In October 2026, David Huff of Genuine Optics stood before engineers at the OCP Global Summit with two competing visions for AI networking.
One promised better energy efficiency by moving light closer to the chip.
Then Huff displayed his readiness scorecard.
The more efficient architecture scored zero.
The Most Efficient Optical Design Scored Zero
The numbers initially favored co-packaged optics, or CPO.
According to Genuine’s architectural estimates, conventional fully retimed optical modules consume approximately 16 picojoules per bit. Near-packaged optics (NPO) could reduce that to 4–6 pJ/bit.
CPO could reach just 1–4 pJ/bit.
These were supplier estimates, not identical-condition product measurements. Still, the engineering logic was compelling: shorter electrical connections generally require less power to maintain signal integrity.
Then came the surprise.
Genuine awarded CPO a zero for reliability, availability, and maintainability (RAM).
Not because CPO had demonstrated zero reliability. This was Genuine’s qualitative assessment, reflecting its own industry position.
The concern was practical.
When a conventional optical module fails, a technician can replace it. When optics share an expensive switch-chip package, repairing or replacing the affected assembly can become more complicated.
The architecture that saves the most electricity might not deliver the lowest lifetime operating cost.
But Huff’s argument went beyond maintenance.
The Factory May Decide the Winner Before the Lab Does
Genuine’s proposed 6.4T NPO engine sits near the switch chip, rather than inside its package.
It uses external lasers and offers three connector configurations: LGA, CPX, and Mezz.
That flexibility matters when customers haven’t agreed on a common interface.
Instead of redesigning the entire optical engine for every system, Genuine aims to adapt its connections while reusing core technology.
Then the problem gets bigger.
In September 2026, the Open CPX group released specifications supporting standardized 6.4T and 7.2T optical modules.
Importantly, those specifications cover both NPO and CPO.
The next competitive advantage may come from making optical engines easier for multiple suppliers to manufacture, qualify, and replace—not simply making them faster.
But a published standard doesn’t guarantee interoperability, manufacturing yield, or volume shipments. Genuine’s early-2027 commercialization target remains a plan, not a proven production milestone.
And CPO could narrow the gap through improved packaging and serviceability.
For now, the industry faces two different engineering challenges.
CPO seeks to shorten the distance between electronics and light. NPO seeks to shorten the distance between the laboratory and the factory.
The winner may be whoever can deliver at scale first.
CONTINUE THE RESEARCH
TECH VISION MAGAZINE · NOVEMBER 3, 2026
Why the Best Optical Technology Might Lose to the Runner-Up
Explore the complete engineering analysis of 6.4T NPO optical engines, connector architectures, manufacturing constraints, and 2027 standardization milestones—and discover what could determine the next generation of AI optical interconnects.