Four bundled offerings — now available through HPE's global seller and partner network — give operators a validated path from pilot to subscriber-scale AI that lives on the line, runs at the network edge, and remembers every interaction.
Today's large language models reset with every session. On a carrier network, that statelessness shows up everywhere the operator engages with the customer. The assistant on a subscriber's personal line treats a ten-year customer like a stranger. The SMB front desk can't recall the callers it greeted yesterday. The Network Operation Centers (NOC) copilot loses incident history between shifts and sites. The agent on a connected device doesn't know the SIM it lives on. Nothing accumulates on the asset the operator uniquely owns — the line.
Personal AI inverts that. Memory lives where identity already resides: bound to the number, resident at the network edge, and compounding with every call, text, and event. This is the architecture that transforms AI from something that performs well in demos into something that gets smarter and stickier with every subscriber interaction. By pairing persistent memory with distributed inference, it remains durable at carrier scale rather than becoming brittle in production.
That's why we're expanding our partnership with HPE to bring memory-based AI to operators globally. Personal AI has joined HPE's Virtual OEM (V-OEM) program, offering four versions of our AI Memory Platform through HPE's trusted global seller and partner network. For carriers, that means one trusted vendor relationship, validated hardware and software, and a structured path that scales from a single line of business to the full subscriber base.
Most AI platforms are built on general-purpose LLMs, which are stateless by default—they do not retain memory of previous conversations unless memory is layered on top. Personal AI is built differently, from the ground up:
A memory core. Every subscriber, business, and connected device is anchored to a persistent memory profile that accumulates knowledge across all touchpoints — calls, texts, documents, preferences, and historical interactions. The result is a stateful AI that retains context, recognizes intent, and develops a real relationship over time. No resets. No stranger danger.
Purpose-built Small Language Models (SLMs). Rather than deploying oversized general-purpose LLMs, Personal AI uses domain- and business-specific models that are 40× cheaper than hosted LLMs at scale and deliver 27× faster time-to-first-token. That efficiency comes from model distillation, LoRA-based training, Cache-Augmented Generation (CAG), and Retrieval-Augmented Generation (RAG) — enabling up to 40 specialized AI models to run simultaneously on a single GPU.
Telephony-native voice. Voice is built in from the ground up — not bolted on. The full pipeline (ASR → SLM → TTS) completes end-to-end in under 500 ms, with sub-300 ms bidirectional latency. In a recent benchmark at 5× burst load, the complete pipeline clocked 424.3 ms P50 — well inside the conversational SLA that keeps subscribers engaged.
Distributed unit economics. The platform is designed for edge deployment with efficient distributed inference, reducing cost-per-million-tokens versus centralized cloud environments. This makes AI economically viable at scale across consumer, SMB, and enterprise segments.
Privacy and compliance by design. Self-contained, self-hosted deployments ensure subscriber data never leaves the network — critical for CPNI compliance and regulated segments. Personal AI does not retain customer data, giving operators full ownership and control.
This partnership didn't start on paper. At NVIDIA GTC 2026, HPE announced the HPE AI Grid — an end-to-end solution connecting AI factories and distributed inference clusters across regional and far-edge sites. The launch showcased Personal AI SLMs running in live field trials on HPE ProLiant servers with NVIDIA GPUs, deployed across a highly distributed carrier network to deliver AI services for small businesses.
The results from that distributed architecture are measured, not modeled — captured under a 5× burst load at 500 concurrent users on a NVIDIA-accelerated HPE infrastructure:
These offerings target the places where memory-based AI is business-critical for the operator:
Through this partnership, AI becomes a revenue line for operators, not a cost center. The line is the one asset only the carrier owns, and memory-based AI turns it into the identity, billing, and distribution layer for AI — monetized through existing relationships and infrastructure. The payoff: stronger retention, net-new SMB revenue, leaner network operations, and fresh ARPU from connected devices — all packaged, metered, and billed via token-based plans, exactly like voice and data.
"Telecom AI needs more than intelligence — it needs memory. By combining Personal AI's memory architecture with HPE's enterprise reach and V-OEM program, we're giving carriers AI that is easier to buy, safer to deploy, and simpler to scale — AI systems that continuously learn, adapt, and improve with every subscriber interaction." — Suman Kanuganti, CEO & Co-founder, Personal AI
Procurement complexity, deployment risk, and scaling challenges often slow AI initiatives. Pairing the Personal AI Memory Platform with HPE’s infrastructure and global partner ecosystem gives customers validated solutions that accelerate time-to-value and reduce operational complexity.
The four bundled offerings let organizations start with targeted use cases, prove business value, and scale confidently from pilot to production. Whether deployed in carrier data centers, at the network edge, or across distributed environments, operators benefit from enterprise-grade infrastructure, global support, and a streamlined path from trial to production — including sovereign and on-premises deployments where subscriber data never leaves the network.
These offerings power the next era of AI, where personalization and continuity are business-critical: telecom providers deploying 24/7 executive assistants, digital receptionists, account representatives, and network operations copilots; business-specific AI agents, sales associate augmentation, ordering agents, and product-knowledge experts; and, across the device lines they provision, troubleshooting agents for robots, connected devices, connected cars, and connected homes, plus delivery robots and other autonomous agents.
To learn how to get started with the Personal AI Memory Platform and HPE V-OEM offerings for telecommunications, contact your HPE account representative or visit hpe.com/oem.
