---
title: AI Dominates Technology Live! 2026 Conference
url: https://www.elseif.net/ai-dominates-technology-live-2026-conference
published: 2026-11-22T20:04:42+00:00
language: en
section: Agents
source: https://crn.pl/artykuly/technology-live-2026-wszystkie-drogi-prowadza-do-ai/
organizations: WEKA, Nvidia, OpenRouter, 451 Research, SemiAnalysis, OpenAI
publisher: elseif
---

# AI Dominates Technology Live! 2026 Conference

Artificial intelligence themes dominated the Technology Live! conference held in London, where providers showcased tools designed to maximize the utility of AI models.

WEKA, a provider of storage memory designed for AI models, highlighted inefficiencies in how graphics processing units are used. Gareth Whiting, the vice president of sales for the EMEA region at WEKA, stated that while control panels may show 93 percent utilization of graphics processing units, much of that work is wasted. He explained that graphics processing units in inference clusters frequently recalculate data they have already processed because AI agent context grows faster than high bandwidth memory.

Whiting distinguished between AI model training and inference. Training is a process with a defined start, end, and budget that is write-heavy and dependent on throughput. Inference is a continuous process where every query generates a real cost. Inference is primarily read-heavy and is determined by latency and the number of simultaneous users. Whiting argued that slow inference costs revenue and customers, meaning useful work per watt, rack, and dollar must be measured rather than just chip intensity. He noted that inference is currently limited by memory rather than computing power.

The rise of AI agents has increased data demands compared to simple chatbots. AI agents perform tasks, use external tools, and read the entire history of their work in each turn. Data from SemiAnalysis on 250,000 coding agent sessions showed a typical session consists of 6 turns and 90,000 tokens, while the longest 10 percent reach 300 turns and 400,000 tokens. OpenRouter reported that weekly token volume grew 50-fold from approximately 4 trillion a year ago to 210 trillion in early November. 451 Research estimates that agents will generate 72 percent of tokens in North America by 2028, compared to 13 percent this year.

Infrastructure spending is shifting toward inference. 451 Research reports that inference accounts for 225 billion dollars of AI infrastructure spending this year, which is about one third of the total. This amount is expected to increase approximately threefold to 746 billion dollars by 2030. Current AI growth is hindered by power, memory, and rack space. Data center energy consumption in the United States was about 4.5 percent in 2023 and is projected to reach 7 to 12 percent in the coming years.

WEKA introduced NeuralMesh to address storage limitations. NeuralMesh is a software-based, distributed storage platform for AI and high performance computing environments. It combines resources from multiple servers and NVMe media into one parallel file system. The architecture eliminates the need for controllers or separate metadata servers, as every server in the cluster handles data, metadata, and input-output operations simultaneously. The software organizes, shares, and protects data in a distributed infrastructure to avoid hot spots by automatically splitting large files into smaller fragments.
