Point2 Technology has closed a $136 million Series B financing round to bring to market a new class of cable that could fundamentally change how AI data centers are wired. The round was led by LB Investment, with new strategic participation from Arm and continued backing from existing investor Maverick Silicon. The company's investor roster now includes Nvidia, Arm, UMC Capital, Molex, and Bosch Ventures — a lineup that spans chip designers, chip manufacturers, connector specialists, and automotive sensor makers. The breadth of support signals that the industry's most critical bottleneck is no longer just silicon, but the electrical and optical pathways that connect it.
AI's hidden bottleneck: the wiring between chips
As artificial intelligence models grow larger and training clusters expand from thousands to tens of thousands of accelerators, moving data between processors has become as important as computing on it. Most AI workloads rely on distributed training across hundreds or thousands of GPUs or custom ASICs, and those chips must exchange enormous volumes of data at terabit-per-second speeds. The physical layer that enables this data transfer — copper cables for short reach, optical transceivers for longer links — has not kept up with the exponential growth in compute density.
Copper cables are cheap and reliable but limited in reach and bandwidth. Their signals degrade quickly over distance, and they generate significant heat, requiring active retiming and heavier shielding. Optical fiber solves the reach problem but introduces its own complications: lasers and photodetectors are expensive, fragile, and power-hungry. In a data center where power is already a primary constraint, every watt spent on moving data is a watt taken away from computation. The industry is caught between two imperfect options, and that gap appears to be exactly what Point2 Technology intends to fill.
What is Point2's e-Tube platform?
Point2's core technology, called e-Tube, uses RF signaling over plastic waveguides instead of traditional copper wire or optical fiber. Unlike optics, e-Tube does not convert electrical signals to light and back. Instead, it transmits electromagnetic waves in the radio frequency spectrum through a specially designed plastic channel. This approach preserves the benefits of electrical signaling — lower latency, lower power, and lower materials cost — while extending the reach beyond what copper can achieve.
According to the company, the e-Tube platform delivers 10 times the reach of copper at comparable cost, 5 times lower weight, and 2 times less cable volume. When compared against optics, Point2 claims 3 times lower power consumption, 3 times lower cost, and 1,000 times lower latency. Perhaps even more importantly, e-Tube avoids the reliability problems associated with failing lasers, which are a major operational headache in hyperscale data centers.
The numbers are starkly compelling, but the real test lies in whether the technology can be manufactured at scale and integrated into the world's largest AI infrastructure projects. Point2 says the new funding will accelerate commercialization of three product lines: Active RF Cable for existing rack architectures, near-packaged e-Tube for chiplet and high-bandwidth memory integration, and co-packaged e-Tube for future processor and switch designs.
Why Nvidia, Arm, and Bosch all want in
When a startup attracts investment from Nvidia, Arm, Molex, and Bosch Ventures, it is not just a financial endorsement — it is a strategic alignment across the entire computing stack. Nvidia dominates the AI accelerator market and needs interconnects that can keep up with its newest GPU platforms. Arm licenses chip architectures and has a vested interest in enabling high-speed data movement between cores and accelerators. Molex is one of the world's largest manufacturers of connectors, and Bosch builds automotive sensors that increasingly rely on high-speed data links.
Each of these companies sees e-Tube as a potential standard that could lower cost and power in future systems. For Nvidia, the appeal is obvious: rack-scale AI clusters demand massive amounts of cabling, and any technology that reduces power and cost while improving reach could protect profit margins and performance. For Arm, the technology could open new possibilities in chiplet-based design and advanced packaging, where data movement between dies is often the limiting factor. Bosch's interest may stem from industrial and automotive applications that need rugged, low-latency interconnects in harsh environments.
“As AI systems scale and bandwidth demands reach terabit-per-second speeds, interconnect has become the defining bottleneck,” said Sean Park, CEO of Point2 Technology. His remark reflects a growing consensus among infrastructure engineers: no matter how fast a single accelerator operates, an AI cluster is only as fast as its slowest data link.
The race for better interconnects is heating up
Point2 is not alone in seeking to disrupt the interconnect market. Earlier this month, Lumilens raised $700 million for optical interconnect technology, suggesting that a large portion of the industry still believes light is the long-term answer. Other startups and established players are exploring co-packaged optics, silicon photonics, and even free-space laser links. But Point2's thesis is that RF over plastic waveguides can do the same job cheaper, cooler, and faster at rack scale, while sidestepping the most painful aspects of optical systems.
The tension between optical and RF approaches is not new. Optical fiber has won every battle for long-haul networking because it offers virtually unlimited bandwidth at extremely long distances. But within a single data center, the typical link length is just a few meters. At those distances, copper is often sufficient but suffers from weight, heat, and cable congestion. Optical transceivers are over-engineered for such short links — they convert electrical signals to light and back with high power overhead and latency. RF over plastic waveguides occupies a middle ground: faster than copper over distance, but cheaper and more robust than optics.
Power savings are the ultimate selling point
The AI data center buildout has consumed hundreds of billions of dollars in chips, power, and cooling infrastructure. With power costs rising and electricity availability becoming a limiting factor for new data centers, companies are scrutinizing every watt. A single hyperscale cluster can draw tens of megawatts, and a meaningful portion of that is consumed by interconnect switches, retimers, and optical transceivers.
Point2's pitch is that every watt saved on interconnect is a watt that can go to compute. If e-Tube truly reduces the power consumption of interconnects by a factor of three compared to optics, a data center operating at 50 MW could potentially shift several megawatts from data movement to actual computation. Over the lifetime of a facility, that translates into significant cost savings and potentially improved model training times.
The company's technology also addresses mechanical challenges. Hyperscale data centers are becoming so cable-dense that airflow is obstructed and maintenance is difficult. A 5x weight reduction and 2x volume reduction would make cable routing easier, improve airflow for cooling, and allow operators to pack more compute into the same physical space. That is a crucially important benefit in older facilities where floor space and cooling capacity are already fixed.
The execution question remains
Despite the promising metrics and strong investor support, Point2 faces significant execution risks. Hyperscalers are notoriously conservative about adopting new interconnect technologies, because a failure at that layer can take down entire training clusters. The technology must demonstrate not only performance in a lab, but reliability in production environments, with failure rates comparable to or better than existing components.
Another challenge is manufacturing scale. Plastic waveguides must be produced with high precision and at enormous volumes to meet the needs of data center operators. The materials used for the waveguides need to maintain signal integrity over time, under temperature variations and electromagnetic interference. Point2 has not disclosed detailed technical specifications or third-party validation, so it remains to be seen whether the claimed 10x reach, 3x power savings, and 1,000x latency improvement can be validated in real-world deployments.
There is also the broader competitive landscape. Copper manufacturers are actively developing advanced materials and active retiming techniques to extend reach. Optical vendors are working on co-packaged optics that integrate lasers into the same package as switches, reducing power and cost significantly. As those technologies mature, the advantage that e-Tube currently claims may narrow or shift.
Still, the investor list suggests that the industry is willing to bet on a new approach. With Nvidia and Arm both backing Point2, the technology has a credible path into future platforms. The funding round will enable Point2 to hire engineering talent, expand manufacturing capabilities, and run qualification programs with early customers.
For a world where AI systems are growing faster than the infrastructure that connects them, Point2 represents a bold attempt to solve the wiring problem with a novel physical medium. Whether RF over plastic waveguides will become the standard for rack-scale AI compute is still uncertain, but the momentum behind the company makes clear that many of the industry's most important players believe copper and optics have reached their limits. The next few years will show whether e-Tube can live up to its promises and become a foundational building block of the AI era.