The Great AI Reset: 2027 Isn't A Spending Story—It's A Recalibration Story

Girish Joshi, SVP of technology at Collabera, has led Fortune 500 digital shifts for 25+ years and now drives AI and agentic transformation.

gettyWill enterprises keep spending on AI in 2027 at their current pace? Likely not, and that’s not a verdict on AI. It’s a verdict on how the last three years were spent, and what enterprises learned from spending them.

The reset will be less about belief in AI and more about recalibration: which experiments, architectures and vendor dependencies have earned a permanent place in the business, and which haven’t.

Some of the reset is straightforward. PwC’s 2026 Global CEO Survey found that 56% of CEOs had seen no meaningful revenue or cost benefit from AI, while only 12% had seen both. Gartner found that only 28% of AI use cases in infrastructure and operations fully met ROI expectations, while 20% failed outright.

Projects that cannot justify their return and keep consuming capex and opex without a credible path to value will be shut down. This part of the reset barely needs debating. The harder question is everything else.

A smaller number of AI use cases will do the opposite: scale into differentiated products, platforms and offerings that become part of the business, not a tool inside it. Deloitte’s 2026 State of AI in the Enterprise research found that only 34% of organizations are truly reimagining the business with AI, while another 30% are redesigning key processes around it. The significance is how few organizations have moved beyond using AI to optimize what already exists.

The next phase won’t be defined by more AI initiatives. It will be defined by fewer AI capabilities that customers, competitors and the business itself find difficult to live without.

There is a deeper reason many enterprises will pull back, and it has little to do with ROI. The excitement around new, cheaper models has worn off. For enterprise buyers, it now feels less like progress and more like a burden, since every significant shift can mean redoing evaluation, integration and governance from a point already passed.

Few enterprise technology categories in the last two decades have moved this fast. ERP, cloud and mobile evolved, but production architectures were expected to stay stable enough for multi-year investments. AI has challenged that: a new model can change an implementation’s economics, a cheaper model can trigger another evaluation and a new capability can make yesterday’s architecture look dated.

That pace was tolerable during experimentation. It gets harder once AI becomes a production dependency, since today’s commitment may not be right a year from now.

AI’s cyber risk is also intensifying. In July, an independent investigation by METR found that roughly 700 of the AI agents OpenAI was testing were involved in a coordinated attack on Hugging Face, and that some attempted to conceal their activity from investigators.

The significance isn’t that an AI agent breached a system. It’s that increasingly autonomous systems can interact with tools, infrastructure and external environments in ways conventional software and security tools were never built to handle. Gartner predicts that more than 40% of agentic AI projects could be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. The natural consequence will be more spending on AI security, governance and containment. That spend is part of the cost of running AI, not an afterthought.

None of this makes 2027 a budget exercise so much as a test of the fundamentals enterprises built over the past three years: which architecture decisions, vendor bets and operating assumptions held up, and which were convenient at the time.

AI has created an unusual situation: the technology beneath a business capability can change faster than the capability itself, making architectural flexibility and vendor substitutability strategic requirements rather than nice-to-haves.

Large AI companies are responding by integrating vertically, using acquisitions and partnerships to control more of the stack. Nvidia’s $12.9 billion acquisition of Hugging Face is a clear example, extending its reach deeper into the AI developer ecosystem.

Enterprise buyers do not appear as excited. Their landscape is already running, and switching costs are too high to treat every consolidation move upstream as an opportunity. What looks like progress to the vendor can look like tightening dependency to the buyer, making it important to understand whether consolidation reduces complexity or simply increases dependency.​

Vendor lock-in is not new. Most large organizations that committed to Azure, AWS or Google Cloud a decade ago are still there, not because lock-in never happened, but because the platform kept being worth staying on. That’s a trade most CIOs have already made peace with.

AI complicates that trade. The risk was never that an enterprise would get locked in. With any useful technology, that was inevitable.

The real risk is that what gets locked in can stop making sense on a timeline shorter than the contract, the amortization schedule or the business case that justified it. A cloud platform chosen in 2016 is recognizably the same platform today. An AI architecture chosen in 2024 already is not.

That gap is what the earlier fatigue is really about. The enterprises that get 2027 right won’t be the ones that avoided lock-in altogether. That was never realistic. They will be disciplined about what they got locked into, choosing a vendor, model or architecture defensible three years out, not the one that looked strongest when purchased. This is the recalibration underway: three years of evidence, applied deal by deal, deciding what stays, what changes and what should never have been built this way.

The real AI reset is therefore not a question of how much to spend. It is a question of what kind of dependency that spend creates: one the enterprise can still defend in 2029, or one it will quietly be unwinding by then.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?