EDA
System
IP
SolvNetPlus
Training & Education
With AI systems driving interconnect speeds to 224G and beyond, the technology industry has been fixated on the supposed demise of copper cables in data centers. Physics has spoken, the common refrain goes, and the future is optical.
But according to Kandou AI, an AI company with deep roots in high-speed connectivity and SerDes design, copper has much more to give.
“People say copper is reaching the end of its life,” says Amin Shokrollahi, CTO of Kandou AI. “That’s simply not true.”
He points to Shannon’s channel capacity theorem, which defines the maximum rate at which information can be transmitted over a communication channel.
“The Shannon limit shows that there is a lot of untapped capacity in existing copper links,” Shokrollahi explains. “Eight to 20 times more in some channels compared to what is being used today.”
Unlocking that untapped potential wouldn’t just give copper new life — it would open one of the biggest opportunities in AI infrastructure today. One that Kandou AI is bullish on seizing.
AI Doesn’t Scale by Accident
Explore the IP that enables high-performance, scalable AI systems
Hitting the AI walls
Modern AI workloads place extraordinary demands on data center systems. And the faster GPUs become, the hungrier they get. This is when data pipelines become crippling bottlenecks.
“Memory and interconnect bottlenecks are increasingly becoming the limiting factor, especially in AI, where you need access to a lot of memory,” says Srujan Linga, CEO of Kandou AI. “Traditional ways of designing interconnect solutions are hitting the wall.”
Three walls, in fact. Today’s AI infrastructure is increasingly running up against:
- The memory wall — GPU performance outpaces the memory subsystem, starving compute units of data.
- Rack‑scale bandwidth limits — multi‑meter copper cables struggle under the extreme speeds AI now demands.
- Reliability over long‑running workloads — hyperscale environments with mission-critical, always-on AI applications demand not just speed, but resilience.
These challenges have led many to predict an inevitable shift to all-optical infrastructure. But optics come with their own cost, complexity, and power burden.
If copper can be pushed further, it can provide additional options and flexibility for those building AI infrastructure.
Giving copper new life
Rather than accepting today’s copper‑signaling limitations as fixed, Kandou AI is applying deep expertise in both semiconductor design and information theory to fundamentally rewrite the rules.
“We use coding techniques no one else has in the world,” says Linga, “to get higher bandwidth, longer distance, and lower power for copper interconnects.”
Kandou AI’s approach doesn’t fight crosstalk and interference — it harnesses them. By applying advanced coding across copper channels, Kandou AI’s signaling schemes extract far more of the medium’s theoretical potential than traditional SerDes ever could.
“Kandou AI has a better handle on copper than almost anybody else,” claims Jay Shenoy, vice president of architecture at Kandou AI. “We have a physics‑based differentiator, a unique combination of math and analog circuits.”
This differentiator is being used to develop innovative, energy-efficient connectivity products that address the memory wall in AI systems as well as high-speed connectivity solutions spanning 224G, 448G, and beyond.
“We are building revolutionary products that will transform AI system design for hyperscale customers,” says Linga.
Leveraging tightly integrated, silicon-proven IP
While Kandou AI’s differentiators lie in their physics-driven interconnect IP, SerDes, and system design, the company is leveraging Synopsys IP to accelerate the development of its products.
“Synopsys provides full stack IP that is tightly integrated and silicon proven,” says Shenoy.
That combination is enabling Kandou AI to develop and deliver its innovations at hyperscale speed.
“Synopsys is a key collaborator enabling us to reach market as rapidly as possible,” says Subhash Roy, vice president of product and strategy at Kandou AI. “By leveraging their IP, we can keep up with — and even exceed — the accelerating innovation schedules of hyperscalers.”
Free from the burden and time commitment of designing and integrating foundational IP layers, Kandou AI can concentrate fully on its unique physics‑driven signaling technologies that set it apart.
“Systems are very complex,” Shenoy says. “A company like Kandou AI can focus on our differentiators and rely on Synopsys for critical subsystems.”
Continue Reading
Introducing the World’s First AI-Powered Analog Layout Synthesis
Jumpstart Chip Design with Digital Instance 2.0: Ready to Use, Configurable RTL2GDS Flow on Cloud
OFC 2026: IEEE 802.3 Launches a Study Group for 400 Gbps Per Lane Ethernet for AI Networks
