OpenAI is burning through bills like there’s no tomorrow. After raising USD 10 bn last year, the company spent most of it in just 18 months. Now, it’s back with another USD 6.6 bn in funding and USD 4 bn in loans — but here’s the kicker, it’ll need more money in less than two years, the New York Times reports. By 2029, OpenAI’s expenses are expected to climb from the current USD 5.4 bn to USD 37.5 bn.

(Tap or click the headline above to read this story with all of the links to external sources.)

The reason for this financial frenzy? Training AI is expensive. AI development is all about feeding enormous amounts of data into massive systems, and OpenAI’s ChatGPT relies on data centers packed with thousands of GPUs — high powered chips that cost USD 30k each — which run months-long calculations that rack up bills in the hundreds of mns of USD.

But it’s not just chips. The electricity to power them, the data center space to house them, and the amount of hardware needed are all in short supply — and the prices are inflating. Companies like OpenAI, Google, and Microsoft are overspending to manufacture more GPUs, build new facilities, and secure enough power to keep their AI dreams alive. But these tools, aka our friend ChatGPT, are being offered to users for free, and sometimes with a low-cost subscription fee that barely scratches the surface of their costs.

The reason behind the madness: OpenAI is aiming to create artificial general intelligence (AGI), a machine that can think and learn like a human. The tech needed for this, like the new OpenAI o1 system, is bigger and more powerful than anything before it. Earlier this month, OpenAI disclosed their intention to ditch a clause that stops AGI being used for commercial gain, meaning that they may be looking to un-dent their financial reserves after the technology debuts.

Tags: