aaron holmes

AT&T processes 45B tokens/day internally and uses open-source models (Nvidia NeMo, Meta Llama, Google Gemma) for ~40% of workloads, targeting 60-70%, to keep OpenAI/Anthropic spend flat despite growing usage. Open-source is 6-10 months behind frontier but gap is narrowing, creating structural cost pressure on closed-model labs.

Enterprises are increasing cybersecurity budgets 20-30% to counter AI-powered attack capabilities, creating a structural tailwind for AI model labs and incumbent vendors building AI-native defense tools, while disrupting legacy vulnerability scanner business models.

Large enterprises like AT&T are systematically replacing 40-70% of frontier model API calls with self-hosted open-source models (Llama, Gemma, NeMo) to control costs, with the performance gap narrowing to 6-10 months behind the frontier. This creates a structural headwind for OpenAI/Anthropic enterprise revenue growth.

Large enterprises like AT&T are systematically replacing OpenAI/Anthropic API calls with self-hosted open-source models (Llama, Gemma, Nemotron) for routine tasks, keeping spend flat while usage grows 3x, though hidden maintenance costs ('puppy is free' analogy) and model lag (6-10 months behind frontier) remain risks.

AT&T's 100k employees use AI for coding, support, HR — processing 45B tokens/day. They route sophisticated tasks to OpenAI/Anthropic APIs but offload summaries and routine work to self-hosted open-source models on own data centers, targeting flat foundation model spend despite growing usage.

AT&T's shift to 40-70% open-source usage on own infrastructure demonstrates a scalable enterprise playbook: customize Llama/Gemma/Nemotron, run on owned GPUs, avoid per-token API fees, and gain data control — threatening closed-model revenue growth.

CISOs perceive attackers gaining advantage from AI coding tools, forcing near-term spending surges on foundation models for defense and new AI agent tools from incumbents, while legacy vulnerability scanners face disruption from LLM-native alternatives.

Large enterprises like AT&T are moving 40-70% of AI workloads to customized open-source models on private infrastructure to avoid hundreds of millions in closed-model API costs, prompting OpenAI and Anthropic to launch new enterprise privacy features in response.