Is Photonics the Bottleneck or Is Packaging the Bottleneck?
What a short exchange between Jensen Huang and Martijn Heck reveals about the future of AI infrastructure
As artificial intelligence scales toward ever larger computing clusters, photonics is increasingly viewed as a critical technology for the next generation of AI infrastructure. Yet a growing debate suggests the real bottleneck may not lie in photonic chips themselves, but in the industrial systems required to package, assemble and integrate them at scale.
Artificial intelligence has become a story of scale. Every new generation of AI models requires larger computing clusters, greater networking capacity and exponentially increasing volumes of data moving between processors. As a result, attention is rapidly shifting beyond the GPU itself toward the infrastructure that connects modern AI systems.
This shift helps explain why Nvidia has invested billions of dollars across the photonics ecosystem in recent months.
Optical interconnects are increasingly viewed as one of the few viable solutions capable of supporting the scale demanded by future AI infrastructure. As traditional copper connections approach practical limits in energy consumption, heat generation and bandwidth, photonics is emerging as a critical enabling technology.
Yet a recent exchange between Nvidia CEO Jensen Huang and photonics expert Martijn Heck reveals a deeper question beneath the headlines.
The question is not whether photonics matters. The question is where the actual bottleneck resides.
The Photonics Argument
When discussing the future of AI infrastructure, Jensen Huang recently warned that silicon photonics capacity requirements may soon exceed global production capabilities.
The concern is understandable. As AI clusters continue to grow, communication increasingly becomes the dominant challenge. Every additional processor requires more connections, more bandwidth and more energy to move information across increasingly complex systems.
Photonics offers an attractive solution. Instead of transmitting data through electrical signals, photonic systems use light. The result is lower power consumption, reduced heat generation and significantly higher bandwidth.
For an industry attempting to connect hundreds of thousands of processors into unified AI systems, these advantages are becoming increasingly difficult to ignore.
From this perspective, more photonics appears to be the obvious answer. Yet not everyone agrees with the diagnosis.
The Heck Correction
Commenting on a discussion surrounding Huang’s remarks, Professor Martijn Heck of Eindhoven University of Technology offered a concise but revealing observation.
“I don’t believe this. I think Jensen Huang is confused. Maybe he is referring to the assembly and packaging and not the actual chips?”
Prof. Martijn Heck
Professor of Integrated Photonics
Eindhoven University of Technology (TU/e)
Source: LinkedIn discussion
At first glance, the distinction may appear technical. In reality, it changes the entire discussion. Heck is not arguing against photonics. Nor is he suggesting that Nvidia’s investments are misplaced.
Instead, he appears to be questioning where the actual industrial constraint exists. Because manufacturing a photonic chip and deploying a photonic system are not the same thing.
Beyond the Chip
Building a photonic chip is only one stage of a much larger process. The chip must be tested. Lasers must be aligned. Optical fibers must be attached. Components must be packaged, assembled and integrated into larger systems capable of operating reliably for years.
Unlike conventional semiconductor devices, photonic systems introduce entirely new manufacturing challenges. Tiny alignment errors can dramatically reduce performance. Optical testing procedures are more complex than electrical testing. Production yields can fall rapidly when technologies move from laboratory environments to industrial scale.
In other words, producing the chip may not be the hardest part. Deploying the chip may be. This is where the discussion shifts from photonics toward packaging.
The Bottleneck Behind the Bottleneck
The semiconductor industry has encountered this pattern repeatedly. Technological breakthroughs often shift bottlenecks rather than eliminate them.
For years, advanced lithography represented the industry’s primary constraint. Then came substrates. Then advanced packaging emerged as one of the most strategically important capabilities in semiconductor manufacturing.
Photonics may now be following a similar trajectory. The challenge may not be the availability of photonic designs or even photonic wafers. The challenge may be the industrial ecosystem required to transform those components into scalable infrastructure.
Seen from this perspective, Huang and Heck may actually be describing different layers of the same problem.
Huang is looking at future demand. He sees AI infrastructure expanding beyond what current optical ecosystems can support.
Heck is looking at industrial reality. He sees the practical challenges involved in turning photonic components into reliable, mass-produced systems.
Both observations can be true simultaneously.
Why This Matters for Europe
The distinction carries important implications for Europe. Much of the public debate surrounding semiconductors continues to focus on fabrication capacity and manufacturing scale. Yet if packaging, assembly and integration become the next critical bottlenecks, the strategic landscape changes considerably.
Europe possesses deep expertise in photonics, precision engineering, industrial automation and semiconductor equipment. Around Eindhoven, Leuven and other technology clusters, researchers and companies have spent decades building capabilities that sit precisely at the intersection of optics and advanced manufacturing.
The lesson may be similar to the one Europe learned through ASML. ASML became strategically indispensable not because it manufactured chips, but because it controlled a critical enabling technology required to manufacture them.
A similar opportunity may now be emerging elsewhere in the value chain. Just as ASML became indispensable through lithography, Europe’s next strategic position may emerge around the highly specialized machinery, testing systems and integration technologies required to package and scale photonic infrastructure.
If AI increasingly depends on photonic systems, the companies enabling those systems may become just as important as the companies designing the chips themselves.
Looking Beyond Photonics
The debate between Huang and Heck ultimately points toward a broader reality. Artificial intelligence is no longer simply a software story. Nor is it merely a chip story. It is becoming an infrastructure story.
Energy systems.
Cooling systems.
Networking systems.
Photonics.
Packaging.
Integration.
The future of AI will likely be shaped by the interaction of all these layers. The history of technology is often written around breakthroughs. The history of industries is usually written around bottlenecks.
The question facing artificial intelligence may no longer be how to build more powerful chips. It may be how to industrialize the systems surrounding them.
Because the next bottleneck may not be the chip. It may be everything that comes after it.
Credit
Illustration by ChatGPT for Altair Media
Caption
The future of artificial intelligence may depend less on processors and more on the systems connecting them. As photonics moves from research labs into industrial deployment, packaging and integration are emerging as critical layers of next-generation AI infrastructure.
