IBM is heading into a potentially rough stretch after its own chief executive, Arvind Krishna, pointed to a shift in how large enterprise customers are allocating budgets—one that could temporarily weigh on IBM’s deal momentum even as the broader market accelerates toward artificial intelligence.
Multiple reports indicate that IBM shares could fall sharply, with estimates pointing to a decline of around 23%. While stock moves are always driven by a mix of expectations and positioning, the underlying story is straightforward: customers are still spending heavily, but they are spending differently. Instead of committing to certain types of large, traditional IT contracts at the pace IBM had anticipated, many buyers are redirecting money toward the foundational components required for AI systems—chips, servers, and memory. That reallocation may not be “bad” for IBM in the long run, but it can be disruptive in the short term, especially when revenue recognition and contract timing depend on the shape of deals.
What makes this moment notable is that it isn’t simply a generic “AI demand is strong” narrative. Krishna’s comments, as reported, suggest a more specific phenomenon: large deals slipped because customers are prioritizing the physical and infrastructural building blocks of AI deployments. In other words, the market is not only buying AI software or services; it is also buying the hardware substrate that makes AI possible at scale. And that substrate is often purchased through channels and vendors that don’t map neatly onto IBM’s historical strengths.
To understand why this could hit IBM’s near-term outlook, it helps to break down what “large deals” typically mean for an enterprise technology provider. Large contracts often bundle multiple elements—consulting, integration, managed services, software licenses, and infrastructure commitments—into a single procurement cycle. When those cycles slow, even if overall spending remains robust, the provider can feel it quickly. The procurement process is also increasingly complex in the AI era: buyers must evaluate performance, power consumption, supply availability, interoperability, and total cost of ownership across a stack that spans accelerators, networking, storage, and orchestration software.
Krishna’s framing implies that some customers are pausing or reshaping parts of their procurement plans while they secure the most constrained resources first. Chips and high-performance server configurations are frequently the bottleneck in AI rollouts. Memory capacity and bandwidth requirements can also become gating factors, particularly for workloads that demand rapid access to large model parameters or data sets. If customers are moving budget from “end-to-end transformation” packages toward “get the compute now” purchases, IBM may see fewer large, bundled deals closing during the period investors care about most.
Yet the deeper question is whether this is a temporary mismatch—or a structural shift that changes IBM’s role in enterprise AI.
IBM’s position in the AI ecosystem has always been somewhat dual. On one hand, it has long marketed itself as an enterprise-grade platform and services company, with capabilities spanning hybrid cloud, data management, security, and consulting. On the other hand, it has also been associated with specialized hardware and systems—most notably through its infrastructure offerings and its involvement in AI-adjacent compute architectures. In the current environment, however, the center of gravity for AI spending appears to be moving toward the components that deliver raw compute throughput and memory bandwidth.
That doesn’t automatically exclude IBM. But it does mean IBM may need to compete more directly for portions of the stack that are increasingly dominated by other players—especially where customers are buying accelerators and server configurations as standalone items before they decide how to integrate them into broader platforms.
There is also a timing issue. Even when customers intend to use IBM’s software or services later, they may delay those decisions until they have locked in the hardware. Hardware procurement can be faster when the buyer knows exactly what performance targets they need and when supply constraints make “first mover” behavior rational. Software and services integration, by contrast, often requires more design work: mapping workloads to architectures, validating compatibility, and negotiating terms that reflect performance and operational requirements. If the hardware arrives first, the integration phase follows—but that follow-on can land in a different quarter than investors expect.
This is where the “slip” matters. A deal that would have closed as a large contract might instead be broken into smaller phases: first the compute, then the orchestration layer, then the data pipeline, then the managed operations. Each phase can still be valuable, but the revenue profile changes. For a company whose financial guidance and investor expectations are sensitive to the timing of large contract wins, even a modest shift in procurement sequencing can translate into a disproportionate market reaction.
The market’s reaction also reflects a broader tension in enterprise AI: the gap between excitement and implementation.
Enterprises are eager to deploy AI capabilities, but they are cautious about doing so in a way that creates operational risk or runaway costs. Many organizations are experimenting with pilots, but scaling from pilot to production requires a disciplined approach to architecture. That architecture depends on hardware choices that can be expensive and difficult to reverse. As a result, buyers often treat chips, servers, and memory as strategic purchases—things they want to get right early. Once those decisions are made, the rest of the stack can be tuned around them.
In that context, Krishna’s comments can be read as a signal that IBM is seeing customers prioritize the “hard parts” of AI deployment first. That includes not only the compute itself but also the supporting infrastructure that ensures the system can run efficiently: memory subsystems, storage performance, and the networking fabric that keeps GPUs or accelerators fed with data. If customers are reallocating budgets toward these areas, IBM’s share of the initial spend may shrink—even if IBM remains involved later.
Still, it would be a mistake to interpret this as a simple “IBM loses to AI hardware vendors” story. The enterprise AI stack is not a single product category; it is a system. Chips and servers are necessary, but they do not deliver business outcomes by themselves. Enterprises need orchestration, governance, security, observability, data integration, and lifecycle management. They also need to ensure that AI models can be updated, monitored, and audited in ways that satisfy regulatory and internal compliance requirements.
IBM’s opportunity, therefore, may lie in being the integrator and operator of AI systems once the hardware foundation is in place. The challenge is that integration and managed services often come after the hardware decision, which can delay revenue recognition. Investors may be focused on the near-term slip rather than the longer-term conversion of AI infrastructure spending into IBM-led deployments.
There is also a strategic nuance: customers may be shifting not only where they spend, but how they structure procurement. Some enterprises are moving toward modular purchasing—buying compute and storage from one set of vendors, software from another, and services from a third. This can reduce the attractiveness of large bundled deals. It can also increase the importance of partnerships and ecosystem positioning.
If IBM’s large deals are slipping because customers are buying chips, servers, and memory first, then IBM’s next task is to ensure it is positioned to capture value in the subsequent phases. That could mean deeper partnerships with hardware suppliers, clearer packaging of integration services, and more flexible commercial models that align with phased procurement. It could also mean emphasizing performance optimization and operational reliability—areas where enterprises are willing to pay once they have committed to the hardware.
Another angle worth considering is the nature of “AI spending” itself. Not all AI spending is equal. Some customers are investing in training infrastructure, which tends to be compute-intensive and memory-heavy. Others are focusing on inference at scale, which can have different performance characteristics and cost structures. Still others are adopting AI through managed services or via third-party platforms, which can shift the hardware requirements away from the customer’s direct procurement.
Krishna’s mention of chips, servers, and memory suggests that at least some of IBM’s customers are taking a more direct approach to building AI infrastructure. That is consistent with a world where enterprises want control over data, latency, and model customization. But it also means the procurement cycle is more exposed to supply constraints and pricing volatility in the semiconductor and server markets.
When those markets are tight, customers may prioritize the most constrained components first. That can create a temporary imbalance: the infrastructure gets funded, while the integration and transformation work gets postponed. IBM’s reported experience fits that pattern.
The potential 23% drop in IBM shares, if it materializes, would likely reflect investor concern that this shift could persist longer than expected. Markets tend to punish uncertainty, especially when guidance and backlog visibility are affected. If investors believe that customers are not just reprioritizing within a quarter but changing their procurement behavior for multiple quarters, the valuation impact can be severe.
However, there is a counterargument that could temper the pessimism: AI infrastructure spending is not a one-time event. It is the beginning of a multi-year buildout. Even if IBM sees fewer large deals now, the installed base of AI infrastructure will require ongoing management, optimization, security hardening, and continuous improvement. Those are areas where enterprise providers can build durable revenue streams—provided they can convert the initial hardware investment into long-term platform and services relationships.
In other words, the question is not whether AI spending is happening. It is. The question is whether IBM’s business model captures enough of the value chain as customers move from “buy compute” to “run and govern AI.”
IBM’s brand and capabilities have historically resonated with enterprises that want stability, compliance, and integration across complex environments. In the AI era, those needs intensify. Organizations are dealing with data governance challenges, model risk management, and the operational burden of deploying AI responsibly. They also face the reality that AI systems can degrade over time as data distributions shift or as models are updated. That creates demand for monitoring, auditing, and lifecycle management—services that are less glamorous than chips but essential for production.
If IBM can position itself as the partner that turns AI infrastructure into reliable business systems, the near-term deal slip may be a temporary artifact of procurement sequencing rather than a sign of declining relevance.
Still, investors will want evidence. They will look for indicators such as: