AI has a price tag, and four numbers show it

A bank's migration cost, a flat spending line, a model with higher token usage and a valuation under pressure all point at what AI costs.

By @0xNotMarc

Cost is the number to watch

Four recent briefs share one thread, and it is the price of AI, seen from both sides of the bill. We have a bank counting its savings, a spending report with a flat line, a new model that uses more tokens (the units AI services bill by), and investors pushing back on a lab's price tag.

A bank puts a figure on it

Westpac says AI-accelerated migration of its legacy St.George data warehouse cut costs from a budgeted $70 million to around $35 million. CEO Anthony Miller says AI tools sped up code conversion, testing and documentation, completing the migration more than two years early. The migration is expected to deliver around $15 million a year in direct benefits, and Miller says the tools, methods and experience developed can be reused for other migrations.

Usage grows while spending stalls

Ramp's data shows token spend only grew 1.5% across August and September, against 42% in July, even as token volumes rose. Ramp's lead economist Ara Kharazian told TechCrunch that OpenAI and Anthropic price cuts have made it cheaper to use AI. A payments team could compare that flat line with its own customers.

Better results, higher token use

Thinking Machines introduced Inkling 2 Small, a mixture-of-experts model with 594 billion parameters, 17 billion of them active, and says it intends to release the full weights under an Apache 2.0 license after external safety testing. The Decoder says the better results come partly at the cost of higher token usage. For a bank weighing an open model, tokens used per finished task deserves a column next to the price per token.

Investors are pricing the labs

The Decoder reports, citing Reuters, that potential backers doubt Anthropic's planned roadshow can kick off as scheduled. It adds that investors want a steeper discount on the $1.5 trillion valuation because the company is still losing money. One investor reportedly floated just $1 trillion. It is still talk for now, and the outcome may look different once the offering is priced.

Which cost line do you track

Price per token, tokens per task and savings at the end of a project are three different numbers, and each tells a different part of the story. It is worth asking which of them your own team reports today, and which one it would be easiest to leave out.