Multi-Terabit Interconnect Validation for AI Data Centers

Keysight Technologies and Point2 Technology have entered a strategic collaboration to validate next-generation, multi-terabit interconnect technology aimed at removing scale-up connectivity bottlenecks in AI and machine-learning data centers.

Addressing scale-up interconnect constraints
As AI and ML workloads continue to grow, hyperscalers are scaling xPU clusters to unprecedented sizes. In this environment, scale-up interconnects—links between processors, accelerators, and memory within a cluster—have become a critical limiting factor for bandwidth, latency, and power consumption.

Traditional copper interconnects struggle with reach and signal integrity at higher data rates, while optical solutions introduce higher cost, power draw, and latency. The collaboration between Keysight and Point2 focuses on validating an alternative interconnect architecture designed to balance reach, efficiency, and cost at multi-terabit speeds.

e-Tube™ interconnect technology
Point2’s e-Tube™ technology uses RF data transmission over plastic waveguides, targeting short-to-medium-reach, high-bandwidth links within AI systems. The approach is designed to extend reach by up to ten times compared with conventional copper at a comparable cost structure.

From a system perspective, the technology targets significantly lower energy and latency overheads. According to Point2, e-Tube™ interconnects can deliver approximately three times lower power consumption and cost than optical alternatives, while reducing latency by orders of magnitude, making them suitable for tightly coupled xPU clusters where deterministic performance is critical.

Validation and characterization framework
Keysight is providing the validation and characterization environment required to assess whether these interconnects can meet hyperscaler-grade performance and reliability thresholds. The test setup leverages Keysight’s high-speed digital test portfolio, including the M8050A High-Performance Bit Error Ratio Tester (BERT) and DCA-X sampling oscilloscopes.

These instruments are used to generate and analyze complex 120-GBaud PAM4 signals, which are required to support data rates up to 1.6 Tbps per link. Accurate characterization at these speeds is essential to evaluate signal integrity, bit error performance, and margin under realistic operating conditions.

Enabling early 3.2T research and development
A key element of the collaboration is Keysight’s support for early-stage research into 3.2-terabit interfaces. By combining 448-Gbps electrical signal generation with advanced real-time and sampling analysis, the test platform enables engineers to explore next-generation interface designs before formal standards and ecosystem maturity.

This early validation capability allows designers to identify signal integrity limits, optimize modulation and channel characteristics, and reduce development risk as AI infrastructure evolves toward higher bandwidth densities.

Implications for hyperscale AI infrastructure
For hyperscalers, the ability to validate alternative interconnect technologies is critical as AI clusters expand and power efficiency becomes a first-order design constraint. Interconnect choices directly influence system scalability, energy consumption, and achievable performance per watt.

By pairing Point2’s RF-based interconnect architecture with Keysight’s high-speed test and measurement framework, the collaboration provides a pathway to assess whether non-optical, power-efficient interconnects can reliably support multi-terabit scale-up connectivity.

Application context
The validation of e-Tube™ technology illustrates how advances in both interconnect design and test instrumentation are required to overcome emerging bottlenecks in AI data centers. As bandwidth requirements push beyond current electrical limits, rigorous characterization at 1.6T and beyond becomes essential to ensure deployable, reliable solutions.

This collaboration positions both companies within a broader industry effort to redefine scale-up connectivity, enabling denser, faster, and more energy-efficient AI and ML computing architectures.