Over the past eighteen months, AI infrastructure spending has surged. Companies like NVIDIA and Microsoft continue to report record revenues driven by GPUs, cloud compute, and AI platform usage. In 2024 and 2025, AI related capital expenditure by major technology groups reached tens of billions of dollars per year, and a small number of AI focused stocks accounted for more than two thirds of total S&P 500 gains.
At the same time, market signals are becoming more cautious. Despite strong revenues, AI exposed stocks have shown increased volatility and capital rotation. Analysts point out that current valuation levels assume large scale downstream value creation that has yet to fully materialize outside of infrastructure sales.
This tension is rooted in where AI money is actually made today.
The vast majority of current AI revenues come from selling hardware, compute capacity, and cloud services. Infrastructure providers generate revenue as long as experimentation continues, regardless of whether AI applications reach sustainable market adoption.
When looking at AI usage beyond infrastructure, industry data paints a more restrained picture. Across large enterprise surveys from consulting and research firms, results are consistent:
- Around 70 to 80 percent of AI initiatives remain at pilot or limited deployment stage
- Less than 20 percent are scaled into production with measurable business impact
- Many projects are paused or abandoned due to integration cost, reliability issues, or unclear return on investment
In other words, investment has clearly moved faster than value realization.Âą
This gap does not reflect a lack of innovation or model capability but in large part the difficulty of execution. Turning AI into a sell-able, reliable, and profitable product requires robust system design, realistic performance assumptions, disciplined engineering, and close alignment with real user needs.
As expectations normalize, this distinction becomes decisive. The AI efforts that endure will not be those with the most ambitious claims, but those that can consistently convert technology into working products that customers are willing to pay for.
This is precisely where rigorous technical judgement and pragmatic execution focused expertise makes the difference between experimentation and lasting value.
NEXTWave Technologies SRL
Bridging deep tech from experimentation to real products