Technology executives quoted in CNBC's July 12 reporting describe current demand for AI chips and data-center capacity as "almost unlimited," a striking characterization given the increased volatility in AI-linked stocks over recent weeks. The framing suggests that whatever near-term equity-market jitters exist around AI infrastructure names, the underlying buyer-side appetite for compute hasn't meaningfully cooled.
What has changed, according to the same executives, is how enterprises are buying. Rather than simply committing to more raw compute capacity on the assumption that more infrastructure automatically translates into more AI value, buyers are increasingly "valuemaxxing" -- squeezing measurable return out of infrastructure and model spend they've already committed to before adding more. That's a materially more disciplined posture than the land-grab dynamic that characterized much of the 2024-2025 AI infrastructure buildout.
The valuemaxxing framing lines up directly with hard survey data: VentureBeat's own research found 86% of enterprises running their own GPU infrastructure report utilization of 50% or less. Executives describing a shift toward extracting more value from existing spend is effectively buyers acknowledging and responding to exactly the underutilization problem that survey quantified, rather than continuing to over-provision blindly.
“That's a materially more disciplined posture than the land-grab dynamic that characterized much of the 2024-2025 AI infrastructure buildout.”
The result is a genuinely bifurcated market signal: public equities tied to AI infrastructure have shown real volatility as investors debate whether the buildout has outrun demand, while the executives closest to actual purchasing decisions describe underlying appetite for compute as effectively unconstrained. Reconciling those two signals -- volatile pricing against strong underlying demand commentary -- is exactly the kind of gap that creates both opportunity and risk for anyone trading AI infrastructure exposure right now.
For founders selling into enterprise AI budgets, valuemaxxing behavior means the sales conversation is shifting from "more capacity" to "more measurable ROI per dollar already committed," a materially different pitch that rewards vendors who can demonstrate utilization and outcome metrics rather than just raw model capability. For infrastructure investors, the "almost unlimited" demand framing from executives is worth weighing directly against the 86% utilization data -- both can be true simultaneously if the demand is real but poorly matched to current deployment patterns.
The bear case: executive commentary about "almost unlimited" demand is inherently self-interested -- these are people whose businesses benefit from continued AI infrastructure investment -- and should be weighed against harder, buyer-side utilization data rather than taken at face value. What to watch next: whether enterprise AI budgets in 2027 planning cycles show measurable reallocation toward efficiency and utilization tooling rather than pure capacity expansion, and whether AI-linked equity volatility subsides or intensifies through the rest of the year.