Urgent.News

What's breaking now, across thousands of outlets.

AI

AT&T (T) Says It’s Not Scared of the “Token Apocalypse.” NVIDIA Corporation (NVDA)’s CEO Is Cheering the Same Trend

AT&T (T) Says It’s Not Scared of the “Token Apocalypse.” NVIDIA Corporation (NVDA)’s CEO Is Cheering the Same Trend

On August 12, AT&T Inc. (NYSE:T) Chief Data and AI Officer Andy Markus stated that OpenAI models currently account for approximately 25% of the telecommunications giant's overall AI usage, with a projected 70% to 80% increase in the future. Meanwhile, NVIDIA Corporation (NASDAQ:NVDA) CEO Jensen Huang has been leading a consortium of companies in support of open-weight AI models.

AT&T's decision to adopt cost-effective open AI models, alongside NVIDIA's endorsement of the same trend, both indicate confidence that open models won't harm the broader AI economy. This raises an important question: does the open-weight wave pose a threat to AI infrastructure development, or does it merely redirect where funds are allocated?

AT&T's transition from closed, proprietary models to open ones has resulted in cost reductions of 80% to 90% in specific applications. The company handles an average of 45 billion AI tokens daily via a smart router that selects the most affordable suitable model for each task. Markus emphasized, "We're not scared of the token future." In addition, AT&T developed its proprietary open model, OTel, and reported a fivefold return on investment this year from AI initiatives.

Some of the open models experimented by AT&T originate from China, a category of models that critics argue raises national security concerns due to state backing for Chinese AI development. Chirag Dekate, a Gartner analyst, pointed out that even enterprises heavily investing in premium proprietary models eventually face unsustainable cost inflection points. This highlights that today's open-model savings may face their own future cost pressures.

NVIDIA's Cerebras CEO Andrew Feldman expressed the belief that there's no reason for chip stocks to decline when open-source models emerge. This viewpoint is based on the Jevons paradox, which states that if AI becomes cheaper to run, people tend to use it more rather than less, potentially consuming as much or more computing power as before.

Cloud and chip companies stand to profit regardless, as open-weight models still require infrastructure to operate. NVIDIA's open-weight advocacy is predicated on the unproven assumption that usage will grow sufficiently to offset any efficiency gains. If large enterprises increasingly adopt smaller, more affordable open models that genuinely require less compute per query, chip demand growth could slow even as usage volumes increase.

Written by urgent.news from Yahoo Finance's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at finance.yahoo.com →

More in AI

Stripe and Ramp Are Betting on AI Traffic Controllers

Stripe is known for processing payments. Ramp is known for managing corporate expenses. But this week, the two companies made almost identical moves: building the layer that decides which artificial intelligence (AI) model gets used. Ramp launched Router.com on Wednesday (Aug. 19), opening up a tool it had run internally for three years.

More from Thursday 20 August →