Verizon Signs a $2 Billion-Plus AI Compute Framework With AWS While Posting Its Best Quarter in Years, and the Stock Slid Anyway

Verizon used its third-quarter earnings on Monday to announce a commercial framework agreement with Amazon Web Services valued at more than $2 billion, establishing terms to scale distributed compute capacity for what the company calls the real-time demands of agentic and generative AI workloads. The deal, disclosed alongside record quarterly results, is the clearest sign yet that the largest US telecom is repositioning its fiber network as a supply line for the AI build-out rather than just a consumer broadband business.

The AI angle is written into the capital plan. Verizon is holding 2026 capital expenditures at $16 billion to $16.5 billion, inclusive of success-based investments for its AI Connect program, and says it remains on track to surpass 32 million fiber passings by year-end, with a medium-term target of 40 to 50 million premises. More than half of new fiber customers are choosing download speeds of one gigabit or more. Distributed compute, in Verizon's framing, means pairing that fiber footprint with computing capacity positioned close to users, so that agent-driven applications can run with the low latency and reliable throughput that centralized cloud regions cannot always guarantee.

The financial backdrop was the strongest Verizon has reported in years. Total operating revenue rose 1.5 percent year over year to $34.31 billion, and mobility and broadband service revenue grew 2.9 percent to about $23.5 billion, an acceleration from 2.8 percent in the second quarter and 1.6 percent in the first. Net income rose 5.2 percent to $5.3 billion, and consolidated adjusted EBITDA reached $13.8 billion, up 8 percent with a 40.2 percent margin that the company called its highest ever. Earnings per share came in at $1.25, up 6.8 percent, with adjusted EPS of $1.31, up 8.3 percent. Operating cash flow was $11.5 billion and free cash flow $7.8 billion, up 12.1 percent.

Subscriber momentum drove much of it. Verizon added 406,000 postpaid phone customers, an increase of more than 820 percent year over year and its strongest third quarter in five years, plus 137,000 postpaid net accounts, the best in over a decade. Postpaid phone churn improved four basis points to 0.94 percent while promotional acquisition costs fell 41 percent, reflecting a deliberate pivot away from device subsidies. Broadband net additions reached 314,000, split between 159,000 fixed wireless access lines and 155,000 fiber connections, bringing total broadband connections to 17.4 million.

Management raised full-year guidance for the third consecutive quarter. Postpaid phone net additions are now expected at 1.1 million to 1.2 million, up from the prior outlook of the upper half of 750,000 to 1 million; adjusted EPS guidance moved to $5.04 to $5.06, representing 7 to 7.5 percent growth; and free cash flow growth was lifted to about 12 percent from 9 to 10 percent. Chief executive Dan Schulman, one year into a turnaround that scrapped heavy subsidies in favor of simpler plans, told investors that technology leaders are choosing Verizon's fiber network to power the next generation of AI infrastructure.

The market was not persuaded on the day: the stock slipped about 2.3 percent to roughly $43.62 in early trading, weighed down by equipment revenue that fell 9 percent on lower upgrade volumes and by the broader anxiety over AI-linked spending. Still, the AWS framework puts Verizon in a small club of telecoms, alongside AT&T and T-Mobile, selling network-native compute into AI workloads, and $2 billion is only the opening number in a framework agreement designed to scale.

The structural point is bigger than the deal size. If agentic AI keeps pushing computation toward where data and people are, distributed capacity on dense fiber becomes a monetizable asset, and telecoms become landlords for a slice of the AI stack. Verizon is betting its 17.4 million broadband connections and expanding fiber footprint can capture that demand before hyperscalers simply overbuild it. Whether agentic workloads actually need that geography at scale is the assumption the whole thesis, and the guidance raise, now rests on.