Consumption is the model — and the bull case
Unlike the seat-based application layer, these four bill by usage: MongoDB's Atlas scales with database workload, Databricks bills compute in DBUs, Cloudflare meters network and edge traffic, Palantir charges for its ontology-and-AIP platform. The structural advantage is that revenue grows automatically as the customer grows, which shows up as net-retention rates the rest of software envies — Palantir at 150%, Databricks >140%, MongoDB ~120%. Gross margins run 74-82%. The bull case writes itself: own the layer applications are built on, and you compound with every workload the customer adds.
The multiple is the entire risk
What unites the four more than the model is the price. Palantir trades near 70× sales, Cloudflare ~40×, Databricks ~25× its run-rate at a $134B private mark, MongoDB richly for a database. Read the weighings together and each one's controlling question is some version of "can the growth, the moat, and the price all be true at once?" — Palantir's words exactly. These are not businesses where the bear disputes the growth (it's real: Palantir +56%, Cloudflare +30%, Databricks >65%). The bear disputes whether any growth rate justifies the multiple, because the price already embeds years of it continuing flawlessly.
Every moat has the same shape of doubt
The bear cases rhyme: the moat is real but maybe not permanent. MongoDB's document-database lead is being eroded at the edges by PostgreSQL with pgvector; Databricks' lakehouse advantage faces open table formats commoditizing the layer; Cloudflare's edge network is extending into AI compute against hyperscalers; Palantir's ontology moat rests heavily on government and a software-vs-services debate. In each, the thing the multiple pays for — durable lock-in — is exactly the thing a well-funded open-source or hyperscaler alternative is working to dissolve. The cluster's sharpest read: consumption models are wonderful when the customer can't leave, and these multiples assume they can't.
The cash-flow tell separates them
One number cuts through the multiples: who actually converts to cash. Palantir is durably GAAP-profitable; MongoDB's free cash flow is inflecting; Cloudflare is still GAAP-unprofitable at ~40× sales, leaving no room for a stumble; Databricks reports a lower ~74% gross margin on its fast-growing AI workloads, hinting the mix shift compresses the very margin the valuation assumes. Reading the four together, the safest seat isn't the highest grower — it's the one whose consumption revenue is already throwing off real cash, because that's the only thing that defends a premium multiple when growth eventually decelerates.
Where they agree — and where they split
All four agree the data-and-AI layer is a great place to sit, that consumption compounds with the customer, and that retention proves the lock-in is working today. They split on durability and price: the bull reads the moat as structural and the growth as years from maturing; the bear reads identical evidence as a temporary lead at a permanent-moat price, with open-source and hyperscalers closing in. The demand for data and AI infrastructure isn't the question any of these studies doubts. Whether the lock-in lasts as long as the multiple needs it to is.