Quick Answer: Larry Ellison rejects the idea of a SaaS apocalypse, the fear that businesses will use AI to build their own applications and stop buying software-as-a-service products. His position is that AI will improve software rather than make it obsolete. Whether he's right depends on how much internal engineering a company can realistically absorb, and on what it costs to maintain that code.
Oracle's co-founder just pushed back on one of the loudest arguments in enterprise technology. If AI can write working code in an afternoon, why would a company keep paying per seat for a CRM, an HR platform, or a finance suite? Ellison's answer, as Business Standard reported, is that AI makes software better, not redundant. That's a bet worth examining, because the "SaaS apocalypse" framing has real money behind it.
Where the SaaS Apocalypse Idea Came From
The argument goes like this. Large language models and agentic coding tools have made it far cheaper to produce a working application. A mid-sized company with a few capable engineers could, in theory, rebuild a workflow tool internally rather than renew a subscription. If that happens across thousands of companies, the revenue base of seat-based SaaS vendors starts to look fragile.
Investors took this seriously. Software stocks sold off repeatedly during 2026 as the market tried to price in the risk that AI agents would compress the number of human seats that vendors bill for. Most of the nervousness centred on companies whose revenue depends on counting users inside a product.
Here's the part that matters: the fear is not really about code generation. It's about switching costs. SaaS has always been partly a bet that building and maintaining software is harder than buying it. AI attacks the building part. It doesn't automatically touch the maintaining part.
Why Ellison Thinks AI Helps Incumbents
Ellison's counter-argument, as reported, is that AI improves the products incumbents already sell. Oracle's pitch centres on embedding AI into its existing stack, including Oracle Fusion Cloud Applications for ERP, HCM, and supply chain, and Oracle Cloud Infrastructure for the compute underneath. The logic is that a vendor with deep data and workflow integration can deliver AI features a customer would never assemble alone.
This is where most commentary goes wrong. People picture a company replacing its ERP with a weekend project. That's not how enterprise systems fail or change. The realistic failure mode is more mundane: a team builds a useful internal tool, it works for eighteen months, then the engineer who wrote it leaves, the upstream API changes, and nobody owns the audit trail. When that happens, the company has quietly taken on a maintenance liability it didn't price in.
That's the counter-intuitive finding worth holding onto. Building software is cheap in a way that makes it look like the expensive part is over. In practice, the expensive part was always the ongoing work: security patches, compliance changes, edge cases, uptime, and integrations. AI lowers the cost of the first version. It does much less to lower the cost of the fifth year.
Build, Buy, or Hybrid: A Practical Comparison
The honest answer for most organisations isn't "buy everything" or "build everything." It's a mix, and the right mix depends on what the software actually does for the business.
| Approach | Upfront cost | Ongoing burden | Best fit |
|---|---|---|---|
| Buy a SaaS product | Low, subscription-based | Vendor handles patches, uptime, and compliance updates | Regulated processes such as payroll, tax, and financial close |
| Build in-house with AI coding tools | Low for a prototype, rising as scope grows | Your team owns security, integrations, and on-call support | Narrow internal workflows that differentiate your business |
| Hybrid: buy the core, extend with custom AI agents | Moderate | Split between vendor and internal team, with clear ownership needed | Most mid-to-large enterprises today |
The hybrid row is where most real activity is happening. Companies keep the system of record and extend it with custom logic, often through APIs. That's also the model Oracle and its competitors are selling.
Here's where the nuance matters. Seat-based pricing is genuinely vulnerable. If an agent handles a task that previously needed five analysts logged into a tool, the vendor loses four seats. Several SaaS companies have already started experimenting with consumption-based or outcome-based pricing for exactly this reason. Ellison's argument works better for vendors who price on platform value than for those who price on logins.
What Could Make Ellison Wrong
Ellison's position has real weak points, and a responsible reader should weigh them.
First, the cost of building has fallen faster than the cost of maintaining. That gap is the core of the disruption thesis. If AI makes internal maintenance cheaper too, through automated testing, code review, and documentation agents, the incumbents' moat narrows considerably.
Second, procurement behaviour can shift before technology does. Many CIOs are under pressure to cut software spend. A vendor that can't demonstrate AI value in renewal conversations may lose budget even if its product is technically superior.
Third, nobody has clean, public data yet on how often companies actually replace purchased software with internal builds. Anecdotes are everywhere. Systematic evidence is thin, so any confident forecast, from either side, should be treated with caution.
That said, "AI will improve software" and "AI will kill SaaS" can both be partly true. Incumbents with proprietary data, regulatory certifications, and deep workflow integration are better placed than point tools with thin feature sets.
How to Assess Your Own Exposure
If you're a buyer, the useful question isn't whether the SaaS apocalypse is coming. It's which of your subscriptions depend on process knowledge that only the vendor holds, and which are essentially data storage with a login screen. The first group is harder to replace. The second is where build-vs-buy decisions are getting interesting.
If you're a builder or an investor, the question is what your product does that an agent-plus-database couldn't do in a quarter. Integrations, compliance, audit trails, and trusted data are harder to replicate than interfaces. That's the defensible ground.
Most people stop here, but the decision itself is more specific than the headline. Ask your team to list the five software tools with the highest annual spend. For each one, note who maintains it today and what would break if that person left. You'll learn more from that list than from any forecast about the SaaS apocalypse.
Frequently Asked Questions
What is the SaaS apocalypse?
The SaaS apocalypse is a term for the fear that AI will let businesses build their own applications instead of buying software-as-a-service products. It's most worrying for vendors that charge per user, since fewer human seats means lower revenue. Larry Ellison has publicly rejected the idea, arguing that AI will improve existing software rather than replace it.
Will AI replace SaaS companies?
Not entirely, and probably not evenly. AI reduces the cost of writing code, but it doesn't remove the ongoing work of security, integrations, compliance, and maintenance. Vendors with proprietary data and deep workflow integration are better placed than thin point tools. Seat-based pricing models face the most pressure.
How can a company build its own software with AI instead of buying SaaS?
Start with a narrow internal workflow that differentiates your business, not a core system of record like payroll or ERP. Prototype with AI coding tools, then price the full cost, including security reviews, on-call support, and the engineer who maintains it after the first year. If the total cost still beats the subscription, build it.
Why do investors fear AI disrupting SaaS?
Many SaaS companies charge per user, so AI agents that reduce the number of people doing a task can shrink revenue. Investors also worry that cheaper software creation lowers switching costs. Those fears have driven volatility in software stocks, though whether they're justified depends on each company's pricing model and product defensibility.
Is Oracle's view on AI and software credible?
Oracle has a strong position in enterprise databases, cloud infrastructure, and applications, which gives it a real stake in the outcome. That doesn't make its argument wrong, but it does mean you should weigh it as a vendor's perspective. Independent evidence on how often companies replace purchased software with internal builds is still limited.
Pick one software subscription your team pays for this month and find out exactly who maintains it and what it would cost to replace. Share this with someone in your organisation who's debating whether to build or buy, and let the maintenance numbers, not the headlines, drive the decision.