Nearly nine in ten operators — 89% — now consider open-source software critical to their AI strategy, according to Nvidia's State of AI in Telecommunications 2026 survey, and the chip vendor says carriers such as SoftBank Corp, AT&T and Indosat Ooredoo Hutchison are already deploying open models alongside proprietary systems in production networks.

The figure marks a turning point for an industry that has largely depended on closed AI platforms from big technology vendors. In a company blog, Nvidia explained that operators retain full visibility over sensitive network data with open models and can fine-tune them for specific operational needs — a level of control that closed systems rarely allow.

Independent benchmarks reinforce the shift. Nvidia pointed to results showing open models closing performance gaps with proprietary systems in reasoning, coding and autonomous network triage — the core workloads carriers need for network automation.

How are operators splitting the workloads?

The picture emerging from deployments is not a wholesale replacement of proprietary AI. Operators are dividing workloads between the two classes of systems:

• Expensive, proprietary models handle complex tasks where closed platforms still hold a performance edge
• Lower-cost open foundation models run day-to-day network operations

This hybrid approach lets carriers balance capability against cost while keeping control over the bulk of routine automation.

What has SoftBank built?

SoftBank Corp offers the most concrete example. The Japanese operator used Nvidia's Nemotron model family as the foundation for its Large Telecom Model, training it on years of internal operational data accumulated from its network.

Rajeev Koodli, principal fellow at SoftBank and SVP of SB Telecom America, said open models allow the operator to capitalise on rapid global AI progress and its accumulated network expertise. In other words, SoftBank is converting operational history — an asset closed vendors cannot replicate — into a competitive advantage.

What are AT&T and Indosat doing?

AT&T is pursuing a multi-model strategy of its own. Chief data and AI officer Andy Markus said the goal is to intelligently match "every workload to the right combination of performance, cost and control."

Indosat Ooredoo Hutchison has taken open models in a localisation direction. Nvidia noted the Indonesian operator tailored its open models to Indonesian language and culture — an adaptation that global proprietary platforms typically do not prioritise for individual markets.

Vendor interest or operator demand?

Nvidia's enthusiasm for open models deserves scrutiny. The company sells the GPUs and networking infrastructure that power AI training and inference regardless of which model family runs on top, so an industry-wide shift toward open-source AI expands its addressable market rather than threatening it. Nvidia's blog functions partly as marketing for its own AI stack, including the Nemotron models SoftBank adopted.

Yet the operator evidence stands on its own. The 89% figure, drawn from Nvidia's 2026 survey, reflects carrier priorities: data visibility, fine-tuning rights and cost control. The operators named — spanning Japan, the United States and Indonesia — represent three continents and very different market conditions, suggesting the trend is structural rather than regional.

The open-versus-closed question also carries regulatory weight. Data sovereignty rules in markets such as Indonesia push operators toward models they can inspect and host themselves, which open-source licensing permits and closed platforms often restrict.

With SoftBank's Large Telecom Model already in operation and AT&T formalising its multi-model approach for 2026, expect open foundation models to take a growing share of routine network automation workloads over the coming year, while proprietary systems retreat to the narrow set of tasks where their performance edge still justifies the price.