Broadcom adds 50G PON gateway SoC with edge AI engine

Broadcom has introduced the BCM68850, a 50G PON gateway SoC with integrated edge AI acceleration, native Wi-Fi 8 compatibility, and advanced security features.


IN Brief:

  • Broadcom has introduced the BCM68850 50G ITU-PON gateway SoC.
  • The device integrates a neural processing unit, native Wi-Fi 8 compatibility, and advanced security features.
  • Edge AI functions are moving into communications silicon as networks require lower latency and local intelligence.

Broadcom has introduced the BCM68850, a 50G ITU-PON gateway system-on-chip with an integrated neural processing unit, native Wi-Fi 8 compatibility, and symmetric 50G throughput.

The device is designed for next-generation fibre gateway equipment and adds local AI inference capability at the broadband access edge. Broadcom has also included intelligent self-healing functions and advanced security features, including post-quantum cryptography support.

The BCM68850 extends Broadcom’s end-to-end 50G PON portfolio, connecting with the company’s optical line terminal and optical network terminal technologies. The device is sampling to early access customers and is intended to support high-throughput gateways that process more functions locally rather than relying entirely on upstream cloud resources.

Broadband access equipment is often judged by service speed, but the silicon requirements are becoming broader. Gateways are turning into embedded computing nodes, combining packet processing, wireless coordination, AI inference, security, diagnostics, and energy management inside constrained power and thermal envelopes.

The neural engine reflects that change. Local AI can support traffic optimisation, fault prediction, quality-of-service decisions, security monitoring, and adaptive Wi-Fi behaviour. Placing those functions inside the gateway gives them direct access to network data and reduces the latency associated with sending every decision upstream.

AI demand is spreading well beyond the data centre. Rising compute deployment has already contributed to pressure on MLCC supply, while edge equipment, telecoms hardware, industrial systems, and embedded devices are adding their own layers of local processing.

Gateway design now has to balance high-speed interfaces with enough processing and memory headroom for software-defined services. Security is equally important because access equipment sits at the boundary between operator networks, enterprise systems, and end-user environments.

Post-quantum cryptography support addresses a long product-lifecycle problem. Network equipment can remain deployed for many years, while security assumptions may change during that period. Stronger cryptographic support in silicon can help reduce the risk of hardware becoming obsolete before the end of its operational life.

Native Wi-Fi 8 compatibility also reflects the convergence of optical access and local wireless networking. Fibre capacity only delivers its value when customer-premises equipment can distribute bandwidth reliably across devices, applications, and local environments. That requires tighter coordination between optical access, gateway processing, wireless control, and diagnostics.

The BCM68850 shows how communications processors are absorbing AI, security, and self-management functions as networks become denser and more automated. Faster access speeds remain important, but the gateway is increasingly being designed as an intelligent edge platform rather than a passive connection point.


Stories for you