26. June 2026 · AI
From SaaS to GaaS: Why the Next Software Battle Is Over Outcomes
At GTC 2026, Jensen Huang said something that should keep every software CEO awake at night: “Every SaaS company will become a GaaS company.” The next battle in enterprise software, he argued, will not be fought over features, seats, or the slickest dashboard. It will be fought over outcomes.
I have spent most of my career helping large organizations buy, deploy, and get value out of enterprise software. So let me say plainly what I think this means: the unit of value in our industry is about to change. We are moving from software you operate to software that does the work. That is not a feature release. It is a change to the business model that built the last twenty years of technology.
What Huang actually said
The shorthand is SaaS → GaaS. SaaS — Software as a Service — was built for humans: you log in, you navigate, you click, you act. GaaS — Huang’s framing for an agent- and outcome-driven model — is built for AI agents: you state an intent, the system plans the work, and it delivers a result with minimal human intervention.
Underneath sits a three-layer stack that is worth memorizing, because it is where the value will redistribute:
- Systems of record — the data and the source of truth (your ERP, CRM, ITSM).
- An agent operating system — the layer that plans, orchestrates, and executes.
- Outcome interfaces — where humans set goals and approve results, instead of doing the clicks.
If that is the architecture of the next decade, then the most valuable real estate in software is no longer the user interface. It is the orchestration layer on top of your data — and the trust around it.
The earthquake is in the pricing
Here is the part that breaks existing companies: when agents do the work, the seat-based subscription model stops making sense. You cannot charge per human login for work that no human performs. So pricing shifts from access to results, and the early movers already show what that looks like:
- Intercom’s Fin charges roughly $0.99 per resolution — you pay when a ticket is actually closed, not for a license.
- Salesforce Agentforce now runs three pricing models in parallel: per-conversation (it launched around $2), “Flex Credits” at about $0.10 per action, and traditional per-user licenses (Agentforce 1 lists at $550/user/month). Nobody yet knows which one wins — so they ship all three.
- Sierra, built outcome-native, only gets paid when its agent resolves an issue without a human. It reportedly reached $100M in ARR in 21 months and crossed $150M+ by early 2026.
Intercom’s president, Archana Agrawal, described what outcome pricing did internally: it “exposed every weak link.” Sales could no longer optimize for licenses, customer success could no longer hide behind usage, and the product simply had to work — consistently. That is the quiet brutality of outcome-based pricing: it turns your revenue into a direct bet on whether your software actually delivers.
Why this is harder than a price-list change
It is tempting to treat this as a packaging exercise. It is not. Three structural things move at once:
Margins. Classic SaaS enjoyed ~80% gross margins because serving one more user was nearly free. Agents are not free — every outcome consumes tokens and compute, which is real cost of goods sold. Outcome software will carry a heavier COGS line, and many valuations built on software-style margins will need to be re-underwritten.
Moats. For years the moat was the interface and the switching cost of retraining your people. When agents operate the software, that moat evaporates — nobody is loyal to a UI they never see. The durable advantages become your data, your system of record, and the reliability and governance of the outcomes you produce.
Go-to-market. Land-and-expand by adding seats is dead. You now have to prove measurable ROI — continuously — because the customer can watch, in real time, whether they are getting outcomes worth paying for. Enterprises will stop buying all-in-one suites and start assembling composable ecosystems, mixing best-of-breed agents from different vendors per task.
Who wins, who gets eaten — the 5-to-10-year view
My honest read: SaaS does not die. It bifurcates.
The systems of record — SAP, Salesforce, ServiceNow and their peers — sit on data gravity and deep process context. They are best positioned to win, if they move fast enough to own the agent layer on top of their own data. Their risk is not a startup; it is inertia.
The most exposed players are the thin workflow and UI wrappers — tools whose entire value was a nicer way for a human to do a task. When the human leaves the loop, so does the reason to pay. Expect consolidation here, and expect a wave of outcome-native challengers like Sierra that never carried the SaaS legacy and can simply price on results.
And pricing itself will converge on a hybrid: a platform fee for access to the data and the agent OS, plus outcome metering for the work performed. Pure per-seat will look, by 2030, the way pure perpetual licenses look today.
The real battleground is governance
Here is where I disagree with the chip-centric version of this story. The constraint on GaaS is not GPUs. It is trust.
The moment software stops suggesting and starts acting — closing tickets, reallocating budget, sending the email, committing the order — a new set of questions becomes existential. Who is accountable when an agent gets it wrong? How do you audit a decision a model made autonomously? What are the controls, the approvals, the kill switches? Enterprises will not hand real outcomes to systems they cannot govern. The vendors who win the next decade will be the ones who make autonomous work auditable, controllable, and safe — not just fast. Governance, not raw compute, is the gate.
What this means if you run the technology
For the CIOs and CFOs I work with, this is not a 2026 procurement decision — it is a 5-to-10-year re-platforming. The finance function, tellingly, is the early proving ground: PYMNTS Intelligence reports that 90% of CFOs expect meaningful operational impact from agentic AI (43% “high,” 47% “moderate”), and that AI has already cut chronic cash-flow timing uncertainty from 68% of firms to 17%. The shift is underway whether or not your contracts reflect it.
Four things I would start now:
- Re-underwrite your software portfolio on outcome economics. Ask of every major vendor: what am I actually paying for — access, or results? Where I can measure outcomes, I want to pay for outcomes.
- Treat your systems of record as strategic assets. The agent era rewards clean, consolidated, well-governed data. Do not fragment it across point tools.
- Build the orchestration and governance layer deliberately. This is the new high ground. Owning how agents are approved, monitored, and held accountable is more durable than any single tool.
- Run the founder’s test on your own stack. As one Fortune observer put it: if the company were founded today, knowing what AI can do, how would it solve the customer’s problem? Where your answer contradicts your current tools, you have found your roadmap.
Huang’s line will be quoted for years, but the substance is simple. We are leaving an era where we paid for the tool and entering one where we pay for the job done. That rewards whoever can deliver reliable outcomes on trusted data — and it punishes everyone who was selling the clicks in between. The companies that internalize this early will not just survive the transition. They will write its rules.