Via designrush.com
Five new design wins for Silicon One chips and Acacia optics signal Cisco's deepening grip on AI data center networking
Cisco just landed five new hyperscaler design wins for its custom silicon and optics portfolio, and CEO Chuck Robbins says more are coming. The networking giant disclosed the wins during its Q3 FY2026 earnings call, positioning Silicon One chips and Acacia pluggable optics as central to the buildout of AI data center infrastructure.
The company’s hyperscaler AI infrastructure orders are now on track to hit a $9 billion run-rate by the end of FY2026.
What Cisco actually won #
Of the five design wins, three involve Cisco’s networking systems and two are tied to its Acacia optics business. The systems wins include the first commitments for the P200, a deep-buffer router chip capable of handling 51.2 terabits per second. These P200 wins specifically target “scale-across” applications, the networking layer that connects separate AI data centers to each other.
A sixth commitment, another P200 deal, landed just after the quarter closed. Cisco is also seeing traction with its G300 switch silicon, announced in February 2026, which pushes capacity to 102.4 Tbps. Together, the Silicon One family has now shipped over one million units.
Five of the six major US hyperscalers are confirmed customers.
The seven-year bet paying off #
Silicon One didn’t appear overnight. Cisco spent seven years developing the unified ASIC family, which is now manufactured at TSMC’s 3 nm process node. The chips are sold two ways: embedded inside Cisco’s own networking platforms, and as merchant silicon to white-box builders and hyperscale customers who want to design their own systems.
The optics side of the business is growing even faster in percentage terms. Cisco’s Acacia pluggable optics are expected to triple year-over-year.
Prior-quarter hyperscaler AI infrastructure orders had already reached $2.1 billion in a single quarter.
Where Cisco fits in the AI infrastructure stack #
The P200’s focus on scale-across deployments is particularly telling. Early AI infrastructure buildouts concentrated on scale-up networking, connecting GPUs within a single cluster. As models grow larger and training runs span multiple data centers, the bottleneck shifts to the wide-area fabric connecting those sites.
Revenue recognition from these hyperscaler orders is expected to begin flowing during the current fiscal year. Design wins are commitments, not shipments. The lag between winning a design slot and recognizing revenue can stretch months or longer, depending on the customer’s deployment timeline.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our