AT&T and NVIDIA are publicly betting that the shift toward cheaper, open-weight AI models will not harm the broader AI infrastructure economy.
AT&T reported cutting costs by 80% to 90% by switching from proprietary to open models and processes 45 billion AI tokens daily; NVIDIA's leadership argues that lower costs will simply drive higher usage, maintaining demand for chips and cloud infrastructure.
The core question remains unresolved: whether open models genuinely reduce compute demand or merely redirect spending.
What happened
On August 12, AT&T's Chief Data and AI Officer said OpenAI models power about 25% of the telecom's total AI usage, with a target of 70% to 80% over time. NVIDIA's CEO has spent the past month leading a consortium championing open-weight AI. AT&T reported that switching from proprietary to open models cut costs by 80% to 90% in certain applications and processes an average of 45 billion AI tokens a day.
Why it matters
Both companies are betting that cheaper open-weight AI models will not reduce overall spending on AI infrastructure—the theory being that if AI becomes cheaper to run, enterprises will use far more of it, maintaining demand for computing power. However, the assumption remains unproven; if large enterprises shift to smaller, cheaper open models that genuinely need less compute per query, chip demand growth could slow despite rising usage volumes.
What to watch
AT&T says its AI initiatives delivered a fivefold return on investment this year and built its own customized open model called OTel. NVIDIA was held by 275 hedge funds as of Q1 2026, up from 264, while AT&T was held by 72, down from 77.
Ask the AI about this article →
The article frames a core tension in AI economics: whether the rise of cheaper, open-weight models represents a genuine threat to infrastructure spending or merely a redistribution of where money flows. AT&T's experience—cutting costs 80% to 90% while processing 45 billion tokens daily through a smart router—suggests enterprises can extract substantial savings by matching tasks to the lowest-cost suitable model. Yet this cost reduction is precisely what critics worry will erode the infrastructure buildout that has driven recent AI hardware spending.
NVIDIA's public support for open-weight models rests on the Jevons paradox, a historical economic principle suggesting that efficiency gains lead to increased consumption rather than reduced demand. The logic is intuitive: if inference becomes cheaper, companies will deploy AI more broadly, offsetting any per-query efficiency gains. However, the article flags a critical gap: this assumption remains unproven. If enterprises truly migrate toward smaller, cheaper models that require genuinely less compute per task, the paradox may not hold, and chip demand growth could decelerate even as total usage volumes rise.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
London-based Callosum announced a $100 million funding round led by Atomico, with participation from Plural, D…
Anthropic is privately hoping to file for its initial public offering by the end of this month, targeting a ra…
Broadcom is reportedly seeking to borrow up to $100 billion in debt financing to support growth efforts at Ant…
Elice Group, a South Korean AI infrastructure provider, announced the launch of the country's first AI data ce…
Tencent Holdings is receiving shipments of Nvidia H200 AI chips under a new Chinese policy that permits limite…
On August 11, IBM announced a multi-year $240 million agreement with Together AI to deploy NVIDIA HGX B300 sys…
