Economist and author David McWilliams uses “digital lettuce” to describe AI hardware investment he believes can spoil quickly. In a Nov. 20, 2025 Fortune interview, he said, “You’re investing in something that is a perishable good.” His point is a warning about rapid GPU obsolescence, not a verified expiration date for every graphics processor, data center or AI investment.
What McWilliams means by “digital lettuce”
McWilliams is applying a food metaphor to the most fast-changing part of an AI data-center buildout: accelerator hardware, especially GPUs used for model training and inference. Unlike a building, power connection or fiber route, a newly purchased chip can lose its competitive value when a faster, more efficient generation arrives.
His strongest formulation, also reported by Futurism on Nov. 21, 2025, is: “Technological change suggests that if you buy a GPU today, the chip is going to be outdated next year.” That is McWilliams’s forecast about technological change and investment returns. It is not an engineering study showing that every GPU becomes unusable after 12 months.
Why the GPU distinction matters
A chip can remain operational while becoming less attractive economically. Investors and operators may care about several different thresholds:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Technical usefulness: whether the hardware can run the required models and workloads.
- Relative performance: whether newer chips deliver more throughput for the same space, power and operating cost.
- Revenue potential: whether customers will still pay enough to justify the older equipment.
- Resale or redeployment: whether the hardware can move to less demanding jobs after newer accelerators arrive.
Under the “digital lettuce” view, the risk is that spending on GPUs has to be refreshed faster than a conventional infrastructure investor expects. That concern applies most directly to the accelerator inventory, not automatically to the entire data-center asset.
Ed Yardeni’s counterpoint: facilities can outlast a hardware cycle
Fortune’s interview also presents a different way to measure useful life. Economist Ed Yardeni said, “Data centers existed before AI caught on in late 2022, when ChatGPT was first introduced.” He argued that some facilities continued operating with their original chips even after the generative-AI boom began.
Fortune attributes to Yardeni the statement that, “During 2021, there were as many as 4,000 of them in the U.S.” This is his reported estimate in the 2025 interview, not an independently verified official census. His broader point is that data centers are not synonymous with the newest AI accelerators: buildings, electrical systems, cooling, networking and servers can continue serving other workloads when a particular GPU generation loses its edge.
Two different useful-life questions
| Question | McWilliams’s emphasis | Yardeni’s emphasis | What the reporting establishes |
|---|---|---|---|
| How long can a GPU remain competitive? | Rapid technological change may make a newly bought chip outdated the following year. | Not the focus of his counterargument. | A prediction attributed to McWilliams; no universal replacement period is established. |
| How long can a data-center facility operate? | Large AI spending could be exposed if its economics depend on quickly aging hardware. | Facilities existed before ChatGPT and some retained original chips. | The reports describe competing arguments, not an engineering lifespan for facilities. |
| What happens to older equipment? | Its investment value may wilt as newer technology arrives. | Older chips can remain in service for some workloads. | No model-by-model utilization, resale or performance study is provided. |
| How should companies depreciate assets? | Shorter economic lives could matter if hardware really loses value quickly. | Longer-lived facilities could support a different view. | No company-specific schedules or quantified accounting effect is established. |
What these reports do—and do not—prove
The Fortune and Futurism articles document attributed arguments about an AI investment boom and the possibility of a crash. They do not provide:
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →- an engineering test of how long current GPU models function;
- a universal GPU replacement timeline;
- a current, audited inventory of U.S. data centers;
- company-by-company depreciation schedules; or
- a quantified estimate of how changing useful-life assumptions would affect profits, asset values or investor returns.
That distinction is important when a headline turns a forecast into a seemingly precise statistic. “Outdated” can mean technically unable to run a workload, less efficient than a newer chip, or simply less profitable at prevailing prices. Those are different outcomes.
How to assess an AI data-center investment claim
- Identify the asset. Is the spending for land and buildings, power and cooling, networking, general servers, or short-cycle GPU accelerators?
- Ask what useful life is being assumed. Check whether the figure is an accounting estimate, an operating expectation or a prediction about competitive performance.
- Separate capacity from competitiveness. A facility may remain open while particular chips are reassigned, written down or replaced.
- Check workload flexibility. Hardware that can serve more than the newest frontier models may have a longer practical role, although the available reports do not quantify that extension.
- Look for evidence behind the forecast. A credible claim should identify the chip generation, workload, power cost, utilization, replacement spending and customer demand that support its assumed returns.
- Treat crash predictions as scenarios. McWilliams’s warning describes a possible outcome of rapid technological change; it is not a measured result that applies to every operator or investor.
The practical reading of “digital lettuce”
McWilliams’s metaphor is most precise when applied to rapidly changing GPU capital whose economic value depends on staying near the leading edge. Yardeni’s response is most relevant to the broader facility: data centers can predate the generative-AI wave and may continue operating with older equipment or different workloads. The available reporting supports keeping those questions separate rather than assigning one universal lifespan to “AI data centers.”
Quick Recap
Best Value
- Family farms not data design for people against AI server farms, data center expansion, rural land buyouts, corporate agriculture, and industrial tech development replacing farmland and open space. Rural conservation and anti data center message.
- AI protest design for farmers, land conservation supporters, anti AI activists, sustainability groups, environmental advocates, rural communities, and people opposing server farm construction, power grid strain, and farmland destruction.
- Lightweight, Classic fit, Double-needle sleeve and bottom hem
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
