Snowflake's finance chief told analysts the company's AI tools are boosting productivity through what he called a "step function change," even as the stock posted its best single-day gain ever on the back of a $6 billion compute commitment to Amazon, according to CNBC's report on Snowflake's May 2026 rally. CNBC noted the same results cast doubt on a narrative it called the "SaaSpocalypse," the fear that AI coding tools let enterprises build in-house what they used to buy by seat. The tension is real, but the evidence for who wins it is thinner than the vendor earnings suggest.
The vendor results are the strongest data in the file and also the most commonly overinterpreted. What the numbers actually show is platform concentration, not customer defection.
Microsoft's Q3 FY2026 results filing put its AI business at a $37 billion annual run rate, up 123% year-over-year, inside overall revenue of $82.9 billion. Palantir's Q2 2026 results filing reported U.S. commercial revenue growth of 149% year-over-year and raised full-year guidance to 82% growth.
Three implications for the next planning cycle:
The meeting question is which vendor relationships are infrastructure the company is building on, and which are subscriptions it could plausibly replace within a budget cycle. The two categories are being priced identically. They will not stay that way.
The build-versus-buy calculus is shifting for internal tools, according to TechRadar's piece on forces reshaping enterprise SaaS, which names AI-assisted development as one of three forces pressuring the traditional vendor relationship.
Vendors feel this differently depending on where they sit in the stack. Data platforms selling infrastructure to build on are fine, arguably better than fine. Snowflake's stock jumped 36% in a single session after it guided up and announced $6 billion in AWS compute spend, with CFO Brian Robins telling analysts that Cortex Code is producing a "step function change" in AI revenue potential, per CNBC's report on the Snowflake rally.
Palantir, selling both platform and outcome, posted 93% year-over-year revenue growth and a 149% jump in U.S. commercial revenue in Q2 2026, with CEO Alex Karp calling the quarter "otherworldly" in the company's own release, filed with the SEC — Palantir's Q2 2026 results filing.
Cortex Code, Claude Code, and equivalents are, functionally, the assembly labor a SaaS vendor used to charge margin on.
ServiceNow and Snowflake's Zero Copy integration, letting data move between platforms without duplication, is the connective tissue that internally-built tools need to avoid becoming another silo — a partnership the companies announced directly, not one this reporting can attribute an adoption number to.
The bottleneck migrates. It used to be developer time.
Snowflake closed up 36% on May 28, 2026, its best single-session move on record, after telling analysts it would commit $6 billion to AWS compute capacity and posting a first-quarter beat, according to CNBC's report on the Snowflake rally. CNBC's report noted the results "cast doubt on recent worries that AI tools are contributing to a 'SaaSpocalypse.'" That framing is worth sitting with, because it concedes the SaaSpocalypse thesis existed before it declares the thesis wrong. The evidence for either verdict is thinner than either camp admits.
The bear case is simple and has a name attached to a mechanism. AI coding tools are cutting the cost of building software in-house, which erodes the reason enterprises pay a per-seat toll to a vendor for something they could now assemble themselves. The bull case, argued by Citi's Tyler Radke after a software-sector selloff, is that the same AI wave is driving incremental consumption of hyperscale data infrastructure that no enterprise can replicate on its own. The selloff was "a good buying opportunity," Radke said, provided an investor is "selective" and prefers "names exposed to hyperscale data volumes," per Citi Research's commentary reported by The Register's SaaS coverage. Both cases can be true for different layers of the stack. Neither is proven by the numbers this reporting turned up.
Start with what the numbers actually show, because the vendor results are the strongest data in the file and also the most commonly overinterpreted. Palantir's second-quarter 2026 filing reported U.S. commercial revenue growth of 149% year over year and total revenue growth of 93% year over year, with a Rule of 40 score — growth rate plus profit margin, the industry's shorthand for efficient scaling — of 155%, according to Palantir's Q2 2026 earnings release filed with the SEC.
Microsoft's fiscal third-quarter filing put its AI business at a $37 billion annual revenue run rate, up 123% year over year. That figure sat inside total revenue of $82.9 billion, up 18%, per Microsoft's Q3 FY2026 earnings release filed with the SEC.
Those figures describe demand for AI infrastructure and platforms. They say nothing about what enterprise buyers do with that infrastructure once purchased, and neither filing addresses whether customers are building tools they intend to keep in-house, resell, or eventually cancel a SaaS contract to replace.
That gap matters because it is where the article's thesis lives or dies, and the sourcing available does not close it. The most direct claim that AI is pushing enterprises toward in-house builds instead of SaaS subscriptions comes from a TechRadar piece on forces reshaping enterprise SaaS, but the retrievable content is largely site navigation rather than reported substance, and it should be weighted accordingly, per TechRadar's piece on forces reshaping enterprise SaaS. A thesis this consequential for subscription-model valuations deserves a named enterprise that tried it, and none surfaced.
What did surface is a live disagreement about whether centralizing software production inside an organization — the "software factory" model — works at all, and the disagreement is generational. BMC's 2020 framing held up Google and Netflix as proof that companies could push "quality software to market sooner" by industrializing delivery pipelines, per BMC's 2020 blog post on becoming a software factory. Six years and one AI cycle later, Coder's Amanda Phelps used the same term to describe a failure mode inside the Defense Department and intelligence community. She wrote that centralized factories like Platform One and Kessel Run had "quietly become the thing slowing development down," calling instead for smaller, decentralized teams, per Coder's essay on Intelligence Community News. Phelps was writing about classified government environments with authority-to-operate constraints that have no enterprise IT analog, and her argument should not be read as a verdict on corporate data platforms. But it is a useful corrective to any assumption that centralizing build capacity is automatically the efficient move once AI lowers the cost of writing code. Whether that critique extends to commercial data platforms is untested by anything in this reporting.
The infrastructure that would make internal-tool commercialization technically plausible is being built regardless of whether anyone commercializes anything. Snowflake and Palantir announced a partnership giving Foundry and Snowflake's Iceberg Tables "bidirectional, zero-copy interoperability," naming Eaton as a joint customer using it to unify data feeding AI agents, according to Eaton's chief data officer Ross Schalmo. What that buys an enterprise is the ability to build an AI application against governed data without duplicating it into a separate pipeline first, which lowers the fixed cost of prototyping. It does not answer what happens after the prototype works. No source in this reporting documents a company taking that next step, and the honest conclusion is that the plumbing is arriving well ahead of any evidence it is being used the way the thesis requires.