The “AI has killed software” narrative has a handful of very loud beneficiaries and a lot of quiet evidence against it. The companies that will survive the next five years are the ones that refuse to treat the hyperscalers as the new gods.
Whenever I make an affirmation, I like to do my research first, and not to sound like a LinkedIn post. I wish more people in this industry did the same, as there is a prevailing mood where we think that big numbers are the whole story.
When the Black Death came among us, people probably thought it was the end. When wars came to our societies, people thought it was the end. Yet, in a strange way, we have a natural power to overcome obstacles and turn change to our advantage.
When AI started to infiltrate our work, and later our personal lives, a large group of people declared that “AI will replace people,” that this technology, not even particularly new, would conquer our brains, hearts, and work, and lead us where it wanted.
Yet we are still working; people are still writing, thinking, creating, building.
In the last two years, more and more people have been saying that “SaaS is dead.” Of course, this phrase came from someone’s mouth, someone with enough influence to shape general opinion, and everybody was already in black, ready for the funeral.
In August 2024, Klarna’s chief executive, Sebastian Siemiatkowski, sat on an earnings call and mentioned, almost in passing, that the Swedish fintech had “shut down Salesforce.” Workday was next.
Klarna would build its own AI-driven replacements, a lightweight stack unshackled from the bloat of traditional enterprise software. The quote moved markets. Articles followed with headlines about the death of SaaS. Salesforce’s Marc Benioff, on stage at Dreamforce, was asked to respond to a customer who had apparently decided the future was AI and the past was his product. He looked, by his own admission, embarrassed.
Six months later, Siemiatkowski quietly clarified what had actually happened. Klarna had not replaced Salesforce with AI. It had replaced Salesforce with other SaaS: Deel for HR, third-party tools for CRM, the Swedish graph database Neo4j for data consolidation.
Klarna still uses Slack, which is still a Salesforce product. Siemiatkowski himself admitted on X that he was “tremendously embarrassed” by how the story had spiralled.
“No,” he wrote, “we did not replace SaaS with an LLM.”
This is the single most instructive story in enterprise software of the past two years. The distance between what was said and what was done reveals the mechanics of the entire “SaaS is dead” narrative. The headline travelled. The correction did not.
An industry of analysts, venture capitalists, and foundation model CEOs built a year of marketing on the louder half.
Start by asking who gains from the story that software-as-a-service is being replaced by artificial intelligence, because the answer is surprisingly narrow. The hyperscalers do, because AI workloads justify the $660 to $690 billion in capital expenditure the five largest US cloud and technology companies have committed for 2026, according to Futurum Group analysis, nearly double the previous year.
The foundation model labs benefit, because every dollar of enterprise software spend redirected to their APIs validates valuations that are otherwise difficult to defend. OpenAI ended 2025 at around $20 billion in annual recurring revenue. Anthropic crossed $9 billion in January 2026. These are genuinely large numbers. They are also, respectively, about three per cent and a little over one per cent of the hyperscaler capex being spent to serve them.
The venture capitalists benefit because their portfolio repricing depends on the narrative that AI-native companies will outrun the incumbents they once funded. And Nvidia, supplier and financier of the boom, benefits until it no longer does.
In March 2026, CEO Jensen Huang confirmed that his recent investments in OpenAI and Anthropic would likely be the last. The circular financing, Nvidia invests in OpenAI, OpenAI buys Nvidia chips, had reached the point where even the chipmaker was ready to stop calling it a virtuous cycle.
MIT’s Michael Cusumano, quoted by Bloomberg, put the arithmetic bluntly: “Nvidia is investing $100 billion in OpenAI stock, and OpenAI is saying they are going to buy $100 billion or more of Nvidia chips.”
You could call that demand. You could also call it bookkeeping.
The 95% number that should have ended the hype
The harder question is whether any of this is producing business results. Here the data is less generous than the pitch decks.
In July 2025, MIT’s Project NANDA published “The GenAI Divide: State of AI in Business 2025”, based on 150 executive interviews, 350 survey responses, and analysis of 300 public AI deployments. Its headline finding: despite roughly $30 to $40 billion in enterprise generative AI spending, 95% of pilots delivered no measurable impact on profit and loss. Only 5% reached production.
The response from the industry was not to recalibrate. It was to argue that the wrong metric was being used. UC Berkeley published a rebuttal suggesting ROI was an “industrial-era” measurement unsuited to a “cognitive-era transformation.”
This is what every hype cycle says in its late phase, that profit is a distraction, that what is being built is too large for ordinary standards. The same argument was made about WeWork, the metaverse, and blockchain.
Each time, the underlying assumption was that the people with capital and megaphones understood the future better than the people actually trying to run a business.
The 5% of AI projects that did succeed, MIT found, shared specific traits. They were built by specialised vendors, not attempted internally. They focused on back-office automation rather than sales theatre. They integrated deeply with existing workflows. Over half of enterprise AI budgets, meanwhile, were going to sales and marketing tools where ROI was lowest.
This is not a revolution sweeping through the enterprise. It is a lot of companies buying demo-friendly products that do not produce returns, while a minority does the unglamorous integration work that quietly extracts value.
The collapse that did not collapse
Stil, I have to admit that there are genuine signs of stress in the SaaS market. In February 2026, roughly $285 billion in market value evaporated from software stocks in a single trading session, what Wall Street christened the “SaaSpocalypse.”
ServiceNow fell 7%. Intuit dropped 11%. LegalZoom lost nearly 20%. Salesforce is down approximately 30% year-to-date. The business rationale, that per-seat pricing starts to collapse when one employee with AI tools can do the work of five, is not wrong.
But Bain & Company, looking at the broader record, has offered a useful correction: technological transitions rarely produce extinction.
They produce heterogeneity. Desktop survived mobile. Cloud did not kill on-premise so much as push it into specialised niches. The history of software is a history of layers accumulating, not replacing.
SaaS vendors are becoming agent-orchestration platforms. Salesforce has Agentforce. HubSpot has AI tools. Snowflake partners with Anthropic. The incumbents are being forced to adapt, but adaptation is not death.
IDC’s European practice framed it precisely in February: “SaaS is not dead, but it is metamorphosing.”
Pricing shifts towards outcomes. Interfaces become more agent-driven. But the real business logic, the auditing, versioning, compliance, and data gravity, remains where it was. The transformation is real. The extinction event is marketing.
The new gods are not new
Every major technology wave produces a brief period in which the companies at its centre are treated as reinventors of reality. For the cloud, it was AWS. For mobile, Apple. Before that, Microsoft.
The rhetoric around big techs like Nvidia, OpenAI, Anthropic, Meta, and xAI has the same cadence: they are building the new infrastructure of civilisation, rewriting how humans work, inevitable. There is a grain of truth in it. AI, and agentic AI in particular, is a real technological step.
The companies most likely to thrive are the ones already disciplined enough to recognise the pattern. Every enterprise that survived the dot-com crash, the mobile transition, and the cloud migration did so by adopting what was useful and ignoring what was hyped, by measuring outcomes against costs, by refusing to treat platform vendors as infallible.
The companies that went under bought the whole story: that their customers would wait while they rebuilt, that the new paradigm would reward early and total commitment.
We reported in February on a pattern now visible across dozens of SaaS companies between $20 million and $80 million in ARR: shipping AI features while net revenue retention quietly collapses.
Eighteen months after going “AI-first,” one company watched its NRR drop from 108% to 94% and lost $2.8 million in renewals, not because the product got worse, but because everyone was building the future and nobody was watching the present. The AI features were legitimately good. The existing customers churned anyway.
None of this is an argument against AI. Previous AI cycles ended with research freezes, shuttered startups, and survivors who had been quietly doing useful work while everyone else claimed the moon. This cycle will likely end similarly.
Some hype will turn out to be real. Most revenue projections will not. A handful of current “AI-native” startups will become durable businesses. Many will be absorbed or exposed as wrappers.
The companies that come through refuse both extremes. They do not miss the trend, because dismissing AI in 2026 is as serious a strategic error as dismissing mobile was in 2010. And they do not drown in it. They do not empty their engineering teams into AI-first rebrands while their existing revenue base walks out the door. They do not treat the big tech companies as gods, but as what they are: very large commercial entities with very specific interests in what you believe about the future.
Klarna, for the record, is still paying for SaaS. It is also still paying OpenAI. This is probably the honest shape of the future: not the death of anything, but a quieter rearrangement in which the winners are the operators who kept their feet on the ground while everyone else was watching the sky.
The funeral for SaaS has been extremely well-attended. The corpse, on closer inspection, is still breathing.
Now we don’t mean to beat up on the folks at NASA, but browsing through these images, we couldn’t help but be reminded of an article we saw on PetaPixel that discussed the space agency’s haphazard approach to sharing images online.
That might not seem unreasonable at first, but it actually represents a pretty serious hurdle for many users. Consider that the whole advantage of these panels is the convenience: you can simply open the box, plug them in, and start collecting energy. But if you need to install a special outlet, potentially requiring an electrician, the whole concept falls apart. Expect to hear more from us on this particular subject as it develops.
You must be logged in to post a comment Login