The following interview features Vladislav Zaimov, a seasoned telecommunications expert with deep roots in enterprise infrastructure and risk management for vulnerable networks. With over two decades of experience monitoring the cyclical nature of the industry, Zaimov offers a unique perspective on the current “AI gold rush” and its profound impact on the physical layers of our global connectivity.
The discussion explores the massive shift in capital expenditure as operators pivot from the 5G era to AI-driven core upgrades. Zaimov breaks down the financial and operational divides between established hyperscalers and emerging “neo-cloud” providers, the strategic importance of submarine cable ownership, and why the “last mile” remains the ultimate battleground for enterprise AI. He also touches upon the complexities of global supply chains and the geopolitical necessity of international cooperation in semiconductor and optical innovation.
Many telecom operators are pivotally shifting focus after a period of limited investment in their optical cores following the 5G rollout. How are these providers currently rebalancing their budgets to meet AI demands, and what specific technical milestones must they reach to achieve the necessary low-latency connectivity?
For the last five years, most global service providers have been heavily distracted by 5G, which many now view as a bit of a financial disappointment or a “bust.” This has led to a significant period of underinvestment in the actual optical core—the backbone that carries all that wireless traffic. Now, they are realizing that to support AI, they must pivot their budgets back to high-speed optical infrastructure because AI training and inference demand a level of performance 5G simply wasn’t designed to handle alone. To achieve the necessary low-latency connectivity, providers are racing to implement 800G and even 1.6T optical waves to move massive datasets between data centers. They are moving away from traditional routed architectures toward more direct, “bent-pipe” optical paths that reduce hop counts and shave off milliseconds of delay. It is a step-by-step process of stripping out legacy gear and replacing it with high-capacity pluggables and line systems that can “bend physics” to push more bits over existing fiber.
Large-scale cloud providers and newer “neo-cloud” players operate under very different financial structures and capitalization models. What specific operational risks do these emerging players face during an infrastructure shakeout, and how can they realistically monetize their niche language models to survive?
The market is currently split into two very different camps: the well-capitalized hyperscalers and the “neo-scalers” or emerging players. The hyperscalers are essentially their own best customers; they use AI for their own profitable applications, which gives them a built-in monetization engine. The emerging players, however, are often operating on debt-based models and special purpose vehicles, which makes them incredibly vulnerable if the market cooling begins. These neo-cloud players face the “monetization wall”—they have built expensive facilities, but if they cannot find a specific, paying niche for their language models, their funding will dry up. We are likely to see a significant fallout where only the models with clear enterprise utility survive, while those without a distinct path to revenue will be absorbed or shuttered during the inevitable infrastructure shakeout.
Significant portions of the world’s submarine cable capacity have shifted from carrier consortiums to being owned by a few major tech giants. How does this concentration of ownership change the way global data traffic is managed, and what are the long-term implications for international network sovereignty?
We have witnessed a historic shift where hyperscalers now own more submarine capacity than anyone else in the world. Historically, these cables were built by consortiums of traditional carriers, but the sheer scale of data movement required by big tech has forced them to become their own infrastructure providers. This concentration means that global data traffic is increasingly managed by private entities rather than public utilities, which can prioritize their own “east-west” data center traffic over general consumer internet. Long-term, this creates a complex tension regarding network sovereignty, as countries realize their primary data arteries are owned by a handful of foreign corporations. The trade-off is that these tech giants have the capital to innovate at a velocity that traditional carriers simply cannot match, effectively subsidizing the advancement of global connectivity for everyone else.
While massive data centers handle model training, the “last mile” of connectivity remains a primary entry point for enterprise and industrial AI applications. What unique advantages do traditional service providers hold at the network edge, and what specific infrastructure upgrades are required to support real-time inference?
The edge is where AI actually makes money, moving from the theoretical training phase into real-world industrial and enterprise applications. As powerful as the hyperscalers are, they lack the physical presence to be everywhere; they cannot reach into every enterprise office, every factory, or every private home. This is the traditional service provider’s “home turf” advantage—they own the “last mile” of fiber that connects the end-user to the cloud. To support real-time inference, these carriers must upgrade their edge nodes with high-speed pluggable optics and localized compute capacity to ensure that data doesn’t have to travel back to a central hub in another state or country. This transition requires a symbiotic relationship where the cutting-edge optical technology developed for hyperscalers is filtered down into the more localized carrier networks to ensure seamless, low-latency performance.
Modern optical networking relies on a global ecosystem of specialized components and advanced semiconductors sourced from diverse international markets. How do current geopolitical shifts influence the speed of innovation for high-speed connectivity, and what practical steps should companies take to ensure supply chain resilience?
Innovation in this space is inherently a “team game” because no single country owns the entire stack. For example, even a leading US-based firm must rely on the cutting-edge semiconductor fabrication of TSMC to produce the chips that power high-speed optical waves. Geopolitical shifts and new regulations around sovereignty mean that companies can no longer rely on a single source or a single region for their critical components. To ensure resilience, firms are having to diversify their ecosystems and collaborate more closely with international partners on standardized components. The practical step for any infrastructure company is to embrace a “best-of-breed” strategy while maintaining a global footprint, ensuring that their supply chain isn’t tethered to a single point of failure in a fluctuating political landscape.
What is your forecast for global network infrastructure?
My forecast is that we are entering a decade-long “global rebuild” of the entire network fabric. We will see a massive convergence where the technology used in Formula One-style hyperscale data centers becomes the standard for everyday carrier networks, making high-speed, low-latency connectivity ubiquitous. While there will certainly be a “shakeout” of smaller, speculative AI players, the underlying demand for bits is permanent and growing. Expect to see a move toward “sovereign AI” clouds in different regions, where nations invest in their own local infrastructure to ensure their data stays within their borders while still benefiting from global connectivity. Ultimately, the network is no longer just a utility—it is the very oxygen that the AI economy breathes.
