91̽»¨

Sean Kohli, General Partner at a San Francisco early-stage venture fund and Chair of the YoungÌý 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.Ìý

Ìý

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

Meanwhile, the stock market is being reshaped before our eyes in real time. Only last week,Ìý 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 toÌý OpenAI.Ìý

The frenzy recalls the dot-com boom, when venture investment ballooned from $8 billionÌý to more than $100 billion in just five years, much of it lost when the bubble burst. CapitalÌý 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.Ìý Then, as now, the challenge is not spotting where the money is going but identifying whichÌý technologies will create lasting value.Ìý

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

Ìý

VoltaiÌý

Ìý

Ìý

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

Despite the boom, or perhaps because of it, the industry faces a severe shortage of skilledÌý designers and engineers of by the end of the decade. Mistakes in chipÌý design are hugely expensive, and the complexity is only increasing.Ìý

Voltai is tackling that destructive bottleneck head-on by building artificial intelligence modelsÌý trained specifically for electronics design. What this means in practice is tools that let chipÌýdesigners work faster, cut out expensive mistakes, and understand technical documents andÌý IP that often take weeks to analyse.Ìý

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

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 firmsÌý 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 ofÌý innovation and shape the next decade of technological growth.Ìý

Ìý

Ìý

Judgment LabsÌý

Ìý

Ìý

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

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

Judgment Labs is closing that gap by building the infrastructure agents need if they are everÌý to be trusted at scale. Its platform lets developers run rigorous evaluations, monitor howÌý 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 aÌý generation of AI systems that would otherwise remain stuck in the lab.Ìý

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

Ìý

VibecodeÌý

Ìý

Ìý

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

The appeal is clear. Building an app has always been a slow and technical process left toÌý those with specialist skills. Vibecode turns that process into a simple dialogue where youÌýdescribe the idea, test the result, refine it, and share. Already more than 40,000 apps haveÌý been created this way, with $9.4 million in seed funding showing the momentum behind it.Ìý

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 orÌý incubators, but from ordinary people with an idea and a phone.Ìý

Ìý

GodelaÌý

Ìý

Ìý

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

The global simulation software market is expected to and theseÌý 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 itÌý produces results in seconds that would otherwise take weeks.ÌýÌý

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

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

Ìý

Archer AIÌý

Ìý

Ìý

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

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

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

Ìý

What will outlive the boom?Ìý

Ìý

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.Ìý

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.Ìý

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