A supercomputer at your fingertips

I'm now up to five internal Options Technology key note and town hall presentations on "AI transformation" in two weeks, one more to go.

This slide found its way into the decks.

The point, now that we have a dedicated Kimi K3 model (amongst others) running internally it becomes a fixed cost. Very different economics to usage or token based pricing. Simple for CFOs.

The onus is on the team to figure out what to do with it. As one of our AI engineers put it, most firms seem to use a supercomputer to draft emails. The challenge to the team, find the nasty game changing applications.

Equally surprising, that fixed cost is much lower than people expect. CEOs, CFOs and boards are a little bit bamboozled by how to think of investment in AI (with a view to drive productivity gains in the existing business and breakthrough innovation, new products). The current generation of LLMs are the greatest productivity and innovation enhancers ever imagined.

We used a simple heuristic.. it made sense to invest circa 5% of total people costs in the new technology. In a growing business the productivity gains quickly fund the investment, the breakthrough product innovation is the gravy. $500,000 a year makes sense against a run rate of $10m, or $5m against $100m.

The reality now we've actual production cost metrics is closer to 0.5% (lower actually). The economic power of dedicated sovereign infrastructure.

At < 0.5% the mandate to the team is simple.. fill your boots, sweat the investment until it breaks.. or simply, experiment and figure out what to do with the supercomputer. Additional workloads are effectively free.