I’m the ‘dean of innovation’ and I know that nobody can agree what ‘innovation’ actually means

The article offers a framework for deploying AI innovation inside large organizations, arguing that executives often use “innovation” to mean different things (adoption, optimization, invention, transformation) and create “false alignment” around mismatched goals (e.g., cost reduction vs new revenue). It emphasizes that generative AI often cannot be justified with reliable 5-year ROI data ex ante, recommending instead the “next experiment” approach—cheap, fast tests that generate the evidence needed to scale. Overall, it is largely conceptual with no direct financial figures, so near-term market impact is limited.

Analysis

The immediate market implication is a widening gap between AI spend that is easy to announce and AI spend that actually converts into durable revenue. That favors the infrastructure layer: even if large enterprises slow broad deployment, they still need inference/training capacity, data movement, and integration plumbing, which keeps demand concentrated in NVDA. The more interesting loser is the enterprise software stack that is trading on fast monetization from seat-based AI features; if buyers reclassify those projects as workflow experiments rather than strategic rollouts, revenue recognition can slip by 1-3 quarters and multiple expansion can stall.

For MSFT, the risk is not product relevance but valuation timing. Copilots can show usage while pricing power lags, so the key question is whether AI features expand net retention or simply repackage existing spend; that will show up in commercial cloud mix and attach rates, not in headline pilot counts. UPS is a second-order beneficiary only if AI becomes measurable cost takeout in routing and labor planning; otherwise this is too incremental to justify a structural rerate. DPZ is mostly a reminder that mature operators can use AI for process gains, but that does not create a standalone stock catalyst.

The contrarian view is that consensus may be overestimating how quickly AI turns into revenue while underestimating how much budget migrates from big-bang deployments to small experiments. That is bullish for compute demand and bearish for vendors selling "transformational" narratives at rich multiples. What would falsify this view: a clear acceleration in MSFT AI attach rates and Azure AI consumption over the next 1-2 quarters, or conversely NVDA data-center growth slowing despite continued enterprise AI commentary.

Key Decisions for Investors

• Maintain or add to NVDA on pullbacks over the next 1-3 months; the best risk/reward is that experimentation drives persistent infrastructure demand even if enterprise rollout velocity disappoints.
• Relative-value trade: long NVDA / short MSFT into the next 1-2 earnings cycles. Thesis: compute monetization stays ahead of software monetization; exit if MSFT shows sustained Copilot seat conversion or AI contribution to Azure growth inflects faster than expected.
• Do not chase broad AI-premium expansion in enterprise software names on pilot announcements alone; require proof of pricing power, net retention, or incremental ARR before paying a higher multiple.
• UPS is a watch item, not a high-conviction trade: look for evidence of operating-ratio improvement from automation before underwriting any AI-driven rerating; absent that, treat the story as marginal and fade hype-driven strength.
• Set an alert on NVDA data-center growth and MSFT commercial cloud/AI attach metrics for the next two quarters; those are the cleanest falsifiers of the current compute-vs-monetization divergence.