Scale's O'Driscoll: Software isn't dead, $688B AI spend creates sales opportunity

Rory O'Driscoll, 30-year software investor at Scale Venture Partners, ran the numbers on whether AI kills SaaS. His answer: software got harder to win, not obsolete. The math matters for anyone selling B2B software in 2025.

Scale's O'Driscoll: Software isn't dead, $688B AI spend creates sales opportunity

The Math on AI Killing Software

Rory O'Driscoll has been investing in enterprise software for 30 years. Long enough to have backed DocuSign, HubSpot, and Box before they were obvious. When the industry started asking if AI killed software, he spent weeks arguing about the numbers.

His answer: software isn't dead. It got a lot harder to win.

The $688B Bet

In 2026, hyperscalers are spending roughly $688B on AI capex. Revenue coming out the other side: about $110B. That is a $578B gap between what is going in and what is coming out.

The revenue does not catch the spend until around 2032, at roughly $1T in revenue. That is five or six years of capital pouring in ahead of returns.

For anyone selling software: this matters. Cheap, improving models create opportunity. They also create competition from players with deeper pockets than any SaaS startup.

Where the Money Comes From

For foundation models to hit $1T in revenue, they have to eat the knowledge worker wage bill. If all the spend landed in the US, you are talking about 15 to 17% of all knowledge worker dollars. For software developers specifically, north of 25%.

Put concretely: for every $200k developer, roughly $50k going to tokens instead of salary.

That is the bet. Scaling laws showed money in produces better models out. ChatGPT showed people actually want to use it. Capital keeps coming as long as both hold.

What This Means for B2B Sales

The question for software companies: does enterprise buy AI direct from foundation models, or is there room for companies that add real value on top?

O'Driscoll's view: both. But the bar is higher. You need to get close to a specific customer, understand their workflow, and ship something economically differentiated. Generic wrappers around GPT-4 do not count.

For sales teams, this changes the pitch. You are not selling software anymore. You are selling domain expertise that happens to use AI. The product has to solve a real problem better than a prompt in ChatGPT.

The Hiccup to Plan For

When you are spending $578B more than you are taking in, someone eventually asks why. Even if the long-term story plays out, expect a hard breath before 2032. Plan your burn like that moment is coming.

For AEs: quota relief when funding dries up is rare. Build pipeline now while the capital is still flowing.

O'Driscoll co-hosts the 20VC x SaaStr podcast. Scale Venture Partners is based in Foster City, California. The firm has been backing enterprise software for 27 years, now focused on frontier tech for enterprise AI.