Views: 16
There is a strange contradiction in venture right now.
On one hand, software has never been easier to build. A small team can now ship products that would have required a full engineering org only a few years ago. Claude Code, Cursor, GitHub Copilot, Codex, Replit, Devin and the growing family of AI coding agents are compressing the cost of creation. Anthropic describes Claude Code as an agentic coding system that can read a codebase, make changes across files, run tests and deliver committed code. It is explicitly “agentic, not autocomplete.”
On the other hand, software has rarely been harder to fund. That is the paradox of 2026. The cost of building software is collapsing, but the bar for building a fundable software company is rising.
For founders, this is confusing. For investors, it is uncomfortable. For the venture market, it is creating one of the sharpest dichotomies we have seen in years.
You are either on a rocketship, with AI-fuelled growth, expanding usage, strong revenue velocity and the perception that you might become category-defining. In a way, Lovable made it difficult for everyone else… Or you are expected to bootstrap, stay lean, serve customers, and prove that your company can survive without venture capital. The middle is getting squeezed.
AI Coding is a real concern for software moats
It is tempting to dismiss AI coding tools as developer productivity software. That is too narrow.
Claude Code, OpenAI’s Codex and its peers are not just helping engineers write code faster. They are changing what counts as defensibility in software.
Historically, software companies could build meaningful advantage through speed of execution, proprietary workflows, accumulated product complexity, integrations, UX polish and engineering depth. Those things still matter, but AI coding agents are eroding the time it takes to reproduce many of them.
If a competitor can build a working version of your product in weeks instead of months, the moat cannot simply be “we built the software.” If a customer can stitch together an internal tool with AI agents, the moat cannot simply be “we automate a workflow.” If an incumbent can point agents at its existing data, distribution and customer base, the moat cannot simply be “we use AI.”
This is where Chamath Palihapitiya’s point about the “collapse of terminal value” becomes relevant. His argument, in short, is that if AI increases the annual probability that a business can be disrupted or rendered obsolete, the market should discount the long-term cash flows of that business much more aggressively. In his example, even a 20% annual probability of AI-driven disruption can dramatically reduce terminal value.
Venture capital, in its current form, effectively ceases to function. Who funds a pre-revenue company at a $1 billion valuation if there is no terminal value to grow into? The IPO market, built on stories about what a company will become, collapses into a smaller arena for businesses that already generate serious cash. The entire venture capital/growth equity complex – one of the defining financial innovations of the last forty years – would instantly become a historical artifact.
The Collapse Of Terminal Value – What Happens If AI Makes Every Moat Temporary?
That matters because venture is, ultimately, a terminal value game. We fund losses today because we believe the company can compound into a much larger, more durable business later. If durability is less certain, investors need more evidence earlier.
The question used to be: can this team build it? Increasingly, the question is: once built, why will it matter?
SVB’s data shows the split clearly
The H1 2026 SVB State of the Markets report captures the contradiction well. In 2025, nearly $340 billion flowed into US VC-backed companies, the second-highest year on record. But that capital was not broadly distributed. The top 1% of companies accounted for roughly one-third of all VC investment, while the bottom half of US VC-backed companies by valuation accounted for just 7% of investment.
SVB also notes that VC investment was up 53% from 2024 to 2025, but deal count was down 15%. In other words, more capital went into fewer companies. That explains why the market can feel simultaneously hot and frozen. The AI winners are raising enormous rounds. Everyone else is fighting for attention.
The same report says deal count under $100 million is at a decade low, which is probably the better proxy for what most founders actually experience. Mega-rounds make the headlines, but sub-$100 million rounds define the day-to-day fundraising environment for most startups.
And the revenue bar has moved. The attached SVB/Carta chart is blunt: in 2025, the median ARR at time of fundraise is not only higher than before (goal post moved up) but also there’s a huge range between the median and top quartile:
Series A
– Median: $2.8M
– Top quartile: $6.9M
Series B
– Median: $8.4M
– Top quartile: $14.9M
Series C
– Median: $16.1M
– Top quartile: $45.0M
That is a very different market from 2021.
SVB’s broader benchmark section makes the same point: revenue required to raise is higher today, but growth is slower. For companies moving from seed to Series A, the challenge is especially acute, with SVB noting that companies often need to grow revenue 8x to 12x between seed and Series A. Historically, around 20% of seed companies made it to Series A within 36 months. Recently, that has fallen to around 10%.
This is why so many seed companies feel trapped. They raised on a 2021-style venture narrative, but they now need 2026-style proof.
The software market is not dead. It is bifurcated.
The mistake is to conclude that software is no longer venture-backable. That is not what the market is saying. The market is saying that generic software is less fundable. Thin AI wrappers are less fundable. Tools with weak distribution, shallow workflow ownership, limited data advantage and no obvious path to durable value are less fundable.
Gil Dibner of Angular Ventures, made a quip about this, capturing the sentiment well:
But software companies that break out are still getting funded aggressively. In some categories, faster than ever. The difference is that investors are now asking harder questions:
- Can this company grow unusually fast?
- Can it capture a workflow, not just a feature?
- Can it own proprietary data or context?
- Can it become system-of-record, system-of-action, or system-of-intelligence?
- Can it expand from a wedge into a platform?
- Can it survive if the next Claude, Gemini, GPT or open-source model makes the core feature cheaper?
That last question is the killer. I wrote more about it in a recent post on VC Cafe. VCs used to ask: what if Google entered the market, now they worry about all model creators who are increasingly moving from just providing the infrastructure to also building the end products.
If the answer is no, the company may still be a good business. It may generate cash. It may serve a real customer need. It may be a great bootstrapped company. But it may not be a venture-scale company.
This is the distinction founders need to internalise. “Useful” and “fundable” are no longer the same thing.
Why AI makes the bar higher, not lower
There is a common assumption that because AI reduces the cost of building, it should increase the number of venture-backable companies.
In practice, the opposite may be true. When everyone can build faster, the premium moves from building to distribution, insight, timing, taste, trust and market access.
This is not new. Every major platform shift has this pattern. Cloud made software cheaper to start. Mobile made distribution global. Open-source made infrastructure more accessible. Each wave lowered creation costs, but raised the bar for differentiation.
AI is doing the same, only faster. For software startups, the defensibility stack is shifting from code to context. The strongest companies will not be the ones that simply use AI. They will be the ones that use AI to create a compounding advantage in one of four ways:
- First, they will own a high-value workflow where switching costs increase over time.
- Second, they will accumulate proprietary data, feedback loops or domain-specific context that improves the product.
- Third, they will have distribution that is hard to replicate, whether through community, ecosystem, partnerships, embedded channels or founder-market fit.
- Fourth, they will turn AI into margin advantage, not just product novelty.
That is why the “AI-native services” trend is so interesting. In many cases, the next generation of software companies may not look like software companies at all. They may look like lean, AI-enabled service providers that use software and agents to deliver outcomes, not seats.
If the customer wants the work done, not another dashboard, the company that owns the outcome may be more valuable than the company that sells the tool.
The venture response: prove it earlier
The venture market is adapting in real time. At pre-seed, founders can still raise on team, insight and speed. But even there, the best rounds increasingly include some evidence of demand: design partners, usage, waitlists, early revenue, a working demo, or a sharp wedge into a painful workflow.
At seed, “we can build it” is no longer enough. Investors want to know why this team can win, why now, and why the product becomes more defensible with usage.
At Series A, the bar is much more quantitative. The SVB/Carta ARR benchmarks show how much the market has reset. A founder raising a Series A today is not just pitching a story. They are being compared to companies already doing millions in ARR, often with AI-driven growth and lean teams.
The result is a harsher fundraising environment, but not necessarily an irrational one. If AI compresses product development cycles, investors should expect more proof. If AI makes feature replication easier, investors should demand stronger moats. If terminal value is less certain, investors should care more about early evidence of urgency, retention and willingness to pay. The bar has moved because the world has moved. SaaS isn’t dead, but the world, and what is fundable, is definitely changing, FAST.
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.