Scott Hsu, a Technical Program Manager in the Office of the CTO at Rippling, argues that the software industry is completing a cycle that began two decades ago — and that AI has just made the outcome inevitable.
History does not repeat in software. But it does rhyme, and right now it is rhyming loudly.
Twenty years ago, a small number of software giants — SAP, Oracle, Microsoft — controlled nearly everything businesses ran on. They were the default. Then the cloud arrived, and their advantage collapsed almost overnight. Building on cloud infrastructure, startups could peel off individual features from these monoliths — expense management, recruiting, contract signing, device management — and turn each one into a standalone product that did one thing better than the giant ever had. The era of point-solution SaaS was born.
It worked spectacularly. Thousands of narrow, focused software companies were built and funded and, in many cases, made spectacularly valuable. The conventional wisdom calcified: focus on one thing, go deep, win your category.
Scott Hsu watched that era end. As a Technical Program Manager in the Office of the CTO at Rippling, a workforce management platform valued at $16.8 billion, he has had a front-row seat to the forces now running in reverse. "The unbundling happened because the cloud made it possible to build one thing fast and cheaply," Hsu says. "AI has made it possible to build ten things fast and cheaply. The logic that created thousands of point solutions is now destroying most of them."
Why the Unbundling Always Had a Shelf Life
To understand why rebundling is happening, it helps to understand why unbundling happened in the first place — and why its advantages were always temporary.
When cloud infrastructure matured in the 2000s and 2010s, it dramatically lowered the cost of building and deploying software. That cost reduction was the wedge. A small team could build a focused product, get it to market, and iterate faster than an SAP or Oracle ever could. The moat was speed and specialization: you could go deeper on expense reports than any mega-vendor bothered to go, because that was all you were doing.
But that wedge only works as long as the incumbents can't move fast. And it only sustains value as long as customers are willing to pay a premium for best-in-class depth in each individual category.
Both conditions have now expired. AI has collapsed the cost of building features — not to the low levels that cloud infrastructure achieved, but to near-zero. Mary Meeker's Bond Capital AI Report declared in 2025 that the era of the SaaS point solution is approaching its end, with horizontal platforms possessing deep data moats positioned to dominate. When an AI agent can replicate the core function of a point solution — categorizing expenses, drafting contracts, routing support tickets — at a fraction of the cost and with minimal implementation friction, the depth advantage disappears. What remains is the distribution problem: which companies already have the customer relationships, the data, and the platform infrastructure to absorb those use cases when the specialized vendor stops being worth the additional contract?
The Three-Layer Stack That Replaces the Fragmented One
This architecture has a direct implication for the middle of the SaaS market. Point solutions that don't sit at the systems-of-record layer and don't have agent infrastructure are competing for a slice of the outcome interface — the most crowded, least defensible position in the new stack. AlixPartners predicts M&A in the software industry will surge 30 to 40 percent year-over-year in 2026, driven precisely by mid-market companies that can't independently fund both an AI transformation and competitive scale. The consolidation isn't a correction. It's a structural consequence.
"The companies getting acquired or going silent right now built genuinely good products," Hsu observes. "The problem isn't the product. It's that the product exists in a category that a broader platform will eventually serve well enough — and well enough, combined with a procurement relationship that's already in place, beats best-in-class every time."
The CAC Arithmetic Nobody Talked About
There is a financial dimension to the rebundling story that rarely gets discussed alongside the AI narrative, but which may be more decisive in the long run.
Point-solution SaaS companies face a structural disadvantage in customer acquisition that predates AI and is now being fatally compounded by it. Every time a point-solution company sells, it is selling to a net-new customer relationship — the most expensive transaction in software. There is no existing foothold, no established trust, no opportunity to trade on the friction of switching costs already in place. The entire cost of customer acquisition is loaded onto a single product.
Compound platforms — those that sell multiple products into the same customer — can amortize that acquisition cost across every subsequent product. The first sale is expensive. The second, third, and fourth sales are dramatically cheaper, because the relationship and the data integration already exist. This arithmetic produces compounding CAC advantages that widen over time, independent of any AI capability.
When AI is added to this dynamic, the gap becomes a chasm. A compound platform building AI features benefits from proprietary behavioral data across every workflow it touches — payroll, benefits, device management, spend, recruiting — in ways a single-product company simply cannot replicate. As Theory Ventures' Tomasz Tunguz recently observed, buyers no longer want to assemble a best-of-breed stack when models change every six weeks. They want a platform they can trust for three to five years. Trust, at enterprise scale, lives in the companies that already process your payroll, manage your identities, and run your compliance.
What the Cycle Looks Like From Inside It
Hsu is careful not to frame the rebundling as a story about winners and losers — at least not in the way the current market narrative tends to. "There will be enormous amounts of value created in the next decade of software," he says. "The question is just where it accrues. And I think the answer is increasingly clear: it accrues to the platforms that own the data layer and have the distribution to deploy AI against it."
Gartner has predicted that by 2030, 35 percent of point-product SaaS tools will be replaced by AI agents or absorbed within larger platform ecosystems. The more provocative observation is what that implies about the other 65 percent: not that they survive unchanged, but that those which do survive will have made themselves indispensable at the data layer — the only position in the new stack that AI cannot simply replicate or route around.
The cloud unbundled software by making it cheap to build one thing well. AI is rebundling it by making it cheap to build everything — which means the only things worth building independently are the ones that can't be absorbed. The cycle that began in 2004 with Salesforce's rise and the slow dismemberment of SAP's monopoly is completing itself. The companies that understood it early enough to position accordingly will be worth a great deal. The ones that didn't may not get another chance to.
"We've seen this film before," Hsu says. "It just runs faster now."
This article is published under HackerNoon's Business Blogging program.