TECHnalysis Research Blogs
TECHnalysis Research president Bob O'Donnell publishes commentary on current tech industry trends every week at LinkedIn.com in the TECHnalysis Research Insights Newsletter and those blog entries are reposted here as well. In addition, those columns are also reprinted on Techspot and SeekingAlpha.

He also writes a regular column in the Tech section of USAToday.com and those columns are posted here. Some of the USAToday columns are also published on partner sites, such as MSN.

He also writes occasional columns for Forbes that can be found here and that are archived here.

In addition, he has written guest columns in various other publications, including RCR Wireless, Fast Company and engadget. Those columns are reprinted here.

August 18, 2026
How Tokenomics Might Save AI PCs

By Bob O'Donnell

There are some interesting things happening in the PC market these days and, more importantly, the potential for even more impactful changes over the next year or so. At a high level, of course, overall PC shipments have been predicted to and indeed are starting to decline on a unit basis. Driven primarily by the huge increases in memory and storage costs, the average selling prices for PCs have gone up significantly over the last year and that has dampened overall demand, particularly for the cost-sensitive consumer and other low-end PC segments.

Despite this, however, the major PC makers—notably, Dell, HP and Lenovo—have all reported higher overall PC revenues and are expecting PC revenues to be up for 2026. The reason is that demand for higher-end, more capable PCs continues to be strong and the greater revenues that these much more expensive PCs are generating is more than offsetting the revenue losses from lower-priced models. That, in itself, is both interesting and surprising.

However, I believe we could be on the precipice of an even larger shift: a significant increase in demand for AI-capable PCs. The explosive growth in enterprise AI usage is creating an entirely new corporate expense category—tokens—and the need to control that spending could finally provide the economic rationale for AI PCs that the industry has struggled to articulate.

There are numerous structural changes in how businesses think about their computing demand which have occurred—and a few more that still need to happen—that point to this change. First, is the fact that companies have suddenly had to deal with a new multimillion-dollar expense category—AI inference spending, increasingly measured and managed through token consumption. Between trends like tokenmaxxing and the now generally accepted thought that unless companies leverage the new GenAI and agentic AI capabilities as aggressively as possible they stand to fall behind their competition, companies are having to redirect enormous sums of money toward something they hadn’t really even considered before.

Initially, the general excitement and sense of urgency in leveraging GenAI and agentic AI meant that little oversight and analysis of these efforts were being made. Now, however, as companies start to realize that these tokenomics issues are going to be a large and long-term part of their corporate budgets, there’s a great deal more attention being paid to how these costs can be managed.

At first, virtually all the token requests were made to cloud-based services. Early on, that wasn’t a big factor because most of the tokens were being generated for free (a situation that is still primarily the case in other parts of the world, notably China). That situation changed dramatically about a year or so ago when major model providers started charging for these tokens and already, we have major model suppliers like Anthropic supposedly reaching staggering annual revenue rates of $65 billion!

At the same time, there have been several technical advances that are opening new options for generating tokens. Improvements in the performance of smaller models, the dramatic rise in the usage of customizable open weight models, and the growing availability of AI infrastructure offerings specifically designed to be deployed in enterprise data centers are all driving new ways to think about how AI-focused computing demands can be met. Notably, the rise of hybrid AI architectures that combine cloud, on-premises and on-device computing is giving organizations more choices about where tokens are generated. There’s also been the recognition that not all AI requests need to be handled by frontier-level models—smaller and more specialized models can not only handle these requests but, in some situations, even provide better, more accurate responses.

Suitably-equipped AI PCs, “deskside” workstations like those powered by Nvidia’s GB10 chip and AMD’s Ryzen AI Max/Max+ 400x, fit perfectly into this new scenario. They can run many of the more powerful, more compact models that are appearing on a daily basis and provide an intriguing new economic alternative to token generation. In fact, some have argued that they’re essentially capable of running the equivalent of frontier-level models from about a year ago. Plus, because of the growing number of AI applications and agentic platforms that leverage open-weight models, they can be leveraged for more specialized applications and solutions.

But beyond all these technical reasons, the really simple and most compelling reason to leverage these AI PCs is economics—or rather, tokenomics. For sufficiently heavy AI users, diverting even 20% of token consumption from expensive cloud models to local inference could materially shorten the payback period on a $4,000 AI PC—and in some high-usage scenarios potentially reduce it to months rather than years. Toss in the potential to send say another 30% of their token requests to an on-prem, GPU-equipped server (or enterprise AI factory as Nvidia’s Jensen Huang has labelled them) and the savings could be even higher.

Of course, there would be initial capital outlay to make these purchases, but building an ROI model to justify them is getting easier and easier every day. Plus, by tapping into the newly created token funds/budget lines that organizations have been forced to create to pay for those purchases makes the arguments even easier.

As logical as this all sounds, however, there are certainly still obstacles in the way. First, the existing PC procurement process in most organizations is done in a way that could never justify these kinds of purchases. Companies typically have three-to-five year lifecyle plans for their PCs and look to keep their purchase prices within the historical norms of $1,000-$2,000 for corporate PCs. Making a shift to a fleetwide deployment of significantly more expensive machines would require C-suite level changes to well-established (and likely highly protected) processes. The critical change that needs to happen is that enterprises need to stop thinking about AI PCs strictly as endpoint purchases and start thinking about them as distributed AI infrastructure.

Second, the challenge of how to best split up and orchestrate the different elements of a single AI prompt or workflow to leverage each of the different computing resources in a hybrid AI environment has yet to be solved. Many organizations are clearly working on this challenge and industry standards like MCP and A2A certainly help create the interoperability foundation to do more heterogeneous AI computing. What remains immature is the intelligence layer capable of dynamically determining which model and which compute resource should handle each part of a workflow based on capability, cost, latency, privacy and availability.

Thankfully, we are also seeing efforts by some of the largest model providers to start sharing versions of their largest models into different environments. Several months back, Google announced a deal with Dell that allows versions of Gemini to run on Dell AI infrastructure within enterprises. In addition, there are clearly economic incentives for companies like Anthropic and OpenAI to create tools that allow them to maintain some degree of usage within a Hybrid AI environment. Inevitably some amount of most AI prompts will likely require a response from the latest frontier model and by creating tools that make that process seamless, they can guarantee a level of usage even as organizations work to reduce their dependence on the latest frontier models (as they inevitably will). To put it succinctly, better to orchestrate 100% of an enterprise's AI activity and directly monetize 30% of it than insist on processing 100% and risk being bypassed.

To be clear, AI PCs aren’t the only solution to the growing economic challenge that massive token consumption is starting to create. In the same way that early unfettered cloud computing usage led to FinOps platforms for cloud consumption, we’re bound to see a dramatic rise in both the number and variety of solutions designed to allow companies to manage their tokenomic challenges as AI usage continues to grow. From a practical, economic perspective, however, powerful AI PCs (and workstations) can and should prove to be formidable tools in helping organizations achieve those goals. Ironically, then, the killer app for AI PCs may not turn out to be a particular AI application at all. It may simply be economics. If enterprises discover that putting more AI compute on employees' desks can meaningfully reduce the rapidly growing cost of cloud-based inference, the tokenomics of AI could finally provide AI PCs with the compelling ROI story they've been missing.

Here’s the link to the column: https://www.linkedin.com/pulse/how-tokenomics-might-save-ai-pcs-bob-o-donnell-jtrxc

Bob O’Donnell is the president and chief analyst of TECHnalysis Research, LLC a market research firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on LinkedIn at Bob O’Donnell or on Twitter @bobodtech.