The Wires Are Clogged
By 2025, the capital expenditures of the world’s largest cloud companies pouring into data centers will exceed hundreds of trillions of won annually.
A single NVIDIA H100 GPU costs 40 million won, and clustering thousands of them can easily push the cost of one AI training system into the hundreds of billions of won.
People might think most of that cost is concentrated in the compute chips.
They’re wrong.
The most expensive element isn’t the chip, but the time the chips spend waiting for each other.
When thousands of GPUs train a massive neural network, they must exchange intermediate calculation results with each other at every computational step.
Even if one GPU finishes its calculation, it cannot proceed to the next step without receiving a synchronization signal from other devices.
This waiting state is called ‘stall’. And this stall occurs in the wires.
Specifically, it happens at this moment.
At the next-generation ultra-high-speed transmission standard of 1.6 Tbps bandwidth, the effective transmission distance of existing copper cables collapses to less than 1 meter due to the laws of physics.
However, the average height of a real server rack is over 2.5 meters.
-> This means that even a cable connecting the bottommost GPU to the topmost network switch within the same rack cannot be physically connected.
This is what the industry calls the ‘Copper Cliff’.
And optical cables, proposed as an alternative in front of this cliff, require 15-20W of power per port instead of solving the connection distance problem.
For hyperscale data centers operating tens of thousands of ports, this is an electricity bill bomb and a major cause of carbon emissions.
Copper is too short. Light is too expensive.
From the gap of this dichotomy, plastic emerged.
First Wall: Copper Swallows High Frequencies
The Law of Physics: The Skin Effect
Why do copper cables suddenly become useless? The answer lies not in materials science but in electromagnetics.
When alternating current flows through a conductor, as the frequency increases, the charge is pushed towards the surface of the conductor, away from the center. This phenomenon, known as the skin depth (δ), is inversely proportional to the square root of the frequency f.
δ = √(ρ / πfμ₀μᵣ)
Intuitively, if the frequency quadruples, the effective cross-sectional area through which current can flow is halved.
A reduced cross-sectional area leads to increased resistance, and increased resistance causes the signal to dissipate as heat.
In the ultra-high-frequency band above 100 GHz, copper conductors essentially act not as signal transmission media but as heaters that incinerate signals.
This causes the effective distance of passive copper cables in a 1.6 Tbps transmission environment to not exceed 1 meter.
In the 3.2 Tbps generation, this distance further decreases.
The World Beyond the ‘Copper Cliff’
Field engineers have historically used thicker cables to overcome this limitation.
Thicker conductors reduce resistance. However, thicker cables become heavy and stiff, blocking the rear of server racks.
Blocked exhaust vents reduce cooling efficiency, and internal temperatures exceed critical thresholds.
The weight and stiffness of the cables can also physically damage port connectors.
Active Electrical Cables (AECs), introduced as a solution, increased the reach to 2-2.5 meters by adding signal regeneration chips (retimers) at the cable ends. However, that’s all.
2.5 meters is still insufficient to meet the wiring demands of the 1.6 Tbps generation data centers.
Copper cable evolution has already hit the ceiling of physical laws.
There are no faster cables. There are no thinner copper cables. All that exists are higher frequencies and the consequently narrowing effective cross-sectional area of the conductor.
Second Wall: Light is Expensive
The Cost of Optical Fiber Cables
Optical fiber cables transmit signals through light refraction, not electromagnetic loss, so they are completely free from the skin effect that copper suffers from. They can easily span kilometers.
So why aren’t hyperscale data centers fully switching to optical cables?
Converting electrical signals to light requires devices.
A laser diode emits light, a driver modulates it, an optical DSP restores the signal, and a transimpedance amplifier (TIA) at the receiver converts it back to voltage.
This combination of components forms an optical transceiver module.
A single 800 Gbps optical transceiver consumes an average of 15-20W of power.
For a hyperscale data center with tens of thousands of ports, the power consumed by the interconnect layer alone is equivalent to that of a massive power plant.
Cost is also an issue.
Optical transceivers require compound semiconductors based on Indium Phosphide (InP) or Gallium Arsenide (GaAs) and silicon photonics packaging technology.
Their unit cost is 4 to 10 times higher than comparable passive copper cables.
Reliability is also a concern.
Laser diodes age faster in high-temperature and high-humidity environments.
Thermal shock near server rack exhaust vents shortens laser lifespans, and in distributed parallel computing environments, a single connection failure can halt the entire training session, forcing a restart from the last checkpoint. It’s not uncommon for training jobs that ran for days on thousands of GPUs to be nullified by the failure of a single optical transceiver’s laser.
Limitations of Co-Packaged Optics
Co-Packaged Optics (CPO) is an attempt to solve this problem by placing the chip and optical components in the same package.
This minimizes the distance electrical signals travel on the substrate, improving power efficiency.
However, CPO significantly increases manufacturing complexity and cost.
Integrating silicon logic and optical components within the same package is one of the most challenging tasks in semiconductor manufacturing, and ultimately, it does not eliminate the fundamental energy inefficiency of optical-to-electrical conversion.
The Third Material: Plastic Swallowed Millimeter Waves
In 2014, an idea from a lab at the Korea Advanced Institute of Science and Technology (KAIST) directly challenged this dichotomy.
The idea was simple: no copper, no lasers.
Shoot electromagnetic waves into a plastic tube.
Point2 Technology’s ’e-Tube’ is a system that transmits RF signals in the millimeter-wave (100-260 GHz) band by confining them within a specially molded, low-loss polymer resin dielectric waveguide.
Since there is no conductor, the skin effect does not occur intrinsically.
Since there are no lasers, there is no risk of optical-to-electrical conversion loss or laser aging.
What Happens Inside the Waveguide
The physical principle of a dielectric waveguide is similar to that of an optical fiber, but it operates with a different signal.
While optical fibers confine visible light frequencies (hundreds of THz), e-Tube confines millimeter-wave (hundreds of GHz) electromagnetic waves.
The principle is Total Internal Reflection.
A plastic core with a high refractive index is surrounded by a cladding with a lower refractive index. Electromagnetic waves injected inside cannot cross the boundary and propagate along the core.
In this structure, signal attenuation is determined solely by the dielectric loss tangent (tan δ) and dielectric constant (ε) of the plastic material, not by high-frequency metal resistance.
α_dielectric ∝ f · √ε · tan δ
The key point in this relationship, unlike with copper, is that the rate of attenuation increase as frequency rises is much gentler compared to metals.
This provides the physical foundation for signals to travel stably for 10-20 meters even in the ultra-high-frequency bands required by the 1.6 Tbps and 3.2 Tbps generations.
From TX Chip to Waveguide: The Role of the MWT Antenna
Computer systems operate with baseband digital signals.
If these signals are directly injected into a dielectric waveguide, low-frequency components will leak out past the boundaries. They won’t propagate.
Point2 Technology has equipped the plug at the end of the cable with a proprietary RFIC chipset.
-> The transmit chip (TX SoC) receives the baseband digital signal and upconverts it to a millimeter-wave analog RF signal in the 100-260 GHz band.
This generated millimeter wave is injected into the e-Tube through a ‘Microstrip-to-Waveguide Transition (MWT) antenna’ etched onto the chip surface.
The MWT is a transducer that matches the planar electrical impedance of the silicon substrate to the three-dimensional electromagnetic impedance inside the dielectric, directly injecting the electromagnetic wave into the center of the waveguide core without energy reflection or external leakage.
On the other side, the on-chip antenna of the receive chip (RX SoC) captures the incoming electromagnetic wave, and a downconversion mixer restores it to the original digital bus signal.
The host ASIC, such as a GPU or switch, doesn’t even need to know that the e-Tube is in between.
The Number 80 Picoseconds
The result of this analog conversion path is a latency of 80 picoseconds (ps).
For comparison: an optical fiber system using optical transceivers involves electrical-to-optical conversion, digital signal processing, and optical-to-electrical conversion, resulting in latencies in the microsecond (μs) range. e-Tube is 1/1,000th of that.
In distributed parallel computing, latency is not just a speed indicator. When thousands of GPUs perform synchronized operations, the slowest link determines the processing speed of the entire cluster. This is why 80 ps latency is 1,000 times more significant.
Comparison Table by Numbers
The picture becomes clear when comparing five interconnect technologies in parallel.
The space filled by e-Tube is the void between the distance that copper cables abandon and the cost that optical cables cannot bear. Its design goal is to handle rack-level server wiring in the 10-20 meter range.
The Company’s Roots, and Two Cash Cows
Point2 Technology was founded in March 2014.
It was co-founded by CEO Sean Park, who led high-speed communication system chip design at Marvell Semiconductor and Teledyne for over 13 years, and Professor Hyunmin Bae of KAIST’s Department of Electrical Engineering (currently Director of KAIST Entrepreneurship Center), along with his graduate research team. The company is headquartered in San Jose, California, with its core R&D conducted by its Korean subsidiary, forming a dual structure.
However, it took time for e-Tube technology to gain market traction.
Two business lines fill that gap.
25G Range Extender: The Hidden Bottleneck in 5G Base Station Networks
During the construction of 5G communication networks, the fronthaul optical links connecting base stations to the core network are being upgraded from the existing 10 Gbps to 25 Gbps.
-> An unexpected problem arises here.
In optical fibers, ‘Chromatic Dispersion,’ where the refractive index varies with wavelength, distorts the signal in proportion to the square of the frequency ratio.
Optical lines that could transmit 70-80 km without issue at 10 Gbps collapse to an effective distance of 15 km at 25 Gbps. It’s a paradox where faster transmission actually creates a shorter reach.
The traditional solution was to install large external dispersion compensation filter devices at each relay point. -> This involved deploying equipment worth hundreds of millions of won on-site and having technicians manually adjust it according to the line length.
Point2 Technology’s Range Extender solves this problem with a semiconductor chip the size of a fingernail, integrated within the optical transceiver module.
The Electronic Dispersion Compensation (EDC) engine restores the completely collapsed eye pattern in real-time, recovering the effective transmission distance to 60-70 km. No external devices or on-site adjustments are needed. It’s a simple replacement of the optical transceiver.
Based on this technology, Point2 Technology has signed a long-term development partnership with Japan’s Sumitomo Electric Industries and secured a substantial mass production supply contract for core ICs for 25G optical transceivers.
-> This is the company’s first revenue stream.
Smart Retimer P1B121: Half the Power, 1/20th the Latency
The second cash cow is the ‘P1B121’, a 112G PAM4 smart retimer chip for data centers.
A retimer is a chip that cleans and regenerates distorted signals at the end of high-speed electrical cables. It is essential for Active Electrical Cables (AECs) in the next-generation 800 Gbps and 1.6 Tbps bands. The problem is that competing products on the market consume too much power.
P1B121 achieves a driving power of 3.0W, approximately 50% lower than comparable competitors. Its latency is also less than 3 nanoseconds (ns), 1/20th of the industry average. This is the result of a hybrid signal optimization design that maximizes analog signal path processing.
These two numbers—3W, 3ns—mean more than just their values. In large-scale accelerator cluster environments that demand ultra-low latency synchronization, the retimer’s latency is prevented from becoming a bottleneck. At the same time, it provides headroom in the power budget.
Why NVIDIA Invested Directly in Korea
In 2024, an unusual event occurred in the global semiconductor market.
NVIDIA’s strategic venture capital arm, NVentures, made its first direct equity investment in a Korean-based semiconductor startup. The investment target was Point2 Technology.
NVentures makes strategic investments to strengthen NVIDIA’s future platforms and technology ecosystem. The goal is technology acquisition, not financial return.
This is interpreted as NVIDIA’s move to incorporate the e-Tube transmission layout as a core framework in the scale-up fabric of its next-generation accelerator architecture beyond Blackwell.
Including this investment, Point2 Technology has now completed a cumulative Series B funding of $76 million (approximately 112.1 billion won).
The investors include Maverick Ventures, UMC Capital, Bosch Ventures, and Molex. The list of investors carries the nature of building a strategic ecosystem rather than mere financial support.
The participation of Bosch, a global automotive parts giant, hints at the potential for e-Tube’s entry into the automotive electronics market.
The investment from Molex, a wire and cable manufacturer, exemplifies how production partnerships have led to capital relationships.
The 2026 BloombergNEF Pioneers award is external validation of this technology.
The Fabless Paradox: How to Dominate Global Supply Chains Without a Factory
Point2 Technology does not have manufacturing facilities. This is by design.
It does not directly produce finished e-Tube cables.
Instead, it designs RFIC chipsets that control electromagnetic modulation and licenses the geometric structure and bonding design technology of the e-Tube waveguide. -> Actual production is handled by the world’s largest connector manufacturers.
Molex, Foxconn Interconnect Technology (FIT), and Amphenol. These three companies are already proven vendors supplying trillions of won worth of products to the data center supply chains of Amazon AWS, Google Cloud, Microsoft Azure, and Meta.
For Point2 Technology to directly enter the supply chains of hyperscale data centers, it would need to pass years of rigorous supplier audit procedures.
However, by shipping finished ARC (Active RF Cable) products with e-Tube chipsets integrated, leveraging the brand and logistics networks of already established partners, this process can be shortened.
It only designs chips. Manufacturing is done by partners who have already built trust. It uses the partners’ channels. This is the secret to the high operating profit margins produced by the fabless business model.
Shift in Revenue Mix Towards 2030
Currently, Point2 Technology’s revenue comes from 25G Range Extenders and Smart Retimers. These are the fuels that sustain the company today.
However, the company’s direction points elsewhere. The internal roadmap’s core is to transition 90% of total revenue to e-Tube-based products by 2030.
There is a market structure supporting this transition.
Between 2026 and 2028, global data centers will fully transition to the 1.6 Tbps and 3.2 Tbps generations.
At this point, copper cables will inevitably be replaced by ARC (Active RF Cable) or optical cables due to their physical limitations. This is the size of the replacement market that e-Tube targets.
Beyond the external cable stage, deeper integration awaits.
NPE (Near Package e-Tube) brings ultra-thin e-Tube waveguide traces close to the silicon interface inside the accelerator board.
CPE (Co-Packaged e-Tube) integrates RFIC components within the GPU packaging substrate. At this stage, the noise from copper wiring on the substrate disappears, and electromagnetic waves are emitted directly from the accelerator chip into the dielectric tube.
Information that they have signed a non-disclosure MOU with at least one of the world’s top GPU manufacturers and are preparing a prototype demonstration in the second half of 2026 suggests that this roadmap is in the execution phase, not just a declaration.
The penetration path of e-Tube is not limited to data centers.
Terabit-class signal wiring inside ultra-thin premium display panels, noise-shielded wiring between automotive ECUs and multi-channel sensors—wherever electromagnetic interference is a problem and space is limited, the physical principles of e-Tube can be applied.
Ahead of the KOSDAQ Listing
Point2 Technology is preparing for its preliminary review application for KOSDAQ’s technology special listing in the second half of 2026.
Samsung Securities and Korea Investment & Securities have been confirmed as joint lead managers, with a target market capitalization of at least 1 trillion won.
There is a reason for the delay in the official listing schedule compared to the initial plan.
The cross-border governance structure between the US headquarters in San Jose and the Korean subsidiary has caused legally complex issues.
It required a longer coordination period than expected to simultaneously satisfy the corporate governance requirements of US federal law and the technical special listing lock-up requirements and cross-border equity settlement structures of the Korea Securities Depository and the Korea Exchange.
The funds to be raised in the pre-IPO round just before the listing are planned to be prioritized for advance payments for initial mass wafer production by foundries and expansion of on-site engineering support infrastructure in the US.
Wires Make Technology Invisible
It’s worth pausing here to consider something.
The problems that e-Tube solves are interesting in that they only work when the technology itself is invisible. The best interconnect solutions are nonexistent. GPUs don’t care what the cables are made of. The host system doesn’t realize that electromagnetic waves have passed through plastic in between.
This is the essence of infrastructure. Infrastructure functions best when it is invisible.
However, there are moments when infrastructure becomes visible. When wires get clogged, power budgets are exceeded, or latency drags down the entire cluster. This is precisely the situation the data center industry faces today. Copper is blocked, optics are too expensive, and AI continues to demand faster.
The moment of crisis when infrastructure becomes visible is precisely the condition under which new materials can emerge.
The creation of a third space between copper and optical fiber by plastic dielectrics is not simply the invention of a better cable. It is the redesign of the physical constraints of AI infrastructure.
There are many stories about AI changing the world. Stories about what materials the wires running that AI are made of are rare. If a single wire determines the actual performance of a trillion-won cluster—then the question about the company making that wire is ultimately a question about the future of AI infrastructure.
References
- Point2 Technology Inc. — e-Tube RF Dielectric Waveguide Architecture White Paper (2024)
- Open Compute Project (OCP) Global Summit — “Breaking the Copper Wall: RF Dielectric Waveguide for AI Cluster Interconnect”, Speaker: Sean Park (2024)
- KAIST Department of Electrical Engineering, Professor Hyunmin Bae’s Research Team — Theoretical paper on electromagnetic induction propagation within dielectric waveguides based on ultra-high frequency modulation
- Point2 Technology Inc. — 25G Range Extender and Sumitomo Electric Partnership Technical Report (2023)
- Bloomberg NEF Pioneers — 2026 Point2 Technology Technology Assessment (2026)
- NVIDIA NVentures — Series B Investment Announcement (2024)
- IEEE Transactions on Microwave Theory and Techniques — “Dielectric Waveguide for On-Board and Short-Reach Interconnects” (2022)
- McKinsey Global Institute — “The Energy and Materials Underpinning the AI Infrastructure Build-out” (2024)
- IEEE 802.3 Task Force — “200G/lane Electrical Signaling and Copper Cliff Analysis” (2023)
- Molex Engineering White Paper — “ARC (Active RF Cable) Co-Development Framework for 1.6T Data Center Deployments” (2024)
- Dell’Oro Group — “Data Center Interconnect Market Forecast 2024–2029” (2024)
- IEC 62149 Series — Fibre Optic Active Components and Devices: Reliability Standards for Laser Diodes in Data Center Environments
- Harvard Business Review — “Fabless Semiconductor Strategy: How Design IP Generates Outsized Returns” (2023)
- Foxconn Interconnect Technology (FIT) — e-Tube Partnership Product Roadmap Brief (2024)
- Korea Exchange (KRX) — KOSDAQ Technology Listing Special Criteria: Cross-Border Corporate Governance Compliance Guidance (2025)