Jun 26, 2026 · 21 min · 10 segments
This week's Podcast looks at the journey the industry has been on with AI. Over the last few years the only story in town has been Training. Huge investments with no capital return. Now it's the time…
OK, let's unpack this, because to really grasp why the entire tech sector is pivoting so hard right now, we have to look at the tension between sunk costs and variable costs.
Yeah, that tension is the core economic driver of basically everything happening in the AI space right now.
For sure.
For the past few years, the entire tech ecosystem has been fixated on the capital expenditure of creating a neural network.
And that training phase, well, it's a monumental one-time sunk cost.
Right.
Using that factory analogy again, training the frontier AI models, like laying the concrete, importing the heavy machinery, constructing the building itself.
It's a huge upfront bill.
It really is.
But inference.
Inference is the electricity.
It's the raw materials and the labor required for every single individual widget that rolls off that assembly line forever.
That paints the perfect picture, honestly, because when you construct that factory, the pre-training phase, right, you are feeding the neural network billions of parameters so it can learn language and logic and patterns.
The heavy lifting.
Exactly.
But once the doors open, every single time a user queries that AI or every time it generates a line of code or an autonomous agent resolves a customer support ticket, you are executing an inference workload.
And that costs money every single time.
Right.
Inference is a continuous, relentless variable cost.
It literally never starts.
I actually brought some data from the source material and the scale of this variable cost is just staggering.
Oh, the numbers are wild.
Yeah.
Which is huge.
OK, let's unpack this, because to really grasp why the entire tech sector is pivoting so hard right now, we have to look at the tension between sunk costs and variable costs.
Yeah, that tension is the core economic driver of basically everything happening in the AI space right now.
For sure.
For the past few years, the entire tech ecosystem has been fixated on the capital expenditure of creating a neural network.
And that training phase, well, it's a monumental one-time sunk cost.
Right.
Using that factory analogy again, training the frontier AI models, like laying the concrete, importing the heavy machinery, constructing the building itself.
It's a huge upfront bill.
It really is.
But inference.
Inference is the electricity.
It's the raw materials and the labor required for every single individual widget that rolls off that assembly line forever.
That paints the perfect picture, honestly, because when you construct that factory, the pre-training phase, right, you are feeding the neural network billions of parameters so it can learn language and logic and patterns.
The heavy lifting.
Exactly.
But once the doors open, every single time a user queries that AI or every time it generates a line of code or an autonomous agent resolves a customer support ticket, you are executing an inference workload.
And that costs money every single time.
Right.
Inference is a continuous, relentless variable cost.
It literally never starts.
I actually brought some data from the source material and the scale of this variable cost is just staggering.
Oh, the numbers are wild.
Yeah.
Which is huge.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
