David Linthicum on Cloud Economics, Enterprise AI and Ubiquitous Computing

Cloud strategy is shifting from choosing a single destination to deciding where each workload makes the most business sense. David Linthicum, chief cloud strategy officer at Deloitte Consulting, sees companies distributing applications and data across public cloud, private infrastructure, edge systems and specialist providers. For enterprise AI, that flexibility matters: cloud services can get teams started quickly, while high compute and storage costs make careful placement essential.

Linthicum brings more than 30 years of experience in enterprise technology. He leads Deloitte Consulting’s cloud technology strategy and architecture practice, working on solutions for clients across industries. He has written more than 15 books and over 7,000 articles, created more than 50 LinkedIn Learning courses, and teaches as an adjunct instructor at Louisiana State University.

Cloud strategy is becoming a workload-by-workload decision

For years, cloud adoption often began with a broad assumption: moving systems to a major provider would simplify operations and improve economics. Linthicum says that calculation has become more specific. Some applications and datasets don’t run as cost-effectively in public cloud, so organizations are reassessing where computing should happen.

The emerging model is distributed. An application might run at an edge location, in a company’s own data center, with a colocation or managed service provider, or in a cloud tailored to a particular industry or task. Small specialist providers, including those focused on AI services, can also serve selected needs. The public cloud remains part of the picture, and providers are still expected to grow. The change is that companies are spreading workloads across platforms instead of directing everything to one place.

That shift makes cloud decisions more dependent on the workload’s economics and requirements. A platform’s convenience is only one factor; teams also need to weigh operating expense and the value the system returns. Linthicum’s proposed direction is a heterogeneous, ubiquitous computing model: applications and datasets can reside wherever the combination of cost and capability best fits.

For organizations, this turns infrastructure planning into an ongoing allocation problem. They must understand what a workload needs, compare the available environments and account for the cost of running it over time. A familiar cloud provider may offer a useful service, but its brand alone doesn’t establish that it’s the right place for every system.

Generative AI is shaping cloud investment

Linthicum described generative AI as a central focus for cloud and enterprise technology providers in 2024. Providers were investing and retooling to expand their AI capabilities or develop applications that use generative AI. At the same time, many companies consuming those services were still researching possibilities and planning their approach rather than deploying AI broadly.

He sees that slower adoption as appropriate. Generative AI can demand substantial investment, and businesses need time to work out where it fits and what it will cost. His account points to a gap between market activity and deployment: providers were preparing services, while many prospective users were still evaluating how to apply them.

For technical leaders, the practical question is therefore not simply whether to adopt the technology. It’s which workloads justify the expense, which platform can support them, and how the costs will behave once use expands. Early investigation can help companies identify promising applications before committing to larger-scale operations.

Why cloud remains a convenient route to enterprise AI

Cloud services reduce the amount of infrastructure work required to get an AI project running. Teams can provision storage, compute and AI software on demand instead of first purchasing hardware, setting up networks, building databases and installing the necessary tools. Depending on the task, they can access generative AI, machine learning or deep learning systems, along with specialized compute resources.

That ready-made environment can shorten the path from an idea to a working application. Services designed to operate together give teams a place to assemble and deploy systems without having to build every infrastructure component themselves. For a particular application and dataset, cloud resources may also be well suited to the required workload.

Linthicum’s case for cloud is strongest on speed and flexibility. A team can bring a product to market sooner and adjust its approach as needs change. Those benefits may justify the operating cost when the resulting business value is high. But convenience doesn’t settle the financial question: companies still need to assess what it will cost to operate an AI system and whether the return warrants that expense.

For that reason, cloud is a strong starting point for AI deployment, rather than an automatic answer for every stage of a system’s life. The same project that benefits from cloud access to specialized resources might later be better served by another arrangement, depending on its workload and economics. Linthicum’s broader point is to evaluate the platform against the business outcome.

Ubiquitous computing puts workloads where they fit

Ubiquitous computing describes a model in which software and data can run across locations and types of infrastructure. That could mean edge computing, a conventional data center, colocation, a managed service, public cloud, or a personal device such as a phone or wristwatch. The goal is to select a suitable platform for each application, rather than to force all systems into the same environment.

Edge computing is one part of this mix. So are existing data centers and private servers. A workload might even run on a small system close to its users or on equipment already available to the business. Linthicum’s emphasis is on matching applications and datasets to the places where they can operate efficiently and provide the most value.

That placement can reduce cost and improve how well a workload runs, provided companies make the comparison carefully. Public cloud may be the right choice for one system; a private server may be more economical for another. The decision should reflect the application’s needs and the benefit it delivers, rather than a blanket preference for one type of infrastructure.

This approach also changes how companies think about cloud simplification. A more flexible technology stack lets teams configure systems around their own requirements and financial priorities. They can choose among cloud and hardware options instead of adapting every workload to the services offered by a single provider. Linthicum sees that choice as a way to manage complexity and make infrastructure decisions in the business’s interests.

Flexibility matters as AI workloads grow

The case for distributing workloads becomes more pressing as generative AI systems consume large amounts of processing and storage. Linthicum warns that running these systems in public cloud can produce very large bills. Some tasks need accelerators such as GPUs and TPUs. Cloud providers may be a sensible place to access those resources, especially when their surrounding services make deployment easier, but they aren’t necessarily the lowest-cost option for every workload.

Companies have limited budgets, so they need to direct spending toward the systems that return the most value. Linthicum argues that choosing services across providers and environments can give teams more room to do that. A company may prefer one cloud’s tools while finding that another part of its system costs less elsewhere. The relevant measure is what the overall setup contributes to the business.

That calls for a practical form of multi-cloud and hybrid planning: select databases, storage, compute and AI services according to the workload, even when that means combining offerings from different providers or using systems outside public cloud. Restricting every application to one provider can limit access to alternatives that better meet cost or technical needs.

Distribution also comes with a planning burden. Teams need to understand where their systems and datasets reside and compare the cost and performance of the available options. Linthicum doesn’t present one platform as a universal winner; the choice depends on the application and the value the company expects it to produce.

A broader technology base can expand what teams build

For Linthicum, ubiquitous computing is also a way to widen the path to new products and services. If teams can choose tools, platforms and locations according to what they’re trying to build, they have more options for turning ideas into working systems. Some projects may use generative AI on cloud infrastructure; others may combine cloud services with capabilities hosted elsewhere.

Mixing those environments can help control costs while giving teams access to the services they need. It also removes the assumption that every new application must stay within a company’s preferred cloud. Linthicum argues that broader access to technology can support efforts to improve customer experience, offer more value to customers and advance digital transformation.

The shift he describes is less about abandoning cloud than treating it as one part of a wider computing strategy. Cloud remains useful for rapid deployment, integrated services and access to specialized compute. But as AI workloads grow and costs become more visible, companies will need to make choices at the workload level. For technology leaders, the concrete task is to measure where each system performs well, what it costs to run and what business value it returns—then place it accordingly.