SAP Reorganized GTM Around AI That Two-Thirds of SaaS Still Gives Away
Per 2026 SaaS benchmark data, 29% of SaaS companies monetize AI directly. The Information reports, citing server makers, that Nvidia plans to raise flagship AI chip prices 17%.
TL;DR
- SAP's Jan Gilg published an AI-led GTM model built on agents and a shared account layer, while benchmark data shows 67% of B2B SaaS charges nothing for AI.
- GTM is reorganizing around AI almost nobody has priced, and vendors without a data moat or collection path will push that cost onto renewals.
- The test is whether the 67% figure drops under half by mid-2027 as Nvidia's reported 17% chip price increase works through the stack.
SAP just published the org chart, and the industry's pricing pages haven't noticed.
In a first-party piece, Jan Gilg, SAP's global president of Customer Success and an extended board member, described the company's AI-led GTM model: a single AI entry point connecting sales teams to specialized agents for planning, outreach, quoting and content, running on a shared intelligence layer built around each account. The proof point he offers is from a customer, Amadeus, where an autonomous agent reconciled unstructured payment data and cleared around 40,000 incorrect transactions.
Gilg's framing keeps human sellers central, with the agents eliminating friction and the intelligence layer giving teams more context and consistency; this is an augmentation design, not a replacement one. When a mega-vendor's board member publishes the design in public, though, the design stops being a vision statement. It's a benchmark, and mid-market CROs will be asked about it by their boards before the year ends.
Here's the problem with everyone racing to build that org chart: almost nobody has priced the thing it runs on.
The 67% problem
Per 2026 SaaS benchmark data, 92% of SaaS companies have launched or plan AI features, 29% monetize AI directly, and 67% of B2B SaaS companies with AI functionality charge nothing incremental for it. Two thirds of the industry has attached an expanding cost line to its product and declined to attach a revenue line.
That was survivable while the cost line behaved. It's about to stop behaving. The Information reports, citing server makers, that Nvidia plans to raise flagship AI chip prices 17%. A single paywalled source attributing to server makers rather than Nvidia, so treat the exact figure as provisional, but the direction sits on top of every other input trend: inference costs flow downhill from silicon, through the labs, into the per-seat AI features nobody charges for.
Atlassian already showed what the honest version of this trade costs: roughly a third of operating margin, spent deliberately, to give AI away and collect through enterprise expansion. Rovo runs inside more than 80% of the Fortune 500 largely unmonetized, and the number Atlassian actually underwrites is that Rovo customers grow ARR at more than twice the rate of non-adopters.
SaaStr's framing of the decision every vendor now faces is the right one: the question isn't whether you charge for AI, it's which line you want the cost to hit and whether your buyers will tolerate a second meter. Atlassian answered deliberately. That was a priced decision with a collection path. The 67% are mostly running the same margin spend without the path, which isn't a strategy so much as a deferred pricing meeting.
What separates the 29%
The benchmark data's second finding explains who gets to charge. The companies monetizing AI directly are disproportionately the ones with genuine data moats: domain-specific data generic models can't replicate, accumulated longitudinally, tied to measurable outcomes.
Datadog can charge for anomaly detection because it ingests telemetry from 32,700 customers' production environments; the model is the moat's interface, not the product. The valuation market has already priced the difference brutally, with companies above 120% net revenue retention trading around 9.3x EV/revenue against 3.1x for those below 100%. Snowflake, ZoomInfo, HubSpot, Figma and Notion appear on the same list for the same reason: workflow capture that compounds into data nobody else holds.
Which reframes what SAP's org chart actually is. An AI-led GTM org built on a shared account intelligence layer is a machine for accumulating exactly that kind of proprietary data: every plan, quote, outreach and engagement feeding an asset competitors can't download. SAP isn't just reorganizing to use AI. It's reorganizing to own the input that makes AI chargeable.
The case for staying free
The land-grab defense is real. While category share is unsettled, bundling AI free buys adoption, stickiness and training data, and charging early hands rivals a wedge. Atlassian's 35% single-day repricing after proving the model says markets will reward the giveaway when it's working.
The defense has a solvency condition, though: a visible path from consumption to collection. Atlassian could point to enterprise budget it wasn't yet harvesting. A mid-market SaaS vendor bundling free AI into flat seats, with no data moat and no expansion story, meets a 17% input increase with nothing but its own gross margin. For that cohort the giveaway isn't a land grab; it's a countdown.
Free AI needs a collection path, or the 17% lands on renewals
SAP's agents for planning, outreach, quoting and content run on the same inference costs as the free AI features the 67% bundle into flat seats. If a vendor has Atlassian's collection path, the giveaway is a priced decision, paid back by Rovo customers growing ARR at more than twice the rate of non-adopters. If it has no data moat and no expansion story, the 67% unwinding lands on renewal quotes as metered features, a higher seat price or margin the vendor eats until it can't.
Atlassian could afford the trade and still spent roughly a third of its operating margin on it. Its FY27 guidance takes revenue growth down to 13%, while contracted future revenue grew 44%.
Our bet is that by mid-2027 the 67% figure drops under half, because the Nvidia increase works through the stack and makes free AI a board-visible line item.
The next edition of the SaaS benchmark data will show whether the share of B2B SaaS charging nothing for AI, 67% this year, falls under half by mid-2027.
