The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
No—enterprise spending alone does not guarantee AI leadership. Capital can buy computing capacity and specialist expertise, but those inputs become an advantage only when an organization can apply them to valuable work, integrate AI into its operations, and measure the results. Evidence points to uneven adoption and productivity effects, not a reliable formula that turns a larger budget into better outcomes.
What “AI leadership” requires beyond a large budget
AI investment can provide essential building blocks: compute, models, data infrastructure, and people with specialized skills. But buying those inputs is different from deploying them successfully, improving productivity, or earning a financial return. Those are separate outcomes, and one does not automatically establish the next.
Joe Bertolami, co-founder and CTO of Clifton AI, makes this case in a September 10, 2026, opinion piece for CIO. He argues that accumulating GPUs and hiring AI specialists is insufficient without broad adoption, an engineering culture that supports implementation, and conditions that help experienced talent remain. His is an argument about enterprise strategy, not proof of a universal law. Read Bertolami’s opinion piece.
The practical implication is to treat AI spending as an input to a business change—not as evidence that the change has happened. A company may have substantial capacity yet struggle to identify a useful application, adapt a process, or sustain the systems it builds.
#1 Best Overall
What the adoption and productivity evidence shows
Adoption is broad, but adoption is not a return
Stanford HAI’s 2026 AI Index reports that 88% of organizations had adopted AI. That figure indicates widespread uptake, but it does not mean all those organizations have deployed AI successfully, improved performance, or generated a financial return. Stanford HAI’s 2026 AI Index also cautions readers to assess model benchmarks and independent testing carefully.
Productivity gains vary among companies
A 2026 National Bureau of Economic Research working paper reports a survey of nearly 750 corporate executives. The authors find heterogeneous AI adoption and positive but varying labor-productivity gains. They associate the observed gains more with revenue-based productivity and innovation- and demand-oriented channels than with capital deepening alone. Survey findings do not guarantee that a particular investment will cause gains at a particular company. Read NBER Working Paper 34984.
Readiness and fit affect adoption
A Stanford Graduate School of Business study reports that 22.8% of U.S. manufacturing plants said they used AI in 2021. This is a historical measurement, based on a purpose-designed survey of about 28,500 establishments—not a current adoption rate. The study links adoption to more recent digital infrastructure and structured production processes, and identifies cost, lack of an applicable use case, and expertise as barriers. The findings underline why a budget may not be enough: a plant also needs a suitable task and the operational foundations to implement it. See the Stanford GSB study summary.
How cheaper and different models change the investment decision
Falling query costs can make more AI workloads economically plausible, but they do not make every model suitable for every task. Stanford HAI’s 2025 AI Index reports that the cost of querying a model with GPT-3.5-equivalent accuracy on MMLU fell from $20 per million tokens to $0.07 per million tokens between November 2022 and October 2024. The October 2024 example was Gemini-1.5-Flash-8B. This is a benchmark- and model-specific comparison, not a universal estimate of enterprise AI costs. See the AI Index economy chapter.
Hosted frontier models, open-weight models, and localized deployments can involve different trade-offs. Compare them against the actual workflow rather than assuming that the most capable, cheapest-per-query, or most self-hostable option is automatically best.
- Task quality: Does the candidate model perform well on representative examples from the intended workflow? General benchmarks may not predict results on a company’s own tasks.
- Total operating cost: Include the costs relevant to the deployment, not only a quoted query rate.
- Data handling: Determine whether the deployment meets the organization’s privacy and data-handling requirements.
- Integration and maintenance: Assess whether the organization has the expertise and capacity to connect, monitor, and maintain the system.
Bertolami’s opinion piece argues that privacy concerns and operating costs increase interest in smaller or localized alternatives. Those considerations are valid evaluation questions, but they do not establish that an open-weight or locally run model will meet every organization’s quality, security, or operational needs.
Rank #4
A more useful way to evaluate AI investment
Instead of treating spending as a proxy for leadership, evaluate a proposed deployment as a business and operational decision:
- Start with the workflow. Identify a specific task or process and the problem AI is meant to address. If there is no applicable use case, adding compute will not create one.
- Set a measurable outcome. Define what would count as improvement for that process, such as a relevant productivity or revenue measure. Distinguish that result from model access or usage.
- Check organizational readiness. Review the data and digital foundations, process structure, and expertise needed to implement the use case. The manufacturing study’s findings make these practical checks particularly relevant in production settings.
- Test candidate deployments on the task. Compare model quality, operating cost, data requirements, and the organization’s ability to integrate and maintain each option. Do not infer workflow performance from a general benchmark alone.
- Track deployment and outcomes separately. Record whether the system is adopted in day-to-day work, whether the workflow changes as intended, and whether the business measure improves. Adoption is not itself evidence of a financial return.
There is no universal ratio of AI spending to business value established by the evidence cited here. Nor does it prove that talent retention alone determines AI leadership. The defensible conclusion is narrower: investment creates options, while results depend on whether an organization can choose a fitting use case and turn its resources into working, valuable processes.
What large AI capital expenditure does—and doesn’t—tell you
A 2026 NBER working paper reports that the five largest U.S. technology firms spent $380 billion in capital expenditure in 2025, with roughly double forecast for 2026. That figure covers those firms’ capital expenditure; it is not the same measure as the more-than-$750-billion AI infrastructure spending projection cited in Bertolami’s CIO opinion piece. The estimates should not be combined or treated as directly comparable. Read NBER Working Paper 35290.
Large spending figures can signal that companies are building capacity, but they do not establish how well that capacity will be used or what returns it will generate. For enterprise leaders, the relevant question is not simply how much is being spent across the market, but whether a specific investment improves a defined workflow under the organization’s real operating constraints.
Quick Recap
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.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →
