AT&T moved 40% of its AI work to cheaper open models. It wants 70%.
AT&T already runs about 40% of its AI work on cheaper open-weight models, which companies can download and run themselves, the Financial Times reported on 28 September.
AT&T plans to reach 70% within a year.
AT&T processes about 45 billion units of text, called tokens, a day. At that volume, its chief data and AI officer told the paper, cost matters a great deal. AT&T tunes open models on its own data so they match closed models on some tasks.
Open-weight models can run on a company's own infrastructure, with no fee to the model's maker for each token processed. Computing power and hosting still cost money.
Tinder also routes some AI requests to open models. Its chief technology officer said its AI spending rose from an annual rate of $1 million in January to $10 million in July, according to the FT.
In August, open models handled 56% of tokens on Vercel's service for routing requests to AI models, up from 7% in December. Data from AlphaSense, a financial research service, showed that mentions of open models on US earnings calls in August and September rose sixfold year on year.
Which of your AI uses would you move to a cheaper model if the answers stayed as accurate?
Sources
Our file on AT&T
All 2 AT&T storiesEvery story here is open to read. The ERP LEADERS brief goes one step further.
The week in enterprise software, in one email: the stories that mattered, the failures with their figures, one case taken apart. Read a sample or sign up for the brief.
Welcome back. · Issue 01
