VC Comment: Amid The AI Frenzy, It’s The Startups That Will Shape The Future

Sean Kohli, General Partner at a San Francisco early-stage venture fund and Chair of the Young  Entrepreneurs Forum, argues that in a market flooded with capital, the real test is separating hype from technologies that will last. He highlights five AI startups in Kohli Ventures’ portfolio already proving they can meet that test. 

 

Artificial intelligence is attracting investment on a scale few industries have ever seen.  America’s largest technology firms are expected to pour close to $400 billion into AI  infrastructure this year alone. By the end of the decade, global spending on data centres could climb above $3 trillion.  

Meanwhile, the stock market is being reshaped before our eyes in real time. Only last week,  Oracle’s share price leapt more than forty per cent in a single day, briefly making its co founder the world’s richest man, on the back of a deal to supply vast computing power to  OpenAI. 

The frenzy recalls the dot-com boom, when venture investment ballooned from $8 billion  to more than $100 billion in just five years, much of it lost when the bubble burst. Capital  flooded into every internet idea, but few could tell which would endure. Most disappeared  

when the bubble burst, yet out of the noise emerged those that defined the digital age.  Then, as now, the challenge is not spotting where the money is going but identifying which  technologies will create lasting value. 

Hype attracts capital, but only real breakthroughs create value. Through Kohli Ventures’  €50 million investment programme, we’ve backed five of the highest-signal startups from  Silicon Valley. Each of them is tackling a critical weakness in the AI economy, from shortages  of talent and tools to gaps in trust and inefficiencies that slow progress. In a crowded field,  these are the companies enabling others to build, scale and innovate with confidence. 

 

Voltai 

 

 

Semiconductors are the beating heart of modern technology, powering everything from  smartphones to self-driving cars. Their rapid advances, captured by Moore’s Law, have  driven progress, and the market is now expected to grow by more than ten per cent a year. 

Despite the boom, or perhaps because of it, the industry faces a severe shortage of skilled  designers and engineers of more than one million by the end of the decade. Mistakes in chip  design are hugely expensive, and the complexity is only increasing. 

Voltai is tackling that destructive bottleneck head-on by building artificial intelligence models  trained specifically for electronics design. What this means in practice is tools that let chip 

designers work faster, cut out expensive mistakes, and understand technical documents and  IP that often take weeks to analyse. 

Instead of wasting weeks chasing down manufacturing constraints or testing for compatibility, Voltai gives engineers early warnings and rapid validation, allowing them to  focus their energy on design improvements and fresh ideas. 

For an industry under pressure to produce more and better silicon, Voltai is an absolute necessity. The company is already winning the confidence of senior executives in chip firms  who see it as a force-multiplier and its tools could be the difference between progress at  scale and an industry stuck in bottlenecks. If it succeeds, it could accelerate the pace of  innovation and shape the next decade of technological growth. 

 

 

Judgment Labs 

 

 

AI agents are one of the most exciting frontiers in technology. They can plan, reason and  use digital tools to complete complex tasks, from booking travel to writing code. The  trouble is that they are also highly unreliable. Too often they stall, repeat mistakes or  generate results no one can trust. 

That fragility matters because the potential prize is so large. Analysts predict the market for  AI agents could reach $47 billion by 2030, but only if they can progress from eye-catching  demonstrations to systems people can rely on every day. Reliability, not capability, will  decide whether they can become mainstream in the near future. 

Judgment Labs is closing that gap by building the infrastructure agents need if they are ever  to be trusted at scale. Its platform lets developers run rigorous evaluations, monitor how  agents make decisions, and catch errors before they spiral. In effect, they are creating the  

feedback loops that every new technology requires. It’s the quality assurance for a  generation of AI systems that would otherwise remain stuck in the lab. 

The team brings research experience from Stanford, Berkeley and other leading labs, but  the mission is refreshingly straightforward: take AI agents from promising experiments to  trusted tools. Without companies like Judgment Labs, agents risk remaining a curiosity.  With them, they have a genuine chance to become part of everyday life. 

 

Vibecode 

 

 

If TikTok turned an army of young people into short-film makers, Vibecode’s goal is to make  everyone app builders by making the process as easy as posting a story or a clip. Its iPhone  app lets you describe what you want and within minutes you have a working version on  your phone. 

The appeal is clear. Building an app has always been a slow and technical process left to  those with specialist skills. Vibecode turns that process into a simple dialogue where you 

describe the idea, test the result, refine it, and share. Already more than 40,000 apps have  been created this way, with $9.4 million in seed funding showing the momentum behind it. 

Just as social media lowered the barriers to publishing, Vibecode lowers the barriers to  building. If it works, the next wave of startups may not come from universities or  incubators, but from ordinary people with an idea and a phone. 

 

Godela 

 

 

Testing new ideas in engineering is painfully slow. Simulations can take days or even weeks,  and physical prototypes take longer still. Those delays hold back innovation when speed is  critical. 

The global simulation software market is expected to more than double by 2030 and these  delays, that act as a painful drag on solving urgent problems, could get worse. Godela’s answer is an AI physics engine. You ask it a question, upload your design files or data, and it  produces results in seconds that would otherwise take weeks.  

Instead of waiting weeks for a single result, researchers can cycle through countless designs,  probe edge-cases in real time, and refine breakthroughs on a continuous loop. The founders  have lived this problem themselves, having worked at Apple, Google, Intel, Stanford and  Harvard. They know how much time is lost waiting for simulations or waiting for prototypes  to be built. 

Their ambition is to do for physical systems what large language models did for text: turn  delays into instant answers and enable access to insights once reserved for specialists. 

 

Archer AI 

 

 

In the same way that pilots rely on flight simulators before flying a plane, AI agents need  environments where they can safely test ideas, fail, and improve. This is the principle behind  reinforcement learning, which has powered breakthroughs from DeepMind’s AlphaGo to  robotics. 

Archer AI is one of the first companies dedicated to building these training environments at  scale, supplying them to labs such as Anthropic and OpenAI. The stakes are clear.  Autonomous vehicle firms now log billions of miles in simulation each year because relying  only on real-world testing would take decades and expose people to unnecessary risk. AI  agents need the same kind of simulated worlds if they are to move from theory to practice. 

By creating realistic, adaptable worlds for AI to learn in, Archer is helping to turn  reinforcement learning from a research curiosity into a practical tool. It’s still a young field,  but one that could prove as essential to AI’s future as data centres.

 

What will outlive the boom? 

 

What unites these five firms is that they give others the power to build. It’s that potential, not hype or inflated valuations, is what shapes the future. 

That is the real story of AI, not abstract strategies or regulatory frameworks, but those bringing down costs, compressing time, and turning research into tools people can actually  use. They are the ones who will define the next era. 

The challenge is not whether these companies will have impact. They already are. The  question is whether others will support and scale them in time or look back wondering why  they let the opportunity slip away.