AI financing hits a new frontier

How do you price compute - the processing resource needed to execute digital tasks? How do I compute compute? 

These are the billion-dollar questions boggling the minds of some of the world’s leading technologists and financiers as the generative AI paradigm shift enters its next stage. 

That new trend is ‘inference’, where consumers use AI-powered applications to generate answers and content. 

Silicon Valley is terribly excited about it, while others have warned that inference is the next step towards a Terminator-esque dystopian future. 

Most of us, however, are already using this fancy new technology in non-highfalutin ways, including searching for recipes, booking holidays (link) and cheating at your work’s fantasy football competition (not guilty). 

I’m sure we’ve all let the Galaxy Brains of California down with our rather basic large language model (LLM) usage, but it has at least prompted an existential question about computational usage within the technology industry and Wall Street. 

Since it already takes up a lot of power and energy to build LLMs, the backbone software of the generative AI sector, more smartphone and PC-driven calculations will only compound matters, making compute scarcer.

More data centres can be built, of course, but that takes time and there is a growing resistance against this type of infrastructure in the US, UK and Europe (link). 

Meanwhile, amid macroeconomic headwinds, local opposition and technological changes, compute is becoming more expensive, not least because pretty much all digital systems and applications rely on it, not just AI-related hardware and software. 

That means it’s increasingly viewed as a commodity, like oil, gas or the goods fuelling the agricultural industry. And to help reduce the volatility of compute prices (it can cost more than $14 to rent one GPU processor for just an hour, currently), there is a growing movement to create future contracts out of compute. 

CME Group, the well-known derivatives marketplace, (link) is launching its own contracts, and it could bring a new level of transparency and, in turn, further innovation in the AI sector. 

But eyebrows have been raised elsewhere about the industry’s financing methods, not least because the bond market has started to wobble (link), pushing up the cost of borrowing for everyone. 

Some hyperscalers have preferred to use these markets, rather than fundraising on the equity markets, to raise the cash needed to build data centres and power R&D initiatives.

What is more intriguing still is that some of these mega-technology companies have teamed up with private equity houses to create joint-venture vehicles, in PE-bond tie-ups (link). 

Some see this as good money management, since these complex deals keep the risk off Big Tech’s balance sheets and therefore protect shareholder value. 

But detractors don’t like the sound of this financial pivot, wondering if the AI builders can keep - time after time - tapping up the private markets, especially, and predominantly, in the US.