AI’s second act: How enterprise infrastructure and synthetic labor are reshaping the future

As generative AI captures headlines and boardroom attention alike, it’s tempting to think of the movement as a sudden surge sparked by the overnight success of ChatGPT. But as Managing Director George Mathew made clear at the Web Summit Vancouver panel — which also included Cathy Gao, JP Sanday, and moderator Sam Bourgi —  the reality is far more strategic and far more enduring.

Over the past five years, the AI ecosystem has been laying the groundwork for what’s next. While consumer use cases like chatbots and image generators have attracted mass awareness, a quieter revolution has been happening under the surface: Enterprise infrastructure has matured, enabling a new generation of AI applications that are no longer experimental — they’re essential.

More than that, Mathew argues that the market opportunity goes far beyond traditional software. Mathew believes we’re entering an era where AI doesn’t just automate tasks — it could redefine entire labor markets.

In this new environment, picking the right companies, platforms, and teams early isn’t just important — it’s essential.

Watch the full panel below:

A five-year build cycle is bearing fruit

While much of the world’s attention has focused on flashy generative tools, Mathew’s view is that the last half-decade of AI has been defined by something less visible but more important: infrastructure.

This is the “plumbing” layer that makes scalable, secure, and sophisticated AI possible in an enterprise context. Tools like Weights & Biases, an early Insight investment, became vital to model development workflows. “Anyone building foundation models needed ways to track experiments, tune parameters, and manage versioning. Weights & Biases became the de facto toolset,” Mathew said. It was later acquired by CoreWeave for approximately $1.4B.

That moment, he explained, marked a larger transition. “We’re starting to now see the platform sediment at this point,” Mathew noted. With enterprise-ready infrastructure in place, companies can finally build applications at scale, without reinventing the wheel each time.

From code to coworkers: Enterprise AI expands the labor frontier

What comes next, Mathew believes, is not just about better software; it’s about redefining labor itself.

He points to Relevance AI as a glimpse into this future. The company builds synthetic digital workers — AI agents that can handle marketing, sales development, and front-office operations. “You can now have a synthetic version of yourself on Relevance,” Mathew said. “It’s a platform where digital coworkers work alongside human ones.”

That may sound futuristic, but it’s already happening. Relevance is seeing strong enterprise adoption from some of the world’s largest companies. What’s driving that? The ability to reduce or replace outsourced services — not just internal headcount — with AI agents that are always available, highly scalable, and cost-effective.

Critically, this expands the market beyond traditional SaaS. “In AI, you’re not only consuming the software market,” Mathew said. “You’re potentially consuming the labor pool, the services layer, everything around it.” This reframes total addressable market (TAM) not in terms of license seats, but in terms of work itself.

Stability is the constraint, and efficiency fuels demand

Despite this momentum, scaling AI infrastructure remains fraught with risk, especially geopolitical risk. As trade tensions over semiconductors and GPUs escalate, Mathew underscored a growing concern: uncertainty.

“We’re seeing public companies abandon second-half forecasts,” he said. “There’s very little predictability.” What the industry needs most, in his view, is stability. Infrastructure builders like CoreWeave and cloud providers are making massive capital commitments. But if policy volatility undermines their ability to plan, the whole ecosystem could suffer.

Yet, even with constraints, demand keeps rising. The emergence of more efficient models, like DeepSeek, hasn’t reduced pressure. In fact, it’s made it worse. “All of that additional capacity was immediately soaked up,” Mathew said.

The lesson? Efficiency isn’t a release valve — it can also act as an accelerant. Gains in inference speed or model size may open the door to entirely new workloads. In Mathew’s words: “This is only going up and to the right.”

This isn’t just the next platform shift

AI isn’t just another “mobile moment” or cloud transformation. According to Mathew, it’s a reconfiguration of how businesses operate, how value is delivered, and how work gets done.

From infrastructure to inference, from pricing models to hiring strategies, the enterprise is being reshaped at every level. And while the current cycle may feel chaotic, its long-term potential is profound.

“This is the most exciting time to be in venture,” Mathew opined. “But it’s also the hardest. You’ve got to pick the right team — and you’ve got to get there early.”

*Note: Insight Partners has invested in Relevance AI and Weights & Biases. 

Video credit to Web Summit.