The GaaS IPO Watch List: Which Agent Companies Could Actually Go Public

Table of Contents

• Why an IPO Watch List for GaaS Is Different
• The Public-Market Bar a GaaS Company Has to Clear
• Tier 1: The Realistic Near-Term Candidates
• Tier 2: The "Needs Two More Years" Group
• Tier 3: Probably Gets Acquired First
• The Metrics That Will Make or Break the S-1
• What the First GaaS IPO Will Signal
• Insights Most People Overlook
• References

Why an IPO Watch List for GaaS Is Different

Most IPO watch lists are exercises in pattern-matching against the last cycle. You take the SaaS playbook, find the companies at $200M+ in ARR growing 40% with improving margins, and rank them. That heuristic mostly works for software because software economics are stable and well understood.

Agentic AI-as-a-service breaks the template in a way that matters for public-market readiness. A GaaS company doesn't sell seats; it sells completed tasks or guaranteed outcomes. Its cost of goods sold isn't a fixed hosting bill that shrinks with scale -- it's a variable inference cost that moves every time a foundation-model provider changes pricing or the company swaps in a more capable, more expensive model. That single structural fact reorders the entire watch list. The companies that look most impressive on a growth slide are often the ones with the shakiest path to the gross margins a public investor will tolerate. We've covered the underlying economics in our pieces on the burn-rate problem and revenue durability stress tests; the IPO question sits downstream of both.

So this watch list isn't ranked by valuation or hype. It's ranked by something closer to "could this company's finance team survive six months of S-1 drafting and a roadshow without the story falling apart." That is a much shorter list than the funding headlines suggest.

The Public-Market Bar a GaaS Company Has to Clear

Before naming names, it's worth being concrete about the bar. Public markets in this cycle are not the 2021 markets that rewarded growth-at-any-cost. The IPO window that reopened reluctantly favors companies with a credible line to profitability and revenue that doesn't evaporate when a customer's pilot budget gets cut.

For a GaaS company specifically, four things have to be true:

A durable revenue base. Usage-based and outcome-based revenue can be spectacular on the way up and brutal on the way down. Underwriters will want to see that a meaningful share of revenue is contracted, committed, or sticky enough to model. The recurring-versus-consumption tension is the heart of the question we raised in the revenue-quality question.

Gross margins that look like software, not like a services firm. If inference costs eat 50-60% of revenue, the company is going to get valued like an IT-services rollup, not a software company. The market's tolerance here is unforgiving, and it's why margin structure, more than growth, decides which companies make this list.

Net revenue retention above roughly 120%. Land-and-expand is the whole bull case for vertical agents -- you start with one workflow and the agent metastasizes across the org. If that expansion isn't showing up in the retention number, the durability story is hollow.

A real moat beyond a model API. Anyone can wrap a frontier model. The companies that clear the bar have proprietary data, workflow lock-in, regulatory positioning, or a distribution advantage that doesn't disappear when the next model drops. Goldman Sachs and others have repeatedly flagged that the durability of AI-driven revenue is the open question hanging over the entire investment thesis.

Clear these four and an IPO is conceivable. Miss two of them and the company is an acquisition target wearing an IPO costume.

Tier 1: The Realistic Near-Term Candidates

The honest framing here is that no pure-play "agent company" is filing an S-1 next quarter. But a handful of categories are close enough to be worth watching seriously.

Agent infrastructure and orchestration platforms

The "picks and shovels" layer is the most IPO-ready slice of GaaS, and it isn't close. Companies selling the orchestration, observability, evaluation, and reliability tooling that other agent builders depend on have classic software economics: high gross margins, recurring contracts, and revenue that doesn't swing with any single customer's outcome bets. This mirrors the dynamic we explored in the "picks and shovels" funding thesis -- the infrastructure layer monetizes the gold rush without taking the prospecting risk. If a GaaS-adjacent company goes public first, bet on this category. Observability and agent-evaluation platforms in particular have the cleanest revenue-quality story in the entire ecosystem.

Vertical agents in regulated, high-value workflows

The second realistic tier is vertical agents that have buried themselves deep in a single industry's workflow -- think legal, healthcare revenue-cycle, insurance claims, or financial back-office. These companies trade horizontal hype for something underwriters love: contracted enterprise revenue, painful switching costs, and a regulatory moat that keeps the next model-wrapper out. They grow slower than the horizontal darlings, but their revenue is legible. A vertical agent at $150M+ in mostly-committed ARR with 130% net retention is a far cleaner IPO story than a horizontal "do anything" agent at $400M in consumption revenue.

A small number of horizontal agent platforms with enterprise traction

A few horizontal players have enough enterprise distribution and brand to attempt it. The risk is that their revenue quality is the worst of the group -- heavy on consumption, exposed to model-cost compression, and vulnerable to a foundation lab simply shipping the same capability natively. They could absolutely go public on momentum. Whether they stay public at a strong multiple is a different question, and one tied directly to the valuation-multiple debate.

Tier 2: The "Needs Two More Years" Group

This is the larger and more interesting bucket: companies with genuinely strong businesses that simply aren't ready to be public.

These are the Series C and D companies that raised at premium valuations, are growing fast, and have real customers -- but whose financials still have a tell. Maybe gross margins are at 55% and climbing but not yet at the 70%+ that earns a software multiple. Maybe net retention is great but the revenue base is too concentrated in a dozen logos. Maybe the burn is still aggressive enough that a public quarter-by-quarter cadence would be punishing.

For this group, the next two years are about margin engineering as much as growth. The single biggest lever is the one McKinsey describes in its work on the economic potential of generative AI: the gap between a flashy demo and a deployed, margin-positive workflow is enormous, and closing it is what turns a fundable company into a public-ready one. Watch for the companies in this tier that quietly start routing more traffic to smaller, cheaper, fine-tuned models -- that's the margin work that makes an IPO possible, and it's the difference between this group and the next one.

Tier 3: Probably Gets Acquired First

A large share of well-funded GaaS companies will never IPO, not because they're failures, but because acquisition is the rational outcome. This isn't a knock -- it's arithmetic.

When a company has a brilliant team, real technology, and a respectable but sub-scale revenue base, an incumbent's "agent attach" thesis makes a clean offer more attractive than the multi-year slog to public-company readiness. We've written about which incumbents are shopping for agents and the build-versus-buy calculus driving it. For founders staring down the burn rate and the margin work required to clear the IPO bar, a strategic acquisition at a strong multiple is often the disciplined choice. The agent-company graveyard is real, but so is the quieter graveyard of perfectly good companies that simply got bought before they ever filed.

The Metrics That Will Make or Break the S-1

If you're actually trying to handicap this, ignore the ARR headline and read the footnotes. Five numbers will determine how any GaaS S-1 is received.

Gross margin trajectory. Not the current number -- the trajectory. A company at 50% margin moving to 65% over four quarters tells a better story than one stuck flat at 60%. Investors are pricing the slope.

Revenue mix: committed versus consumption. The S-1 will have to disclose this, and it's the single most important line for a GaaS company. A high committed-revenue share de-risks the whole model. A revenue base that's 80% pure consumption invites the question every skeptic will ask: what happens in a downturn when customers throttle usage?

Net revenue retention, scrutinized for source. Is expansion coming from genuine workflow growth, or from price increases and model-cost pass-throughs that customers will eventually resist? Underwriters have gotten good at decomposing this.

Inference cost as a percentage of revenue, and its sensitivity to model pricing. This is the GaaS-specific risk factor that simply doesn't exist in a classic SaaS S-1. Any honest filing will have to disclose dependence on a small number of model providers -- a concentration risk that public investors will price.

Customer concentration. Early enterprise GaaS revenue is notoriously top-heavy. If the top ten customers are 60% of revenue, that's a flashing risk-factor light.

The companies that go public successfully will be the ones whose answers to all five are boring. In this market, boring is the whole point.

What the First GaaS IPO Will Signal

The first genuine GaaS IPO will do more than mint a few fortunes. It will set the comparable that every later filing gets measured against, and it will give the private market its first real public price signal for outcome-based AI revenue. Right now, private valuations are floating on narrative and a premium-valuation logic that has never been tested against a daily-traded stock price.

That first print will also resolve an argument the private market has been having with itself: does the public market value a GaaS company on revenue multiples like software, or does it haircut hard for the inference-cost exposure and revenue volatility? My read is that the first one out of the gate will be priced cautiously -- a discount to peak software multiples -- and that the discount will narrow only as the company proves a few quarters of margin expansion. The market will reward durability over dazzle, and it will do it in public, on a Tuesday, in a way no private round ever has to.

When that day comes, the watch list flips from speculation to comparable. Until then, the most useful thing anyone can do is stop reading valuation headlines and start reading the unit economics underneath them.

Insights Most People Overlook

The first GaaS IPO will probably be a company most people don't think of as an "agent company" at all. The cleanest public-ready financials live in the infrastructure and tooling layer, not the consumer-facing agent brands. The press will be disappointed; the bankers will not. Boring revenue goes public first.

Outcome-based pricing is a public-market liability before it's an asset. Founders love selling per-outcome because it aligns with customer value. But "we get paid when the agent succeeds" is exactly the kind of revenue that public investors discount for volatility. The pricing model that wins deals may be the one that drags the multiple, at least until there's a long enough track record to prove the revenue holds in a downturn.

A foundation-model price cut can move a GaaS company's gross margin more than anything its own team does. This is genuinely unusual. Most companies control their own COGS. A GaaS company's single biggest cost line is set by a vendor it doesn't control and often competes with. An S-1 has to disclose this as a risk factor, and a sharp analyst will model the downside scenario where the model provider raises prices or ships the agent's core feature natively. That's a sword over the category that classic SaaS never faced -- and it's why valuation haircuts from margin compression are a live risk even for strong companies.

Customer concentration will out the over-hyped names. A lot of GaaS revenue is a handful of design-partner enterprises paying generously. That looks like product-market fit until the S-1 forces disclosure of how few logos the revenue actually sits on. Some celebrated companies will discover their revenue isn't as diversified as their valuation implied.

The IPO window may matter more than any single company's readiness. GaaS companies don't control macro. A genuinely ready candidate can sit private for two extra years simply because the window is shut, while a less-ready company sneaks out during a hot stretch. Timing the market is part of the watch list whether anyone likes it or not -- which is why this list is a moving target, not a ranking.

References

• Goldman Sachs: Gen AI -- Too Much Spend, Too Little Benefit?
• McKinsey: The Economic Potential of Generative AI -- The Next Productivity Frontier