If AI Capex Slows, B2B SaaS Multiples Face Two Opposite Forces at Once

AI capital spending adds roughly 140 basis points to US growth in 2026, on Bridgewater's estimate. If that spending decelerates, SaaS multiples get pulled two ways at once. The fear that agents will replace software seats eases, which lifts them. The capital and the growth impulse behind the whole market thins, which pushes them down. Which force wins depends on why capex slowed, not by how much.

What actually sets a B2B SaaS multiple

A revenue multiple looks like a market opinion about a company. It is closer to shorthand for three separate beliefs, and those beliefs move independently.

Growth is still the largest single input

Faster compounding is worth more, so growth carries most of the spread between companies. SaaS Capital reports median growth across its index falling from above 30% at the 2021 peak to the low teens today.

Anyone calling today's software valuations pure panic has to explain away 20 points of lost growth first.

The discount rate sits underneath everything

Software revenue is long-duration. Most of the value in a SaaS model sits in years 5 through 15. The rate used to discount those years therefore matters more than it does for a business earning its money now.

This is the mechanical link between AI infrastructure finance and your valuation. It does not care what your product does, only what money costs.

Terminal value is what AI actually moved

The third belief is whether the business still exists at the end of the forecast period. That is the one 2026 changed.

SaaS Capital's own read on the Q1 2026 sell-off is that the market treated AI as an existential risk to the category, not a company-specific threat. That is why the re-rating was broad rather than selective. The mechanics are covered in the breakdown of the software selloff.

Where B2B SaaS multiples stand in August 2026

There is no single authoritative number for the SaaS multiple. The range between credible sources is wide enough to change a fundraise conversation.

The number depends on which index you read

SaaS Capital's index covers 63 public companies and divides market capitalisation by annualised run-rate revenue. On that basis the firm describes a new normal of 6 to 8 times ARR from 2023 through 2025, then a sharp re-rating to decade-plus lows in the first quarter of 2026.

Comparison sets built on forward revenue read lower. Multiples.vc put the median public software company at 2.2 times next-twelve-months revenue on 19 August 2026, with infrastructure software at 2.8 times, vertical software at 2.3 times and horizontal software at 2.2 times.

Those two readings are not in conflict. They use different denominators.

The market cut multiples while margins improved. SaaS Capital notes that the median public SaaS company now reports an operating profit, the first time in more than 10 years.

Profitability improved. Growth slowed. The multiple fell far more than the growth slowdown alone explains. That gap is the market pricing a shorter assumed life for seat-based software, and it is the part a capex slowdown could reverse.

Why a capex slowdown pulls SaaS multiples in two directions

Most commentary picks one side and writes it up as the answer. Both sides are real, and they act on different parts of the valuation.

Force one, upward: the displacement discount unwinds

AI capex is the physical expression of a belief. Hundreds of billions of dollars of accelerators and floor space only make sense if enterprises are about to route work through models rather than through people holding licences.

If that build decelerates because demand disappointed, the timeline for agents absorbing software seats stretches. Terminal value assumptions repair, and the multiple follows. That is the argument for a cooling AI cycle being good news for incumbent software.

The argument is right in direction and overstated in size. Capex funds capability that has already shipped. The compute needed to compress seats in a mid-market support team exists today.

Force two, downward: the money does not come home to you

The second force works through the market rather than through your product. AI capital spending is now the dominant marginal source of US growth. A slowdown removes the growth impulse funding everyone's budget.

Analysis at The Overshoot found the five largest cloud providers spent about $150 billion in Q1 2026, equal to roughly 2% of quarterly US GDP, against under $80 billion a year earlier. The same analysis attributes about 1 percentage point of the 6% nominal GDP increase over those 12 months to that spending alone.

There is also a credit channel, and it is the faster one. If capex slows because financing repriced rather than because demand softened, spreads widen and every long-duration equity de-rates together. The exposure behind that channel is mapped in the piece on AI debt and private credit.

The third part is the one founders underrate. Capital that leaves AI infrastructure does not queue up to buy horizontal SaaS next. It buys whatever is cheap, cash-generative and uncorrelated with the thing that just broke. Software in 2026 was neither.

Three scenarios, and what each does to your multiple

These are scenarios, not forecasts. Each is built from a different reason capex could slow, because the reason determines the outcome.

Scenario 1: the glide path

Capex growth decelerates roughly along the path Goldman Sachs Research already assumes, with no credit event and no abrupt guidance cut. Rates hold. The AI story stops being a daily repricing event and becomes a set of quarterly numbers.

SaaS multiples then recover from the top of the distribution rather than across it. Companies with strong net retention and revenue attached to AI usage re-rate first. Everything sold purely by the seat stays cheap. This is the consensus path, and consensus paths are mostly priced already.

Scenario 2: the air pocket

Capex slows abruptly because financing got expensive, not because demand fell. A large lease commitment is renegotiated, one operator guides capital spending down mid-year, and credit spreads on AI-linked issuers widen sharply.

Every long-duration asset de-rates together, software included, despite software having caused none of it. SaaS multiples compress further from an already low base, then recover on a lag of two to four quarters once the discount rate settles.

This is the scenario founders are least prepared for, because being right about your own business does not protect you. It is also, in my judgement, the most likely of the three. Debt-funded builds stop for reasons unrelated to whether the thing being built was a good idea.

Scenario 3: rotation home

Capex slows and, in the same period, agent deployments visibly underdeliver inside large enterprises. Seat counts hold. Net revenue retention stabilises. The evidence of displacement fails to arrive on schedule.

Terminal value repairs quickly here, into a market just reminded that capital-intensive assets depreciate while software does not. Profitable, low-capex businesses look very different after an infrastructure disappointment.

This is the highest-upside path and the one with the least supporting evidence today. I would not build a plan on it, only make sure I could take advantage of it.

Each scenario has a test that would rule it out. A sharp move in credit spreads converts the glide path into an air pocket. Capex slowing while spreads stay tight points to a demand cause rather than a financing one. Continued step-downs in reported net retention would rule out rotation home.

The variable that decides which scenario you get

The headline capex number tells you almost nothing. A 20% cut to a $500 billion programme and a 20% cut driven by a funding market closing are the same number and opposite events.

The distinction that matters is whether the slowdown arrives through the income statement or the balance sheet. Demand-led slowdowns are absorbed slowly and mostly help software, because they stretch the displacement timeline. Financing-led slowdowns arrive at once and hurt everything priced on future cash.

So the indicator to track is not capex guidance. It is the cost of AI-linked credit and the mix between cash-funded and debt-funded building, the same signal flagged in the analysis of the payback math behind the build.

Where this argument is weakest

The framework above has three real problems, and one of them could invalidate the whole thing.

The mechanism assumes capex is the market's proxy for displacement risk. That is an inference, not an observation. The market may instead be pricing observed customer behaviour, in seat counts and renewal data, and ignoring capex. If so, a capex slowdown does little to software multiples in either direction, and the framing collapses to one force.

The macro drag may be smaller than the GDP arithmetic suggests. A material part of the growth contribution comes from chipmaker profits unlikely to be recycled into the wider economy, and the build employs very few people relative to its cost.

Bridgewater's own example, which cuts against its headline figure, is stark. A Meta data centre in Texas costing about $1.5 billion is associated with roughly 100 jobs, against 1,620 at a comparably sized battery plant. An investment category that creates almost no employment also destroys almost none when it slows.

That weakens force two. A capex slowdown could hit measured GDP hard while leaving the corporate software budgets that pay your invoices largely intact.

The strongest counter-case is that multiples deserve to stay low. Median growth across public SaaS has fallen from above 30% to the low teens, and the median company now runs at a modest operating profit. A low-teens grower with thin margins is a normal software business, and normal software businesses have historically traded in single digits against ARR. On that reading nothing is broken. The category matured, and the 2021 comparison was always the anomaly.

I do not fully believe that version. The speed of the Q1 2026 move looks like a repricing of risk rather than a slow adjustment to maturity. It is still the strongest argument against this post.

This post deliberately does not give you a single current SaaS multiple. The most credible long-running series publishes its level in a chart and a downloadable file rather than in prose, and the forward-revenue comparison sets use a different denominator. Quoting one figure as the SaaS multiple would have been the easy version and the wrong one.

What this changes if you run a SaaS company

None of this is actionable at the index level. It is actionable at the company level, in three ways.

Plan off runway, not off recovery. Assume the multiple available at your next raise is the one available today. If scenario 2 arrives while you are in market, the round reprices in weeks and your operating performance will not save it. Longer contracts, more prepayment and a lower burn multiple are what convert an air pocket into a survivable quarter.

Move revenue off pure seat counts. The market is discounting seat-based pricing specifically, not software generally. Forbes reported the share of SaaS companies using per-seat as their primary model falling from 21% to 15% in 12 months, while hybrid models rose from 27% to 41%.

That shift answers buyers asking why they need 500 licences when an agent handles the work of several people. The pricing mechanics are worked through in the piece on seat compression, and the competitive angle sits in the analysis of vertical AI against horizontal software.

Instrument what your product replaced. Buyers arrive with a comparison you did not choose: your price against the cost of building the same capability with an agent and an API. The answer that works is a number in their units, from a baseline you recorded before deployment.

Gartner raised its 2026 worldwide IT spending forecast to $6.31 trillion, up 13.5%, with data centre systems growing 55.8%. The money is there. It is being allocated on evidence rather than category loyalty, and vendors who cannot produce evidence are losing renewals inside a growing market.

Frequently asked questions

Related reading