What public cloud gets wrong with AI

AI is not cheap. During the past year, public cloud giants have significantly accelerated their investments in AI. By the close of 2024, Microsoft had invested more than $80 billion in AI infrastructure, partnerships, and development. AWS is on track to spend nearly a year’s worth of its revenue on AI in 2025, with AI data center spending up 13.2% to $9.3 billion in 2024. Major tech companies, including Google and Meta, collectively contribute to a $1 trillion global wave of generative AI investments projected in the next few years, much of it pouring directly into their hyperscale data centers and cloud-based AI services

For the better part of a decade, public cloud providers have touted their AI offerings as the next frontier in digital transformation. With promises of limitless scalability, rich prebuilt services, and democratized intelligence, we were led to believe that AI in the cloud was a simple answer for complex business needs. Yet, after countless conversations and firsthand engagements with enterprises, I see a growing gulf between what the major cloud providers sell and what enterprises truly need.

Business value lost in translation

The biggest marketing failure is not just technical; it’s a disconnect between understanding and defining business value for enterprise customers. When large public cloud providers launch their newest AI models and APIs, they announce partnerships with AI leaders and host summits that highlight the next wave of technology that their products will support. Although these efforts generate excitement, they seldom address the core question every enterprise leader should ask: “How will this improve my business?”

Enterprises aren’t interested in AI for AI’s sake. They care about outcomes: better customer engagement, reduced costs, optimized operations, and new revenue streams. Yet I see cloud providers pushing a technology-first message, focusing on technical capabilities rather than real business cases. The sales pitch often boils down to “You need AI because it’s the future,” rather than “Here’s how AI can solve your unique business problem and deliver measurable value.”

Business leaders are navigating through lists of AI services, from natural language processing to computer vision, trying to connect features to business opportunities. The responsibility falls on companies to bridge the gap between what the technology can offer and what their business actually requires. Rarely do cloud vendors provide verticalized solutions that match the complex reality of large enterprises.

AI costs too much

Here’s the real elephant in the room that mega cloud providers are reluctant to discuss: For many enterprises, the cost of running AI workloads in the public cloud is untenable.

Lift any enterprise rock and you’ll uncover anxiety about cloud bills, especially as AI experiments transition into ongoing production. Training large models, running inference on vast data sets, or integrating generative AI into business processes comes with unpredictable and often eye-popping monthly charges. Cloud providers offer flexible pricing but rarely transparency or predictability, particularly as workloads grow. Many companies realize too late that their cloud AI ambitions are racing toward budget limits.

As a result, an increasing number of organizations are revisiting on-premises deployments or turning to smaller cloud vendors that can offer competitive pricing, flexible deployments, and more willingness to create bespoke, value-based deals. It’s a reversal from the cloud’s early days when cost and agility were assumed to always favor the hyperscale public cloud. Now, as egress fees, specialized hardware (such as GPUs and TPUs), and proprietary lock-in increase costs, many enterprises are questioning whether public cloud AI aligns with their value propositions.

Smaller providers and on-prem solutions can sometimes—and in some cases, often—deliver AI features at a fraction of the cost, especially when paired with strong open source communities and lower operating overhead. As the market evolves, the message is clear: Public cloud AI is no longer synonymous with best-in-class value or even state-of-the-art capability.

Moving from cloud-only to best value

Given these challenges, how should enterprises proceed? I recommend that organizations abandon “cloud-only” thinking and adopt a best-value strategy that focuses on how the technology serves the business objectives rather than where it resides. Here’s how you do it:

Let business problems drive technology choices. Start with a deep understanding of business pain points and desired outcomes. Build a business case for AI rooted in reality, not hype. Then let those requirements drive the evaluation of technology stacks, vendors, and deployment models.

Assess value, not just features. It’s easy to get lost in feature comparisons between cloud providers, but the true differentiator is the business value they offer. Look for solutions, whether from a hyperscale cloud, a niche provider, or even on premises, that can deliver tangible ROI in terms of cost reduction, productivity, customer experience, or risk mitigation.

Be aggressive with cost analysis. Don’t rely solely on estimates or calculators. Model your expected AI workloads, conduct pilots, and compare total costs across different deployment options. Pay close attention to long-term or hidden costs such as egress fees, premium support, scaling limits, and mandatory upgrades.

Demand transparency and partnership. View technology vendors as partners instead of just providers. Insist on transparency regarding pricing, road maps, and support. If a vendor doesn’t understand your business or can’t clearly explain how its technology will help achieve your specific goals, consider looking elsewhere.

Favor flexibility and interoperability. Adopt a hybrid or multicloud approach when it makes sense. Today, the best results usually come from selecting the right tool for each situation: combining public cloud, private infrastructure, and edge solutions as needed. Support integration standards and portability to prevent lock-in.

The era of relying solely on the public cloud for AI innovation is over. Major cloud providers have extensive capabilities but often miss the mark for enterprise buyers because they fail to clearly define, deliver, or price business value. Enterprises shouldn’t feel pressured to adapt their needs to a predefined set of cloud services. Instead, let business problems and goals guide technology selection. Pursue a best-value strategy that leverages the strengths of each platform as appropriate. Ultimately, the right answer is the solution that puts business value first.

Enterprises that get this right will cut through the noise to make technology decisions that unlock real, sustainable advantages in the age of AI. The easiest way to get on that right track is to lead with this question to vendors: “How can your AI products solve my unique business problems and deliver measurable value?”