Software split into two prices this year

Hey {{first_name|Investor}} -

Two software companies. Same asset class, same year, same market.

One trades around 9.2 times its revenue. The other trades around 1.7 times. That's the gap between DevOps software and sales and marketing software on the current Multiples.vc read, using median forward twelve-month revenue multiples built on FactSet consensus estimates.

A 5x spread inside one corner of tech. The market spent this year deciding that some software gets one price and some gets another, and it used a fairly blunt rule to sort them.

The rule is roughly this: if a company charges per employee who logs in, mark it down. If it charges per unit of machine activity, mark it up. The logic being that AI means fewer employees logging in.

The rule is directionally right. It's applied crudely enough that the mistakes are worth finding.

Every software business bills against a unit. Something it counts, then sends you an invoice for.

Some count seats. Fifty employees with logins, fifty times the monthly rate. Others count activity: a text message sent, a minute of a phone call, a gigabyte stored, a transaction processed, a support ticket resolved.

Think of it as a parking pass against a taxi meter. The parking pass costs the same whether you drive or not. The meter only runs when the wheels turn.

That distinction sat in the background for fifteen years because the two moved together. More business meant more employees meant more seats. AI breaks the link, and it breaks it in both directions at once. Fewer people are needed to do a given job, which shrinks the seat count. And far more machine actions run underneath, which grows nearly every other unit there is.

So the first question on any software company is narrow.

Gate one: when AI does more of the work, does more of it run through this company's meter?

On its own, gate one proves very little.

Activity through a meter and money in the business are two separate measurements. Between them sits the question of what the company keeps.

Gate two: does the company hold onto a margin on the extra units?

Gross margin is the plain version of that. Take what a company charges, subtract what it costs to actually deliver the service, and what's left is the money the business gets to run on. A company can double its activity and keep nothing extra if the cost of serving that activity doubles alongside it.

Three ways gate one passes and gate two quietly fails.

Pass-through. You bill the customer, then hand almost all of it to somebody else. It shows up as revenue and stops there. Like a restaurant adding a delivery fee it forwards straight to the courier: the bill got bigger, the kitchen kept nothing.

Bandwidth ($BAND) shows it plainly, and the same mechanic is running through several communications companies right now.

Bandwidth reported 22% revenue growth on July 29. Underneath that headline, its actual Cloud Communications business went from $136 million to $152 million, which is 12%. Meanwhile the messaging surcharges it collects from customers and forwards to phone carriers went from $44 million to $68 million.

Of $40 million in reported growth, $24 million is money passing through on its way somewhere else. That's 60% of the growth.

Gross margin on a reported basis fell to 36% from 40% precisely because of it, since those forwarded dollars carry no profit. The stock dropped about 26% on what looked like a beat-and-raise quarter.

Bandwidth's AI story has substance: every $1 million-plus customer win and expansion in the quarter included its AI or orchestration products. The growth rate is still 12%, and anyone reading 22% is reading somebody else's money.

Yield compression. Volume climbs while the price per unit falls faster, so the spread thins. Marqeta ($MQ) is the live case: payment volume grew 32% last quarter while net revenue grew 17%. More transactions, less kept per transaction.

Cannibalization. The new meter eats the old one. A company selling an AI agent that resolves customer support tickets might simply be converting a subscription its customer already paid for into usage billing, at an unknown spread, while the headline reads like new revenue.

A name clears my bar when both gates pass.

Toast ($TOST) fails it.

Toast runs the software and payment system inside restaurants, and it's a business I like. Payment volume of $60.7 billion, up 22%. Recurring revenue up 25%, recurring gross profit up 28%, and order, menu, labor and payment data across roughly 180,000 locations that would take years to rebuild.

Gate two, comfortably. It fails gate one, because people eat the same number of meals whether or not an agent books the table. AI raises what Toast can sell to a restaurant it already has. The number of restaurant transactions in the world stays where it is.

Good business, no AI catalyst.

The cybersecurity argument is sound and it's crowded, which is why I'm leaving it alone here.

Every AI agent deployed needs an identity, permissions, monitoring and an audit trail, so the companies selling that get a bigger job as agents multiply. CrowdStrike ($CRWD) and Palo Alto Networks ($PANW) are the two names that come up. The market reached that conclusion early and has held it since.

I laid out where machine identity and edge security sit in the stack in my midyear review, including the two names I hold there.

What follows runs the same test on software the market sorted with its blunt rule and never went back to check.

The names that clear both gates compound quietly for years without needing you to be right about which AI model wins or which lab ships the next one. They collect on activity regardless of who provides it.

That's a different quality of holding than betting on a single winner. It's also the kind of position you can hold through a drawdown without checking it daily, which matters when the goal is to stop trading your hours for money.

The catch is that the market has already sorted software into two prices, and the sorting was done with a blunt instrument. Some of what got marked down deserved it. Some of it bills in units that AI multiplies.

Telling those apart takes running both gates on the actual filings.

The rest of this issue is for Insiders.

Below the line I run both gates on 7 companies, using their most recent quarters, and label exactly where the evidence sits on each one.

Two of them you already know. Shopify ($SHOP) had its AI-driven orders triple last quarter, then declined to say what share of its total sales that actually represents. That gap is the whole question. Twilio ($TWLO) had its reported gross margin fall while its underlying margin rose, and both are true at once for a reason worth understanding before you read any communications company's numbers again.

The other 5 are less obvious. One held its gross margin flat while its business mix shifted hard against it, and leaves revenue it's already earning out of its own forecast on purpose. One had its CFO qualify the headline AI figure in the same sentence he announced it. One is the only company here showing you what an AI product's margins look like while they climb out of the hole.

I also cover the shape of software that loses units, and the three things that would tell me I'm reading this wrong, including the company that proved this quarter that usage billing cuts both directions.