Point2 Raises $136M to Tackle AI’s Copper and Optical Wiring Limits

Point2 Technology has reportedly raised $136 million to challenge a stubborn AI infrastructure problem: moving data between accelerators without copper’s reach limits or optics’ latency penalty. The Google News headline puts a large number on that wager, but the more important story sits inside the cable.

Point2 wants to place radio-frequency signaling inside plastic waveguides, creating an active radio cable for rack-scale AI systems. The company says its e-Tube platform combines copper-like operation with longer reach, thinner cabling, and substantially lower latency than optical links.

That pitch targets a gap between two established technologies. Copper remains familiar, inexpensive, and fast over short distances. Optical fiber travels farther, but electrical-to-optical conversion adds power, components, and operational complexity.

Point2 is betting that AI clusters need a third physical connection. Its proposed answer carries millimeter-wave radio signals through a shielded plastic guide, rather than electrical current through copper or light through glass.

The funding figure also requires context. Point2’s latest official announcement says an April 2026 Series B extension brought that round’s total to $76 million. The broader $136 million figure in the Google News headline appears to refer to cumulative capital, not one newly announced financing.

That distinction matters because Point2 is no longer financing a laboratory idea alone. It is funding productization, manufacturing preparation, validation, and the difficult transition into hyperscale infrastructure.

Google News Puts Point2’s Funding Behind a Physical Bottleneck

Point2’s financing matters because AI performance increasingly depends on the links between chips, not only the chips themselves.

Point2 announced its latest Series B extension on April 21, 2026. Maverick Silicon led the extension, while NVentures, Nvidia’s venture investment arm, and UMC Capital participated.

According to the company’s funding announcement, the transaction brought total Series B funding to $76 million. Point2 said it would use the capital to move its active radio cable platform toward deployment.

The company also plans near-package and co-packaged versions of e-Tube. Those designs would move the radio conversion hardware closer to accelerators or switches, reducing the short electrical path between a processor and its external connection.

Point2’s public funding history includes earlier investments from Bosch Ventures, Molex, and other backers. The $136 million total circulating through Google News likely combines several financing events, although Point2’s latest release does not state that cumulative number.

This is not a minor accounting detail. A single new $136 million round would imply a different level of investor commitment than $136 million accumulated over several years.

The underlying signal remains substantial. Nvidia’s investment arm, a semiconductor foundry investor, and major connector companies have all placed themselves around Point2’s commercialization effort.

Those relationships offer more than capital. UMC connects the company to semiconductor manufacturing knowledge. Molex and Foxconn Interconnect Technology bring experience producing cables and connectors at scale. NVentures provides proximity to the accelerator company shaping many rack-scale requirements.

Point2 says the money will support engineering hires, system development, and deployment work. That is where interconnect startups often face their hardest test.

A cable can perform well in a demonstration and still fail as a data-center product. Customers require predictable error rates, qualified connectors, manageable thermals, redundant suppliers, and years of reliability data.

Point2 must therefore convert technical measurements into an infrastructure component that operators trust. Its investors suggest the company understands that challenge, but they do not eliminate it.

The news is best read as a commercialization milestone. It does not establish that active radio cables have displaced copper or optics in production AI systems.

That gap between investment and adoption defines the rest of the story. Point2 has financing and strategic allies, but incumbents still control the racks.

AI Racks Are Running Toward the Copper Cliff

The pressure comes from accelerator density, which forces copper to carry more data across increasingly difficult physical layouts.

Scale-up networking connects processors so closely that they behave like one larger computing system. Unlike scale-out networking between servers, these links prioritize extremely low latency and tightly coordinated data movement.

Copper has traditionally served that job. It avoids optical conversion, uses familiar connectors, and offers predictable behavior across short links.

However, higher signaling frequencies expose copper’s physical limitations. Current concentrates near a conductor’s surface, a phenomenon called the skin effect, which increases effective resistance as frequency rises.

Signal loss, noise, and timing distortion also increase. Engineers compensate with thicker conductors, shorter cables, stronger equalization, and active electronics placed at each end.

Active electrical cables use retimers, chips that reconstruct degraded data and timing signals before retransmission. They extend copper’s usable range, but consume power and occupy valuable space around crowded accelerator trays.

IEEE Spectrum described the approaching limit as the copper cliff. At terabit-class rates, copper links tend to become shorter, thicker, and more energy intensive.

That collision arrives as rack-scale systems grow. Nvidia’s GB200 NVL72 architecture connects 72 processors within one system, while later designs are expected to coordinate even larger accelerator groups.

A larger compute domain creates more links, longer paths, and tighter cable-routing constraints. Liquid cooling hardware competes for the same physical space.

Copper suppliers are not standing still. Credo and other interconnect vendors continue improving active electrical cables, signal processors, and retimers.

IEEE Spectrum reported that Credo had developed an 800-gigabit active electrical cable reaching seven meters. That distance addresses many current rack and multi-rack layouts.

Copper’s mature supply chain remains a major advantage. Operators understand its failure modes, technicians can handle it, and system builders have years of deployment experience.

Point2 must therefore beat a moving target. It is not competing against passive copper frozen at an old performance level.

The company’s opportunity appears when extending copper requires too much power, bulk, or signal-processing complexity. Radio-guided links could create more design space without forcing an immediate move to optics.

This matters to cloud operators because an interconnect is not merely a cable expense. Its power draw affects cooling capacity, rack density, and the number of accelerators a facility can operate.

A failed link can also interrupt a distributed training job. The cost of instability grows with the time and compute already invested in that run.

For accelerator vendors, the wiring problem constrains product architecture. A chip can deliver more compute, but the system gains little if processors cannot exchange model states and intermediate results quickly enough.

Point2’s bet is that physical design pressure will force buyers to consider a new medium. The company does not need copper to disappear everywhere.

It needs copper to become unacceptable across enough high-bandwidth, short-to-medium-distance connections. That threshold will determine whether e-Tube becomes a product category or remains a specialist option.

RF Waveguides Try to Split the Difference

Point2’s central mechanism replaces copper conduction with guided radio waves while retaining ordinary silicon at each cable end.

An e-Tube cable contains multiple plastic dielectric waveguides. A dielectric waveguide channels electromagnetic energy through an insulating material, rather than carrying current through a metal conductor.

At each end, a silicon transceiver converts digital electrical data into modulated millimeter-wave signals. Another transceiver reverses the process at the destination.

Point2’s reported 1.6-terabit design uses eight e-Tube cores. The transmitters operate at 90 and 225 gigahertz, allowing each guide to carry multiple high-speed channels.

The company says the completed cable measures 8.1 millimeters across, about half the volume of a comparable active copper cable. It has also described cable reach of up to seven meters for this configuration.

Point2 presents broader comparisons in its materials. Its April announcement claims up to ten times copper’s reach, one-fifth of its weight, and half its cable volume at comparable cost.

Against optics, Point2 claims one-third of the power, one-third of the cost, and one-thousandth of the latency. These are company comparisons, not neutral benchmarks covering every implementation.

The latency claim concerns conversion overhead, not light’s travel speed. Optical links require lasers, modulators, photodetectors, and associated processing to translate between electrical and optical domains.

An RF link also performs conversions. Point2 argues that direct modulation through its mixed-signal chips adds far less delay than an optical transceiver.

Its approach has another practical attraction. The radio chips can use mature semiconductor processes instead of specialized photonic fabrication.

Researchers associated with Point2 and the Korea Advanced Institute of Science and Technology have demonstrated related circuitry using 28-nanometer CMOS. That process is widely understood and does not depend on a leading-edge manufacturing node.

The waveguide may also fit existing cable production methods. Molex has reportedly produced the design without major changes to its manufacturing lines, while Foxconn Interconnect Technology has partnered with Point2.

Those details strengthen the manufacturing case, but they do not prove high-volume economics. Production yield, connector assembly, shielding consistency, testing time, and supplier capacity will influence the actual cost.

Point2 is also working with Keysight on high-speed validation. Their testing collaboration uses bit-error-rate equipment and oscilloscopes to characterize multi-terabit links.

The companies have discussed testing 120-gigabaud PAM4 signals for 1.6-terabit connections. PAM4 encodes two bits in each symbol using four voltage levels, increasing bandwidth while making signal integrity harder to maintain.

Keysight is also supporting early work around 3.2-terabit interfaces. That forward-looking target matters because an interconnect must survive more than one hardware generation.

The mechanism gives Point2 a coherent technical position. Radio waveguides aim to keep copper’s electronic simplicity while escaping copper’s high-frequency loss.

They also seek optical-like reach without placing lasers inside every short accelerator connection. The result is not wireless networking, since the radio energy remains confined within the cable.

That distinction prevents a common misunderstanding. e-Tube does not send data through open air across a data hall.

It uses radio-frequency physics inside a controlled, shielded path. From an operator’s perspective, it remains a pluggable wired connection.

The Real Opponent Is Optics, Not Copper

Point2 can extend copper’s role, but its larger strategic fight concerns whether optics must move all the way to the accelerator.

Copper already dominates short scale-up links. Point2’s first task is winning connections that copper can no longer serve efficiently.

The larger prize sits closer to the processor. Nvidia, Broadcom, and photonics suppliers are developing optical systems that place optical components beside switches and, eventually, accelerators.

Co-packaged optics integrates optical engines within the same package as a switching chip. Short electrical traces connect the processor to those engines, while fiber handles the longer path.

This architecture reduces the electrical distance and can improve bandwidth density. It also removes large pluggable optical modules from a switch’s front panel.

Optics has an inherent reach advantage. Fiber can carry high-bandwidth signals across rows, buildings, and campuses where radio waveguides are not intended to compete.

The contested zone lies inside racks and tightly coupled systems. There, extreme reach matters less than latency, power, serviceability, and connector density.

Point2 argues that putting lasers near expensive processors introduces unwanted heat and reliability concerns. Optical alignment also requires precise manufacturing, while temperature changes can affect photonic behavior.

Optical vendors have spent years addressing those issues. Their technology already operates across production networks, and major chip companies are investing heavily in integration.

Point2 cannot win by comparing a finished RF cable with an outdated optical module. Buyers will compare e-Tube with improving linear pluggable optics, co-packaged optics, and other low-power designs.

Radio also faces competition from another startup, AttoTude. Its approach carries terahertz signals through narrow dielectric waveguides at higher frequencies than Point2 uses.

AttoTude has demonstrated 224-gigabit transmission over four meters at 970 gigahertz, according to IEEE Spectrum. It has discussed potential reaches near 20 meters.

The company’s system was less integrated when those results were reported. Point2 appears closer to producing a complete pluggable cable, giving it an important timing advantage.

Still, AttoTude shows that Point2 does not own the radio-waveguide category. More than one architecture could occupy the space between copper and optics.

The central battle is therefore a route decision. System designers must choose how long to preserve electrical signaling before converting data into another physical form.

Optical suppliers want that conversion to occur increasingly close to the processor. Point2 wants radio to take over before copper fails, delaying or avoiding optics within the scale-up domain.

This opponent map clarifies why NVentures matters. Nvidia can observe several interconnect paths without committing its entire roadmap to one startup.

An investment provides access and influence, but does not equal a product selection. Nvidia continues developing and promoting optical networking alongside copper-based rack systems.

Point2 must earn sockets through qualification, not investor logos. Hyperscalers will likely test competing media against their own workloads, temperatures, cable geometries, and failure requirements.

The company’s advantage is that it offers an incremental deployment path. A pluggable active radio cable can fit familiar system boundaries before near-package versions arrive.

That path could make evaluation easier. Buyers may test e-Tube as a cable first, then consider deeper integration after reliability data accumulates.

If that sequence works, Point2 can enter without asking customers to redesign every processor package. It could also build operational confidence before challenging co-packaged optics directly.

The Performance Claims Still Need Production Proof

The unresolved issue is not whether radio can carry data through plastic, but whether thousands of links can operate reliably inside real AI systems.

Point2’s comparison numbers are striking. Ten times copper’s reach and one-thousandth of optical latency create an appealing headline.

Yet each ratio depends on the reference product, link length, data rate, error correction, and system boundary. A comparison against one optical transceiver cannot describe every optical architecture.

Power accounting also needs consistent boundaries. Measurements may include cable-end chips while excluding host-side electronics, cooling overhead, or management components.

Latency can be presented as conversion delay, end-to-end link delay, or application-visible communication time. These values answer different questions.

Point2’s claims should therefore remain claims until customers or independent laboratories publish comparable results. The Keysight work is useful, but the announced collaboration does not substitute for public qualification data.

Bit-error rate deserves particular attention. Faster signaling creates less margin for noise, connector variation, reflections, and manufacturing defects.

Forward error correction can recover corrupted data, but it consumes power and adds latency. A low raw error rate could let Point2 use lighter correction, while a noisy deployment could weaken its latency advantage.

Thermal behavior is another open question. AI racks contain accelerators, power systems, cooling components, and cables operating under sustained load.

A laboratory cable may perform differently when routed beside hot components or bent through a dense chassis. Plastic materials and RF characteristics must remain stable across the supported temperature range.

Electromagnetic isolation also matters. The waveguides are shielded, but operators will want evidence that many adjacent cables do not create harmful crosstalk.

Connector durability presents a more ordinary risk. Data centers require repeatable insertion, secure mating, field replacement, and diagnostic visibility.

A novel medium may need new test procedures for technicians. Operators will ask how they identify a failing guide, inspect a connector, and replace a link without disrupting neighboring equipment.

Supply resilience poses another challenge. Mature copper and optical ecosystems contain multiple component vendors, manufacturing regions, test tools, and repair processes.

Point2 has capable partners, but a new category begins with fewer qualified sources. Large buyers often resist dependence on one transceiver design or one specialized material supplier.

Standards will influence adoption. Proprietary cables can enter tightly controlled systems, but standardized electrical interfaces, form factors, and management behavior make broader purchasing easier.

Point2’s pluggable approach helps because it can connect through established interfaces. However, interoperability involves more than fitting into the same port.

Customers will expect predictable link training, telemetry, firmware behavior, error reporting, and compatibility across switches and accelerators. Qualification cycles can consume years.

The financing gives Point2 time to address these concerns. It does not shorten the evidence requirements imposed by operators running expensive clusters.

This is where the $136 million headline becomes less celebratory and more practical. Building a semiconductor-enabled infrastructure category requires sustained capital before meaningful volume revenue appears.

Teams evaluating the technology should preserve test results, supplier disclosures, and architecture decisions in a searchable engineering knowledge base. The important comparison will evolve across multiple hardware generations.

Point2’s strongest evidence will not be another percentage. It will be a named production deployment with disclosed capacity, stable operation, and repeat orders.

Until then, e-Tube remains a credible contender with an incomplete commercial record. That is normal for infrastructure entering qualification, but it limits how confidently anyone can declare a winner.

Three Signals Will Decide Whether Point2 Escapes the Lab

Point2’s next phase will be judged by manufacturing, customer qualification, and performance at the next interface speed.

The first signal is a production-ready 1.6-terabit active radio cable. IEEE Spectrum reported that Point2 planned to begin manufacturing the supporting chips during 2026.

A finished product should provide more than peak throughput. Buyers need supported reaches, power limits, error rates, connector specifications, environmental tolerances, and expected service life.

Volume availability would strengthen Point2’s claim that mature silicon and existing cable lines create a manufacturing advantage. Repeated delays would suggest that integration remains harder than the architecture implies.

The second signal is an identifiable hyperscale or accelerator deployment. Point2 says it is engaging large customers, but public material has focused mainly on partners and validation.

A customer qualification announcement would show that e-Tube survived requirements beyond Point2’s own laboratory. A revenue-generating deployment would carry even more weight.

The best evidence would include the link’s role within a system. Accelerator-to-switch, accelerator-to-accelerator, and rack-to-rack connections impose different constraints.

Operators should also watch whether Nvidia’s relationship progresses beyond NVentures. Design adoption within an Nvidia platform would strongly validate Point2’s route.

The absence of such adoption would not automatically invalidate e-Tube. Nvidia supports multiple suppliers and develops competing copper and optical technologies.

The third signal is credible 3.2-terabit validation. Point2 and Keysight have already described early research at that speed.

A successful demonstration with useful reach, manageable power, and low error rates would show that the platform scales beyond its first product generation. That would reinforce its position against improving copper.

A result requiring heavy error correction or sharply reduced reach would weaken the central thesis. It could confine e-Tube to narrower applications.

Optical progress must be tracked at the same time. Broadcom, Nvidia, and photonics vendors will keep reducing power and simplifying packaging.

Point2 does not compete in isolation. Its advantage must improve faster than alternative technologies within the specific distances used by rack-scale AI.

Copper may also retain more territory than expected. Better materials, active electrical cables, and advanced retimers can postpone the point where system designers need another medium.

That is why the most useful question is not whether RF beats copper and optics universally. No single medium wins every distance, cost target, and reliability requirement.

The real question is whether AI scale-up systems contain a large, durable connection zone where copper is insufficient and optics is unnecessarily complex. Point2 has built its company around that zone.

The Google News framing captures the ambition but compresses the uncertainty. Capital, technical claims, and strategic partners make Point2 worth watching, yet deployments will determine its place in the stack.

For developers and AI users, the impact will arrive indirectly. Better interconnects can raise accelerator utilization, support larger coherent systems, and reduce infrastructure power spent moving data.

Enterprise buyers should watch whether those gains translate into better service capacity or lower operational constraints. Hardware improvements do not automatically produce cheaper AI access.

Point2 has identified a real wiring problem and proposed a technically plausible third path. Now it must prove that radio-guided plastic cables work reliably at the scale where AI infrastructure decisions become permanent.

Watch the product specifications, the first named customer, and the 3.2-terabit results. Those signals will reveal whether the funding bought Point2 a bridge into production or merely more time beside the copper cliff.