Views: 56
From where I sit as an early-stage VC, the honest answer is: probably a bit of both.
There is clearly excess in parts of the market. Some rounds are surreal. Some narratives are getting funded before the products are ready. Latest example is Recursive Intelligence, a 4 month old London-based frontier-lab that raised $500M at a $4 billion valuation in an oversubscribed round that might grow to $1 billion. And yes, there is no shortage of AI companies that look more like thin wrappers than enduring businesses.
But underneath the noise, something much bigger is happening.
The best visual I’ve seen for that is the chart on page 8 of ARK Invest’s Big Ideas 2026 report. It maps the great capital expenditure waves of modern history, from railroads and telephony to electrification, cars, semiconductors, ecommerce and software, and then shows the next mountains now forming: AI software, terrestrial AI data centers, space and robotaxis. In other words, AI is no longer just a software theme. It is starting to look like a broad investment cycle that spills into compute, energy, logistics and the physical world.
That macro picture is now showing up in venture funding in a way that is hard to ignore. Crunchbase found that Q1 2026 was the biggest quarter for venture funding on record, with $300 billion invested globally into roughly 6,000 startups. AI alone captured $242 billion, or 80% of total funding for the quarter. Four companies, OpenAI, Anthropic, xAI and Waymo, raised a combined $188 billion, equal to 65% of all global venture investment in the quarter.
So yes, the boom is real.
But the more interesting question for founders is not whether AI is big. Of course it is. The real question is what kind of market this creates for a startup entering now.
Because this is not a normal cycle.
It is a super-cycle with very sharp filters.
The biggest challenges for founders entering AI today
The first challenge is obvious: do not compete with the frontier labs where they are strongest.
If your strategy depends on outspending OpenAI, Anthropic or xAI on training, compute or general-purpose model capability, you are probably dead on arrival. The giants are spending sums that are simply not available to 99.9% of startups, and they are doing so in a market where scale itself compounds performance and distribution.
Part of the challenge is that frontier labs are moving from being pure infrastructure to the application layer themselves, by launching products. I gave a few recent examples by Anthropic, including Claude Design, in my recent post.
The second challenge is that product half-life is collapsing.
When inference gets dramatically cheaper, models get better, and agent performance improves this quickly, what felt magical last quarter can start looking like a feature by the next one. ARK notes that AI agents’ reliable long-duration task completion increased 5x during 2025, from 6 minutes to 31 minutes. That is great news for the ecosystem, but it is brutal for shallow products.
This is why I worry about knowledge and product homogenisation. If your company is just packaging the output of a frontier model in a prettier UI, the moat will be thin and the shelf life may be shorter than you think.
The third challenge is one many technical founders still underestimate: enterprise adoption is still a human problem.
Yes, the tools are improving fast. But procurement cycles, security concerns, internal politics, workflow changes and trust still move more slowly than model progress. Founders often assume that once the technology works, adoption will follow automatically. It rarely does. Especially in enterprise software, the bottleneck is often not intelligence. It is implementation, risk tolerance and behavior change.
Where early-stage founders can still win
The good news is that the opportunity is enormous for exactly the same reasons the market is hard.
The first opportunity is the most important one: the application layer is still wide open.
I do not believe the main lesson of this market is “raise more.” I think the lesson is “be more precise.” Let the frontier labs spend the billions on model training and infrastructure. Your job is to build where domain context, workflow ownership, proprietary data, compliance, trust or distribution create defensibility.
That could mean vertical software. The deeper the workflow integration and data moat, the better.
That could mean AI-native workflow tools.
That could mean compliance, orchestration, QA, procurement, healthcare, legal, finance, customer support, internal tools or industrial operations.
But it probably does not mean “another generic copilot.”
The second opportunity is the collapsing cost of intelligence.
When model usage gets cheaper that fast, early-stage founders can experiment more aggressively, ship faster, and serve use cases that would previously have required services-heavy businesses. This lowers the cost of starting a company and expands the range of products that can reach viable gross margins.
When infrastructure spending explodes and the cost of intelligence collapses at the same time, entirely new product categories become economically viable. Things that were too expensive, too brittle or too labor-intensive even 12 months ago suddenly become buildable by small teams.
That is usually when the most interesting startups are born.
The third opportunity is the move from chat to agents.
It’s impossible to open any social media platform without hearing about OpenClaw, Claude Code, OpenAI’s codex etc autonomously coding or creating entire products from scratch.
With it also comes a whole new slew of challenges, but the web is basically being re-invented FOR (and in some case BY) AI Agents.
So what should founders do?
If I were building an AI startup today, I would try to win on one of five things:
- Distribution
- Proprietary data
- Workflow depth
- Trust
- Speed of execution
More specifically, I would look for a problem that is painful, repetitive, valuable and still underserved by software.
I would use the abundance of intelligence created by the labs, but I would build the company where context matters, where integration matters, and where the customer does not just want answers but outcomes.
So are we in a bubble, or in a super-cycle?
My answer is still both.
There is froth. There is hype. There will absolutely be casualties.
But the broader arc looks real. The capex is real. The infrastructure is real. The cost declines are real. The adoption curves are real. And for founders, that creates one of the most exciting moments in decades to build a meaningful company.
The catch is that this is not an easy market. It only looks easy from a distance.
For founders entering AI today, the playbook is not to build the broadest story. It is to build the sharpest wedge.
He is a former General Partner at Google Ventures (GV) in Europe, former head of Google for Entrepreneurs in Europe, and founding head of Campus London, Google’s first startup hub. Eze writes on Israeli tech, venture capital, artificial intelligence, and founder strategy.
He is also the founder of Techbikers, a nonprofit that brings together the startup ecosystem on cycling challenges in support of Room to Read.