Why network issues are holding back enterprise deployments [Q&A]

While AI promises lightning-fast transformation, many organizations are running into problems when outdated, complex networks bog down deployments, forcing costly redesigns and leaving businesses waiting months instead of weeks to implement AI at scale.

We spoke to Amir Khan, CEO at Alkira, to learn why fixing the network has become foundational to unlocking real AI value.

BN: Why is the AI boom exposing hidden weaknesses in enterprise networks?

AK: AI workloads place unprecedented demands on network infrastructure that many enterprises have not fully anticipated or addressed. AI applications require immense data throughput, low latency, and seamless connectivity across hybrid, multi-cloud, and edge environments to function effectively. Traditional enterprise networks, often built for legacy traffic patterns and localized environments, struggle to keep up with these requirements, leading to performance bottlenecks and scalability challenges.

Moreover, the complexity and distribution of modern IT environments highlight gaps in network visibility, control, and security. AI-driven applications depend on real-time data transfer and dynamic resource allocation, which strains existing network capacity and capacity planning methods. At the same time, the evolving threat landscape means networks must secure AI workloads against sophisticated cyberattacks that exploit these new data flows and computational demands.

In essence, the rapid adoption of AI stresses network infrastructures not designed for such scale, speed, and security needs. This mismatch forces enterprises to re-evaluate and rethink their network strategies to ensure agility, resilience, and security in the face of AI-driven transformation. AI acts as both an accelerator and a stress test, revealing hidden weaknesses in enterprise networks that require modernization to fully leverage AI's potential.

BN: In what ways is traditional networking slowing AI innovation across industries?

AK: This is primarily because traditional enterprise networks are not built to handle the massive scale, speed, and complexity that modern AI workloads demand. What should be a rapid, transformative deployment is instead hindered by network limitations that cause costly delays and prolonged timelines. Enterprises often face months-long network redesigns and infrastructure upgrades just to support AI’s intense data processing and real-time requirements, significantly slowing down the adoption and scaling of AI-driven innovations.

Because AI workloads involve transferring huge datasets across hybrid environments and require ultra-low latency for real-time processing, networks that are slow or fragmented create bottlenecks that reduce AI performance and efficiency. The result is a frustrating experience where businesses struggle to realize the speed and agility promised by AI, hampering their ability to quickly develop new products, streamline operations, or improve customer experiences.

This slowdown in network readiness leads to a state where AI’s potential is seemingly within reach but hindered by underlying infrastructure gaps. Consequently, innovation is slowed across industries as AI-driven initiatives are delayed, limiting competitive advantage and digital transformation possibilities.

BN: What unique challenges do large enterprises face when running AI across complex, hybrid, and global infrastructures?

AK: One major issue large enterprises face is infrastructure strain. AI workloads, particularly sophisticated models like large language models, require extremely high-speed data movement and low latency between data centers, clouds, and edge locations. Traditional networks and data architectures often cannot keep pace with these demands, creating bottlenecks in memory bandwidth, compute access, and data transfer that limit AI performance and scalability.

Data fragmentation is another key obstacle. Enterprises typically have data spread across numerous siloed systems, whether it’s on-premises, in multiple clouds, or in edge environments, making it difficult to unify and govern data effectively for AI training and inference. This fragmentation leads to inefficiencies, duplicate data storage, compliance challenges, and inconsistent model inputs causing drift and reduced accuracy.

Security and governance add further complexity. The expanded attack surface created by distributed AI workloads increases vulnerability to cyber threats, demanding robust, AI-specific security frameworks and fine-grained access controls. Meanwhile, the shortage of specialized skills to manage complex AI infrastructure, integrate diverse systems, and ensure data quality and compliance poses a significant operational challenge.

Finally, the scale and geographic distribution of AI environments introduce challenges in capacity planning and cost management. Data centers face physical limits on power, cooling, and space, while global operations require resilient, agile networks that minimize latency and support compliance with regional regulations.

BN: How can organizations enable AI adoption without costly, time-consuming network overhauls?

AK: Organizations can take deliberate, strategic steps to modernize their network environments gradually and efficiently. Rather than replacing entire network infrastructures, enterprises benefit from assessing existing assets to identify bottlenecks and critical points that impact AI workloads, focusing investments where they will have the most immediate and measurable effect.

Investing in flexible, scalable connectivity options helps support dynamic AI workloads without disrupting core operations. These technologies facilitate low-latency, high-throughput data flows essential for AI training and real-time inference while reducing the need for full network rebuilds.

Additionally, adopting automation and AI-powered network management tools enhances network visibility, optimizes traffic routing, and simplifies troubleshooting, enabling faster responsiveness to changing AI demands without extensive manual intervention. Strengthening security with zero-trust frameworks and continuous monitoring ensures AI workloads and sensitive data are protected without adding complexity.

Modernizing incrementally through pilot projects and prioritizing hybrid, multi-cloud, and edge deployments allows organizations to evolve their networks alongside AI adoption at a manageable pace. This measured approach minimizes risk and cost while laying the foundation for scalable, secure AI operations capable of driving business innovation.

BN: What lessons has have you learned from helping Fortune 500 companies deploy AI initiatives at scale?

AK: Alkira has learned several key lessons. One critical insight is that speed and agility are paramount. Enterprises benefit greatly when they can rapidly connect and scale their global, hybrid cloud environments without the traditional months or years of network redesign and deployment. Simplifying complex network operations and providing centralized visibility empowers IT teams to respond quickly to changing business needs and avoid bottlenecks that slow AI projects.

Another lesson is the importance of security and control at scale. The largest organizations must be able to apply consistent segmentation and policy enforcement across diverse environments, from on-premises to multi-cloud, to protect sensitive AI data and workloads while enabling seamless connectivity. This granularity in control helps maintain compliance and mitigate risk even as AI adoption accelerates.
Fortune 500 enterprises also value operational efficiency gains. Transitioning from manually intensive network management to fully codified, automated infrastructure management reduces the need for large engineering teams and minimizes human error. Faster provisioning times and the ability to easily onboard partners and cloud resources translate directly into faster innovation and improved business outcomes.

These lessons highlight the critical need for a modern network foundation that enables frictionless AI adoption and sustainable innovation.

Image credit: akarapongphoto/depositphotos.com