Why AI Networks Need OTN DWDM, 800G & Quantum Security | Omdia Interviews PacketLight
Video TL;DR
At OFC 2026, PacketLight CEO Koby Reshef had a discussion with Omdia's Sterling Perrin, on how AI is reshaping optical transport, increasing demand for high-capacity, low-latency, and secure DWDM/OTN networks. Koby highlights PacketLight’s focus on open, standards-based solutions, the 800G evolution, and Layer 1 encryption with quantum-safe readiness, positioning the company as a flexible optical transport partner for AI, subsea, enterprise, and critical infrastructure networks.
Koby Reshef outlined how the company is positioning itself around three major shifts in optical networking: the rise of AI infrastructure, the need for secure transport, and the growing importance of open, standards-based networking. According to Reshef, PacketLight’s long-term relevance comes from staying close to customers, moving quickly, and adapting products to the needs of different vertical markets including carriers, financial institutions, utilities, video, and encryption-focused environments.
A customer-driven approach to optical transport
Reshef credits PacketLight’s 26-year longevity to a practical operating model: listen closely to customers, identify what each market needs, and develop products quickly without excessive hierarchy. He emphasized that PacketLight tailors features for specific use cases rather than building generic products detached from operational needs.
That customer focus has played out across a fast-changing optical market. PacketLight began in the era of Gigabit Ethernet and 10G chassis systems, and now delivers platforms reaching up to 8T in 1RU. In Reshef’s view, optical transport has seen one of the industry’s most dramatic technology shifts over the last two decades.
Why AI is changing transport requirements
Reshef described AI as a major force shaping transport networks, not only because of raw bandwidth demand, but because of strict performance requirements. AI workloads require large-scale movement of data during both training and inference, but the real challenge is not just throughput. Latency and jitter also matter, because idle GPUs mean wasted compute resources.
This is where PacketLight sees value in DWDM and OTN. Reshef argued that these are predictable transport layers: operators know the latency profile, and traffic is not exposed to the variable behavior of routing and switching layers. For AI environments, that predictability helps keep compute infrastructure fully utilized. PacketLight’s role, he said, is to move data from one side to the other quickly, predictably, with low latency and low jitter, while scaling fiber utilization across the C-band, L-band, Flex Grid, and higher optical rates.
The role of OTN in longer-distance AI transport
While Ethernet may dominate very short-reach environments inside the data center, Reshef made the case that OTN becomes essential at longer distances. He pointed to distances above roughly 80 to 100 kilometers, especially when networks must pass through multiple amplifiers or use Raman amplification. In those scenarios, forward error correction becomes necessary to maintain a reliable, error-free link.
He also extended that point to terrestrial and subsea environments. Terrestrial networks connecting data centers, as well as long-haul and subsea infrastructure spanning thousands of kilometers, require OTN because native Ethernet cannot reliably traverse those paths on its own. For PacketLight, OTN is not just a legacy transport layer. It is a practical requirement for deterministic, long-distance, AI-ready connectivity.
Openness and interoperability are becoming standard expectations
Another theme in the interview was the shift toward more open optical networks. Reshef noted that a few years ago, at 100G and 200G, interoperability standards were less mature. With 400G, that changed. Standards such as ZR, ZR+, and Open ROADM helped create a more interoperable ecosystem across DSP vendors and optics suppliers.
On the client side as well, Reshef said adherence to standards allows optics from different vendors to interoperate more reliably. That gives customers stronger second-source options and more freedom to optimize for price, power, or performance. PacketLight’s own approach reflects that philosophy: the company now uses standard pluggable optics only, and avoids non-standard options that could become unsupported after a short time in market.
800G demand is rising, but migration is not frictionless
Reshef said PacketLight is beginning to see real customer demand for coherent 800G, especially among early adopters and AI-focused buyers. He framed the momentum around economics: lower cost per bit, lower power per bit, and higher density. One example he gave was that an 800G module today can cost about 30% less than two 400G modules, making it especially attractive in greenfield builds.
Still, he cautioned that the move to 800G is more complex than prior migrations from 100G to 200G to 400G. Those earlier steps fit more easily into the same spectral spacing, while 800G requires 150 GHz spacing. That means brownfield environments may need Flex Grid support or filter upgrades, making the transition slower than previous generational shifts. Greenfield networks, by contrast, are better positioned to adopt 800G immediately.
Demand is broadening beyond hyperscalers
Although hyperscalers remain a major driver of optical demand, Reshef described a much broader landscape of buyers. He mentioned government and sovereign AI projects, neo-cloud providers, e-commerce companies, and terrestrial networks linking subsea landing points to colocation facilities and data centers such as Equinix and Global Switch.
This matters because it suggests the AI buildout is no longer limited to a handful of cloud giants. PacketLight sees transport growth across multiple segments that need more bandwidth, more secure infrastructure, and more scalable optical connectivity.
Security is moving from compliance issue to strategic requirement
PacketLight has long focused on Layer 1 encryption, and Reshef explained that the original demand came from regulatory compliance. Sectors such as healthcare, financial services, and government needed to protect sensitive data, especially across remote data center links where fiber can be physically tapped. His point was blunt: if someone bends a fiber and extracts even a small percentage of the signal, all transmitted data can be exposed unless it is encrypted.
To address that, PacketLight implemented AES-256 Layer 1 encryption and pursued certifications such as FIPS and Common Criteria. According to Reshef, those certifications are essential because encryption claims need independent validation. PacketLight continues to certify each relevant product, despite the time and cost involved.
Preparing for post-quantum and quantum-safe transport
Reshef also pointed to increasing interest in quantum-safe security. He said customers are increasingly aware of the “harvest now, decrypt later” risk, where attackers record encrypted traffic today in hopes of decrypting it in the future once quantum computing capabilities mature.
PacketLight’s response is to stay interface-ready rather than force a single path. Reshef described support for both QKD and PQC-based approaches, allowing the transport layer to work with different key delivery systems depending on customer preference. He even noted that systems can operate with both approaches in parallel, adding multiple layers of security.
From his perspective, the quantum-safe market is still evolving, but momentum is building as the ecosystem matures, costs improve, and more vendors enter the space.
PacketLight’s position in the market
Taken together, the interview presents PacketLight as a company focused on practical transport challenges rather than broad market hype. Its strategy centers on predictable optical performance for AI workloads, standards-based interoperability, support for long-distance OTN use cases, and certified Layer 1 encryption that can integrate with emerging quantum-safe frameworks.
The message from OFC 2026 was clear: as AI traffic scales, transport networks will need to deliver not just capacity, but determinism, flexibility, and security. PacketLight sees that combination as its opportunity.
FAQ
1. What does PacketLight say is the reason for its long-term relevance?
PacketLight’s CEO says the company has remained relevant by staying close to customers, understanding vertical-specific needs, and moving quickly to deliver products without excessive hierarchy.
2. How does PacketLight describe its role in AI infrastructure?
PacketLight says its role is to transport data quickly, predictably, with low latency and low jitter, which is critical for AI training and inference environments where idle GPUs reduce efficiency.
3. Why are latency and jitter so important for AI networks?
According to the interview, AI workloads are sensitive not just to bandwidth but also to latency and jitter, because compute resources such as GPUs should not remain idle waiting for data.
4. Why does PacketLight emphasize DWDM and OTN for AI transport?
The company argues that DWDM and OTN provide predictable transport behavior, including known latency characteristics, which makes them well suited for high-performance AI connectivity.
5. At what distance does OTN become especially relevant?
PacketLight says OTN becomes especially relevant at distances above roughly 80 to 100 kilometers, particularly when networks use multiple amplifiers or Raman amplification and require strong forward error correction.
6. Why does PacketLight say Ethernet alone is not enough for some long-distance links?
In the interview, PacketLight explains that native Ethernet cannot reliably traverse longer terrestrial and subsea paths on its own, especially where amplified long-haul transport is required.
7. What is PacketLight’s view on open networking and interoperability?
PacketLight says standards such as ZR, ZR+, and Open ROADM have improved interoperability, allowing customers to connect optics and DSPs from different vendors more easily than in earlier generations.
8. Does PacketLight use proprietary optics?
No. PacketLight says it now uses only standard pluggable optics that adhere to industry standards, in order to avoid long-term support risks tied to non-standard modules.
9. Is PacketLight seeing demand for 800G?
Yes. The company says it is starting to see 800G demand from early adopters, especially driven by cost-per-bit, power efficiency, and density advantages.
10. What is slowing 800G adoption in brownfield networks?
PacketLight says 800G requires 150 GHz spacing, which means some existing line systems and filters need upgrades. That makes migration more complex than earlier shifts from 100G to 200G or 400G.
11. Which customer segments does PacketLight see driving growth?
The interview points to a broad mix of demand, including hyperscalers, neo-cloud providers, sovereign AI projects, government environments, e-commerce players, and terrestrial networks linking subsea landing points to data centers.
12. Why does PacketLight stress Layer 1 encryption?
PacketLight says Layer 1 encryption is essential because fiber links between remote locations can be physically tapped, exposing sensitive data unless it is encrypted.
13. What encryption standards and certifications did PacketLight mention?
The company said it supports AES-256 Layer 1 encryption and has pursued certifications including FIPS and Common Criteria.
14. What is the “harvest now, decrypt later” concern mentioned in the interview?
It refers to the risk that attackers could capture encrypted traffic today and decrypt it later when quantum computing becomes capable enough to break current key exchange methods.
15. How is PacketLight preparing for quantum-safe networking?
PacketLight says its systems can interface with both QKD and PQC-based approaches, allowing customers to choose their preferred method and even combine them for added security.
16. What does PacketLight say customers should look for in a quantum-safe transport solution?
According to the interview, customers should look for Layer 1 optical encryption, onboard randomness generation, standard interfaces to third-party key systems, and certified transport equipment that proves encryption is actually implemented correctly.