Forget 1999—Dan Ives insists we’re in 1996, when infrastructure spend was ramping but software monetization hadn’t exploded. Snowflake, MongoDB and Palantir are the three consumption-driven names he says still trade like laggards while $3 trillion in AI wallet-share shifts from chips to workloads.
Wedbush senior analyst Dan Ives just ripped up the 1999 bubble playbook. In a pre-earnings note circulated to institutional clients, he told investors to “rip the rear-view mirror off” and view the current AI cycle as mid-1996, when CapEx was surging but software monetization had yet to reach escape velocity. His math: roughly $550 billion in committed AI infrastructure spend is already locked in, yet enterprise consumption revenues are only beginning to inflect.
The $3 Trillion Shift From CapEx to Consumption
Ives’ field checks across Microsoft, Alphabet and Amazon show demand quotes for GPU clusters running 12–18 months deep. The takeaway: hyperscalers are no longer experimenting; they are pre-paying for capacity that must be rented out. That rental phase—measured in on-demand compute, storage and workload execution—is where Snowflake, MongoDB and Palantir capture incremental dollars.
- Total AI ecosystem wallet-share expected to reach $3 trillion by 2028, with 65% moving to consumption-based software pricing models.
- Enterprise customers re-forecast AI budgets up 8× versus 2023 levels, but only 12% of spend is currently live in production workloads.
- Ives sees “a 1996 moment”: hardware bets look frothy, yet software revenue pools are under-modelled by an average 20–35% across the group.
Why Q4 Earnings Are the Prove-It Quarter
Fourth-quarter results kick off the first cycle where AI budgets exit pilot status. Ives expects guidance raises to centre on “consumption dollar per GPU hour” rather than traditional seat licences. That metric favours data-platform vendors that meter usage instead of locking customers into flat subscriptions.
Three Software Plays Still Priced for Disappointment
Snowflake: The Purest AI Consumption Proxy
Snowflake Inc (NYSE: SNOW) earns revenue every time a query touches its data cloud. Ives notes that GPU-generated embeddings can increase query intensity 4–6× versus traditional BI workloads, yet the Street models revenue growth decelerating to 22% in FY-27. His estimate: low-30% is achievable if AI workloads scale as anticipated. At 10.5× NTM sales—a discount to faster-growing SaaS peers—shares price in a slowdown that field checks say is not happening.
MongoDB: The Developer’s Real-Time Data Layer
MongoDB Inc. (NASDAQ: MDB) dominates modern AI stacks that require flexible, JSON-like documents for large-language-model context windows. Atlas cloud revenue grew 36% YoY last quarter, but Ives argues pricing has not yet reflected vector-search premiums now in beta. With 40% long-term growth pencilled in by buy-side models, a ~200 bps beat each quarter through 2026 would drive 25% upside to consensus EBIT estimates without heroic assumptions.
Palantir: From Pilot to Mission-Critical at Scale
Palantir Technologies Inc (NASDAQ: PLTR) already pockets $100 million-plus deals in defense and pharma. Ives cites a recently disclosed $10 billion Army contract that consolidates 75 legacy systems into its AI platform as evidence that procurement officers view Palantir as infrastructure, not software. Margins expand as deployments shift to SaaS; he models 500 bps of operating leverage over the next two years, a trajectory the Street still treats as speculative.
Valuation Disconnect Creates Asymmetric Risk/Reward
Across the trio, the average EV/NTM revenue multiple has compressed from 18× in 2021 to 11× today despite accelerating AI-related bookings. Ives frames the setup as “1996 Cisco”—the year network hardware spend exploded, yet application-layer stocks hadn’t rerated. History shows the software cohort outperformed hardware by 3.2× from 1996 to 2000 once usage ramped.
Bottom Line for Portfolios
If Ives’ timeline is correct, hardware names have captured the early AI dollars, but software will capture the durable ones. Investors still underweight consumption-based models risk repeating the 1990s mistake of owning the pickaxes while missing the gold rush. Earnings guidance that even nudges consumption metrics higher could force a violent multiple re-rating for Snowflake, MongoDB and Palantir before the market looks up.
Stay ahead of every turn in the AI trade—bookmark onlytrustedinfo.com for the fastest, most authoritative analysis on onlytrustedinfo.com.
