PostHeaderSection

Symbol

Global NavBar

PostContentSection

Global Footer

Blog Detail Page

Beirit Harvey

At Web Summit Vancouver 2026, leaders came together to discuss what it will take to build and scale the next generation of AI infrastructure. Here are seven of the most important infrastructure shifts shaping the next era of AI. 

The Future of AI Infrastructure: 7 Key Industry Takeaways from Web Summit Vancouver 2026

At Web Summit Vancouver 2026, leaders from across the AI infrastructure sector came together to discuss one of the industry’s biggest questions: what will it take to build and scale the next generation of AI infrastructure?

The panel featured TensorWave CEO Darrick Horton, Rodrigo Liang, Co-founder and CEO of SambaNova Systems, and Ami Badani, CMO of Arm, in conversation with Harold Munro of the Vancouver Sun. 

The biggest takeaway: AI infrastructure is entering a new phase. The challenge is n

At Web Summit Vancouver 2026, leaders from across the AI infrastructure sector came together to discuss one of the industry’s biggest questions: what will it take to build and scale the next generation of AI infrastructure?

The panel featured TensorWave CEO Darrick Horton, Rodrigo Liang, Co-founder and CEO of SambaNova Systems, and Ami Badani, CMO of Arm, in conversation with Harold Munro of the Vancouver Sun. 

The biggest takeaway: AI infrastructure is entering a new phase. The challenge is no longer just access to GPUs. Infrastructure teams are now balancing power availability, inference efficiency, distributed deployment models, open ecosystems, sovereign AI requirements, and real-time workloads simultaneously.

Here are seven of the most important infrastructure shifts shaping the next era of AI.


1. AI Infrastructure Is Becoming a Power Problem

One of the clearest themes throughout the discussion was that scaling AI is increasingly constrained by energy availability.

Liang described how inference demand, especially with agentic AI systems, is driving the need for more efficient infrastructure designed specifically for large-scale inference rather than training alone. The goal is delivering large-model inference that is both fast and power efficient.

Horton added that the growing power shortage is already reshaping how AI infrastructure gets deployed globally. As a result, infrastructure providers are moving toward more geographically distributed deployments rather than relying solely on massive centralized clusters.


2. The Rise of Distributed and Heterogeneous AI Infrastructure

Horton explained that while training workloads still benefit from tightly connected large clusters, inference workloads are changing the architecture of AI infrastructure.

“For inference… it’s much more acceptable to have a diversified portfolio of sites.”

That shift allows inference to run closer to end users while enabling providers to distribute workloads across multiple facilities. TensorWave’s role, Horton explained, is abstracting that complexity away from customers.

The conversation also highlighted the growing move toward heterogeneous compute environments, where different chips and architectures are used for different workloads.

Liang argued that inference is accelerating this shift because many inference workloads no longer depend on CUDA-specific environments. Instead, open frameworks and PyTorch-based ecosystems are enabling a broader mix of architectures across modern AI infrastructure.


3. AI Is Moving Beyond the Cloud

Badani emphasized that AI will not remain centralized in cloud environments forever.

Drawing parallels to the evolution from mainframes to personal computing, she described how AI workloads will increasingly move to edge devices such as smartphones, wearables, and smart glasses.

Badani explained that this shift is being driven not only by performance, but also by privacy, security, regulatory, and cost considerations.

She pointed to Arm’s long-standing focus on power-efficient compute as foundational to enabling AI workloads across both cloud and edge environments.


4. Sovereign AI Is Becoming a Strategic Priority

The discussion also focused heavily on sovereign AI and why governments around the world are investing in localized infrastructure and models.

Liang described sovereign AI as the ability for countries to operate AI systems aligned with their own language, culture, laws, and data governance requirements. He noted that countries including Japan and nations across the Middle East are already training sovereign models tailored to local linguistic and cultural needs.

Badani added that different countries are approaching sovereign AI differently depending on their risk tolerance, capital availability, and long-term economic strategy.

Horton framed sovereign AI from the perspective of governments themselves becoming AI users:

“They’re not going to want to send their data to OpenAI… they need to invest in their own data centers.”

He also emphasized that sovereign AI is not just about hardware and data centers. It requires countries to invest in local AI labs and also sustainable power sources.  


5. Open Ecosystems Will Define the Next Phase of AI

Another major theme throughout the conversation was the importance of openness and competition in the AI ecosystem. Badani argued that long-term innovation depends on avoiding centralized control by only a handful of companies.

Horton echoed that view directly:

“When we founded TensorWave, we did not partner with NVIDIA. We instead partnered with AMD because they are committed to fostering an open ecosystem.”

Rodrigo Liang also challenged one of the industry’s most persistent assumptions: that NVIDIA’s CUDA ecosystem remains an unavoidable barrier across all AI workloads. He argued that inference is changing that dynamic significantly because many modern inference workloads are increasingly built around open frameworks and portable model architectures.

The panel broadly agreed that the future of AI infrastructure will be heterogeneous, multi-vendor, and increasingly optimized around workload-specific architectures rather than one universal stack.


6. CPU vs. GPU: Why AI Infrastructure Is Becoming More Balanced

As AI workloads evolve beyond training into large-scale inference and agentic workflows, the balance between CPUs and GPUs is starting to shift. While GPUs remain critical for training large models and handling accelerated compute tasks, the panelists explained that CPUs are becoming increasingly important for orchestration, agent management, and real-time inference operations.

Horton noted that TensorWave is already seeing increased CPU demand driven by agentic AI workloads, adding that TensorWave’s partnership with AMD positions the company well for this shift given AMD’s strength across both CPUs and GPUs.

Horton explained that TensorWave is now supporting increasingly flexible CPU-to-GPU ratios depending on workload requirements:

“We are seeing an uptick in CPU demand for some of these agentic workflows as well.”

Badani described this moment as a “renaissance” for CPUs, particularly as AI agents become more pervasive and require highly power-efficient compute for orchestration and continuous execution.

The panel agreed that GPUs are not going away, but future AI infrastructure will increasingly rely on a more balanced CPU-to-GPU ratio as workloads become more distributed, real-time, and inference-driven.


7. Real-Time AI Will Drive the Next Wave

As the session wrapped, the panelists looked ahead to the next five to ten years of AI infrastructure. 

Horton predicted that one of AI’s biggest long-term impacts will be accelerating global investment in power infrastructure:

Badani pointed to three major challenges ahead:

 * power infrastructure
 * capital requirements
 * supply chain resilience

Lastly, Liang closed the discussion by emphasizing that AI is rapidly becoming continuous, real-time infrastructure rather than something users interact with occasionally.


TensorWave Cloud: Engineered for AI Velocity

TensorWave is the industry’s premier AMD-exclusive cloud provider, delivering memory optimized AI infrastructure for training, fine-tuning and inference. 

Our platform combines the raw computational power of AMD's latest GPU architectures with innovative liquid cooling technology and ultra-high-speed networking, creating an optimized environment for the most demanding AI workloads. From large language model training to real-time inference, TensorWave provides the performance, scalability, and cost-efficiency that modern AI initiatives demand.

Ready to get started? Connect with a sales engineer.

The Future of AI Infrastructure: 7 Key Industry Takeaways from Web Summit Vancouver 2026

Product

Solutions

Resources

Company

Product

Resources

Solutions

Company