because of inadequate data foundations. Not inadequate AI. Inadequate data. The companies pulling back from AI aren’t discovering that the technology is bad. They’re discovering that their data, their processes, and their organizational structures weren’t built for it.
What happened to Salesforce’s AI team?
Salesforce’s Agentforce story is the clearest case study of the pullback pattern. The company invested heavily in agentic AI, positioning it as the future of customer relationship management. Benioff called it transformational. Then reality intervened.
In early 2026, Salesforce laid off under 1,000 employees across four core teams, including the Agentforce division. Adam Evans, the EVP and GM of AI, left the company. He was the fourth senior executive to depart within three months. The restructuring consolidated Agentforce and Slack under Joe Inzerillo, the President of Enterprise AI and Technology.
The numbers tell a complicated story. Agentforce had 18,500 customers by mid-2026, with 9,500 paying and customer counts growing 50% quarter over quarter. Those aren’t failure numbers. But they apparently weren’t enough to justify the original organizational structure or headcount either. Salesforce didn’t abandon AI. It recalibrated how many people and how much money the bet required.
That recalibration is happening everywhere. April 2026 was the worst month of tech layoffs this year, with companies like Oracle cutting 30,000 roles and Meta eliminating 8,000 positions. The layoffs weren’t always labeled “AI-related,” but the pattern is consistent: companies are reallocating headcount budgets toward AI infrastructure while discovering the infrastructure doesn’t yet replace what it displaced.
Is the AI bubble bursting in 2026?
No, but the air is leaking. The distinction matters for founders.
Global AI capex is still climbing. Goldman Sachs revised its 2026 estimate upward from $465 billion to $527 billion. Amazon, Meta, Google, and Microsoft are expected to invest a combined $650 billion in AI within a single year, most of it directed toward data centers, chips, networking, and energy. That’s not what a bursting bubble looks like.
What it looks like is a reckoning. A Morgan Stanley Research analysis mapping 3,600 stocks for AI exposure found that only 21% of S&P 500 companies could cite a measurable AI benefit, even as adoption rates climb across the board. The gap between “we’re using AI” and “AI is making us money” has become a chasm.
MIT’s research from late 2025 was even more damning: 95% of generative AI pilots failed to reach production. Not failed to deliver results. Failed to even get deployed. A National Bureau of Economic Research study from February 2026 found that 90% of firms reported no impact of AI on workplace productivity, while their executives projected AI would increase productivity by 1.4%. The expectations and the outcomes aren’t in the same time zone.
For founders, the practical takeaway is this: some AI companies are still growing explosively. Anthropic’s revenue grew 80x in a single quarter. The pullback isn’t about AI being useless. It’s about most companies not knowing how to use it.
The energy backlash nobody saw coming
Here’s the angle that hasn’t gotten enough attention. AI’s energy appetite is creating a political problem that’s starting to kill projects before they launch.
According to Bloomberg analysis, wholesale electricity costs have surged 267% over five years in areas near U.S. data center clusters. Those costs get passed directly to residential customers. Baltimore residents saw monthly bills jump more than $17 after a PJM power auction hit record highs. Another increase of up to $4 is expected by mid-2026.
The backlash is real and it’s quantifiable. Between March and June 2025, community opposition blocked or delayed $98 billion worth of data center projects. At least 25 projects were canceled in response to local objections. A Consumer Reports survey of 2,146 U.S. adults found that 78% are concerned that new data centers will increase their energy bills.
Bloom Energy projects that U.S. data center energy demand will nearly double from 80 to 150 gigawatts between 2025 and 2028. A Fortune report from April 2026 noted that data centers now account for half of all new U.S. electricity use, exactly as public opinion on AI is souring.
For founders building AI-dependent products, this creates a new constraint. Your cloud costs aren’t just a line item. They’re connected to a political fight that’s getting louder every month. Google Cloud’s $462 billion backlog already signals compute scarcity. The energy backlash could make that scarcity worse.
How Fidelity shows the “quiet pullback” in action
Fidelity’s restructuring is worth studying because it shows how the pullback works when companies are smart about it. In May 2026, Fidelity cut 800-1,000 jobs, about 1% of its 80,000-person workforce. But the company explicitly said AI wasn’t part of the equation. The changes were about staffing people best equipped to work on newer tech platforms.
At the same time, Fidelity planned to add 3,300 new jobs and fill 2,000 open positions, including 2,000 early-career engineering roles. The company scrapped its agile squad model in favor of larger teams designed to ship faster on key initiatives.
What Fidelity actually did was reorganize around what worked and cut what didn’t, without tying the narrative to AI at all. Compare that to companies that laid off workers explicitly “because of AI” and then discovered the AI couldn’t do the work. Fidelity’s approach looks more honest: restructure for the tools you’ve actually built, not the tools you hope to build.
The pattern holds across the BILL Holdings restructuring, where the fintech company cut 30% of its workforce while authorizing a $1 billion buyback. The story isn’t “AI replaced these jobs.” The story is “we’re reallocating capital, and AI is one of many factors.”
Why are companies that replaced workers with AI seeing worse results?
The Gartner finding deserves its own section because it contradicts the dominant narrative. Companies that used AI to eliminate jobs did not outperform companies that kept their teams. The data is unambiguous.
The explanation is straightforward once you see it. AI tools in 2026 are good at narrow, repeatable tasks: summarizing documents, generating first drafts, routing support tickets, flagging anomalies in data. They’re terrible at judgment calls, relationship management, creative problem-solving, and the thousand small decisions that keep a business running. Companies that fired the people doing those judgment-call jobs and expected AI to backfill discovered the gap within months.
HBR reported in January 2026 that companies were laying off workers because of AI’s potential, not its performance. The layoffs were anticipatory. Executives saw what ChatGPT could do in a demo and assumed that capability would translate to their specific workflows. It rarely did. The result: organizations with fewer people, the same workload, and an AI tool that could handle maybe 30% of what the departed employees used to do.
CFOs are starting to admit the scale of the problem privately. A Fortune survey from March 2026 found that CFOs project AI-driven layoffs will be 9x higher this year than publicly acknowledged. But the same survey revealed what they call the “productivity paradox”: the cuts aren’t producing the efficiency gains that justified them. Companies are quietly rehiring for roles they eliminated six months ago, often at higher salaries because the talent pool has moved on.
What should founders learn from the AI pullback?
The pullback reveals a split. On one side are companies treating AI as a cost center: buying tools, running pilots, cutting headcount to fund it, and seeing flat returns. On the other side are companies treating AI as a capability multiplier, keeping their teams and using the technology to make those teams faster.
The Gartner data is clear: companies using AI as “people amplification” reported the highest gains. The ones that replaced workers with AI and expected the same output didn’t. This isn’t a philosophical argument. It’s showing up in the financials.
Here’s what the data suggests founders should do:
Kill your pilots fast. The 95% GenAI pilot failure rate means long evaluation cycles are burning cash. Set a 90-day production deadline for any AI initiative. If it’s not in production by then, it won’t be. Gartner predicts 60% of AI projects lacking AI-ready data will be abandoned through 2026. Don’t be the company that takes 18 months to learn what should’ve taken 3.
Stop replacing people with AI. The data is unambiguous. Companies that cut staff to fund AI saw identical returns to those that didn’t cut anyone. Use AI to make your team 2x more productive instead of cutting your team in half. The math isn’t the same, and the companies that understand this are the ones seeing 188% median ROI instead of negative 72%.
Budget for energy costs as a strategic risk. If your product depends on GPU compute, your margins are exposed to an electricity cost shock that’s already 267% above five-year baselines in major data center markets. The infrastructure pressure between OpenAI, Microsoft, and Amazon is a preview of what’s coming for everyone.
Measure before you announce. Salesforce announced Agentforce as transformational before the numbers justified the claim. The S&P 500 data shows that markets reward companies that can prove AI pays and punish those that just talk about it. Morgan Stanley found that adopters delivering measurable results are seeing cash flow margin expansion at roughly 2x the global average. The keyword is “measurable.”
Watch who’s still doubling down. The pullback isn’t universal. Even xAI’s co-founder exodus hasn’t slowed Musk’s AI spending. Shopify is embedding AI deeper into merchant tools. The companies that survive the reckoning will be the ones that built real products instead of running pilots. Pay attention to the companies that can show revenue attributable to AI, not the ones that just talk about potential. That’s the dividing line between the next wave of winners and the companies that spent three years learning what didn’t work.
