However, in our experience over the last year dealing with Mid-market and Enterprise customers, the consumption-based or outcome-based pricing is not preferred by cos. They would prefer predictability in contracts and this is where SaaS cos including Salesforce are now trying to bundle the AI credits in per-seat pricing.
Thanks, that's a great point. Consumption-based pricing looks like the rational solution for SaaS companies, but as you say, their customers are used to the predictability of seat-based licensing. Will be interesting to see how it all shakes out.
Really like your application of financial tools to technology.
Did you read the Michael Burry / Dwarkesh piece? Burry was rightly and unsurprisingly concerned about all of this layered and recursive debt blowing up. He also specifically called out the unusual way depreciation applies to GPUs; not only is it probably accelerated but can depend on shocks like a new class of chip being released. He argued that these investments are more accurately expensed than capitalized.
He mentioned that companies could use an accounting treatment where capital equipment could be called “unused” and delay depreciation. Which makes sense for, say, a large construction company buying several excavators and storing them in a warehouse, in shrinkwrap from the factory, for a year until they are needed. But totally inappropriate for cutting edge technology still advancing rapidly.
I am sure there are many smart CFOs who could create pools of technically unused but obsolete AI hardware and not depreciate it, eventually writing it off, probably to fiscally catastrophic consequence to their stock price.
This is a smart point. The key distinction is economic depreciation vs accounting depreciation. Even if a CFO wants to treat GPUs as 'not yet in service' auditors probably care about when the asset is actually available for use. And the bigger issue is impairment if obsolescence or secondary market prices move against you.
Also, a financing unwind of the kind that Burry is concerned about could make AI cheaper, faster: distressed GPU supply, lower inference costs, more aggressive bundling. So the business model pressure on seat-based SaaS can intensify even if the capex/debt layer gets messy.
I appreciate your formulaic thinking. I’d value your thoughts on advantages to software companies by size. If scale economics matter less, is customer acquisition cost and brand value remaining structural advantages of the big boys?
I think AI weakens scale in production (features are cheaper to replicate). But it strengthens scale in distribution (owning the surfaces), trust/procurement, and bundling. So, yes, CAC & brand remain, but the edge moves to being the default tool and already being approved by the CISO office (or whatever equivalent there is in a given enterprise). Specialists can win by going deep on a workflow & integrating tightly & pricing to outcomes.
Such a detailed breakdown.
However, in our experience over the last year dealing with Mid-market and Enterprise customers, the consumption-based or outcome-based pricing is not preferred by cos. They would prefer predictability in contracts and this is where SaaS cos including Salesforce are now trying to bundle the AI credits in per-seat pricing.
Thanks, that's a great point. Consumption-based pricing looks like the rational solution for SaaS companies, but as you say, their customers are used to the predictability of seat-based licensing. Will be interesting to see how it all shakes out.
Some Twitter smarty pants said “ you can’t have a bull market in AI and a bull market in SaaS at the same time.” Great post Dave.
Thanks. It has been interesting to watch people react to this dynamic, to say the least.
Really like your application of financial tools to technology.
Did you read the Michael Burry / Dwarkesh piece? Burry was rightly and unsurprisingly concerned about all of this layered and recursive debt blowing up. He also specifically called out the unusual way depreciation applies to GPUs; not only is it probably accelerated but can depend on shocks like a new class of chip being released. He argued that these investments are more accurately expensed than capitalized.
He mentioned that companies could use an accounting treatment where capital equipment could be called “unused” and delay depreciation. Which makes sense for, say, a large construction company buying several excavators and storing them in a warehouse, in shrinkwrap from the factory, for a year until they are needed. But totally inappropriate for cutting edge technology still advancing rapidly.
I am sure there are many smart CFOs who could create pools of technically unused but obsolete AI hardware and not depreciate it, eventually writing it off, probably to fiscally catastrophic consequence to their stock price.
Thanks.
This is a smart point. The key distinction is economic depreciation vs accounting depreciation. Even if a CFO wants to treat GPUs as 'not yet in service' auditors probably care about when the asset is actually available for use. And the bigger issue is impairment if obsolescence or secondary market prices move against you.
Also, a financing unwind of the kind that Burry is concerned about could make AI cheaper, faster: distressed GPU supply, lower inference costs, more aggressive bundling. So the business model pressure on seat-based SaaS can intensify even if the capex/debt layer gets messy.
I appreciate your formulaic thinking. I’d value your thoughts on advantages to software companies by size. If scale economics matter less, is customer acquisition cost and brand value remaining structural advantages of the big boys?
FYI-typo here: “jon-to-be-done”
Thanks, appreciate it.
I think AI weakens scale in production (features are cheaper to replicate). But it strengthens scale in distribution (owning the surfaces), trust/procurement, and bundling. So, yes, CAC & brand remain, but the edge moves to being the default tool and already being approved by the CISO office (or whatever equivalent there is in a given enterprise). Specialists can win by going deep on a workflow & integrating tightly & pricing to outcomes.