Venture at the Speed of Software

Four years in, how has AI reshaped Venture?

Written by: George Patin | Research Contributor, London Venture Capital Network

At LVCN, we have often written about where venture is focusing its attention. A parallel question, though, is how VC itself has changed. As with many other industries, AI has begun to reshape venture from the inside out. We track how, and what's next for the AI-driven VC.

Global startup investment reached $510 billion in the first half of the year, already above the $440 billion deployed across all of 2025. That sounds like a broad reopening of the market. It is not quite that. OpenAI and Anthropic alone accounted for $217 billion — 43% of the global total — while more than 70% of Q2 funding went to AI companies. Venture has entered a new record cycle, but also its most concentrated one.

At the same time, the industry's OS is undergoing rapid reinvention. Sourcing platforms can scan millions of companies; frontier models can read every deck, assemble a market map and produce a first-pass memo; portfolio systems can autonomously collect metrics, and at this rate may very well raise them in board meetings on their own. AI hasn't quite automated venture; rather, it is layering itself on top of its fundamentals and pushing them into overclock.

The more interesting question, then, is not whether AI will replace venture capitalists. An industry comprised of AI-designed pitch decks reviewed by AI VC models is a rather dreadful state of affairs. Rather, the trend is in which parts of venture become software, what remains stubbornly human, and where the advantage moves once every firm has access to roughly the same toolkit.

The Market Post-Reset

The post-2021 correction is, in headline terms, over. Funding recovered through 2025 and accelerated dramatically in 2026; Q1 set a new quarterly record at roughly $300 billion, followed by another $205 billion in Q2. Yet the underlying market looks less like the broad software boom of the 2010s and more like a barbell.

At one end sit frontier-model labs and capital-intensive companies in compute, defence, robotics, biotechnology and energy – many of which we have covered in previous issues. Their rounds increasingly resemble infrastructure finance: enormous cheques, strategic investors and long time horizons. At the other sits a dense layer of seed-stage companies made cheaper to build by AI. In the middle, particularly when it comes to conventional SaaS without a clear AI advantage, capital either remains selective or is dwindling altogether.

The geography is similarly uneven. US companies captured two-thirds of global startup capital in Q2. Europe deployed $25.6 billion across 1,636 deals in the same quarter, with AI and defence leading alongside biotech and energy. The region remains scientifically credible and increasingly ambitious, but its growth-stage capital stack is still far smaller than America’s.

The Capital Landscape

Concentration is happening at the fund level, too. In the US, Andreessen Horowitz, Thrive Capital and Founders Fund captured 48.1% of all capital raised by venture firms through the first half of 2026. LPs have returned to brand, scale and realised track record, leaving many emerging managers in a much harder fundraising market even as startup investment breaks records.

What has emerged is a three-part structure. First are the multi-stage platforms — a16z, General Catalyst, Lightspeed, Sequoia, Accel, Thrive — able to lead billion-dollar rounds and offer founders recruiting, policy, go-to-market and capital-markets support. Second are specialist firms such as Lux, DCVC, Eclipse and Lowercarbon, whose advantage comes from technical depth and networks in difficult sectors. Third are smaller, thesis-led seed funds, scouts and solo GPs, competing through speed, access and a precise view of an emerging category.

AI cuts across all three. Large firms can spread the cost of proprietary platforms across billions in assets. Specialists can encode a narrow thesis into better search and diligence. Small funds can operate with a team that would have been impossibly lean five years ago. The same technology therefore reinforces incumbents and lowers the minimum efficient size of a new fund. Venture is getting bigger and smaller at the same time.

The Investment Stack is Shifting

Sourcing is changing fastest. The old model relied on introductions, events, inbound decks and an associate’s ability to keep a market map in their head. The new one has now advanced from its n8n automation days into something like partial autonomous observability: company registries, hiring velocity, GitHub activity, patents, web traffic, founder movements and financing signals, all scored against a firm’s thesis.

Harmonic and Specter turn the private company landscape into a searchable signal graph; Evertrace looks for founders before a company is publicly announced; Affinity maps the warmest route through a firm’s network. Internal systems go further. SignalFire’s Beacon spans hundreds of millions of people and tens of millions of organisations; InReach says its DIG platform sources the majority of its investments; EQT’s Motherbrain and Moonfire’s Launchpad join external data to internal history. And these are just the systems that have been publicly disclosed.

The result is a much wider top of funnel. Data Driven VC mapped 235 firms building serious data capabilities in 2025, up from 151 two years earlier. A third said more than 40% of their deal flow was already generated by data tools.

Quantity is, as always, not quality. On that front, AI is also starting to occupy an outsized role.

Diligence is the next layer. Tools such as Deckmatch and AlphaLens structure inbound decks, enrich company records and test them against an investment thesis. Hebbia and Rogo can search data rooms, compare claims across documents and draft sections of an investment memo. General-purpose models can build a first market map, produce customer-interview questions or stress-test a financial model in minutes.

This is a genuine productivity shift, but not yet an autonomous investment committee. Early-stage company data is sparse, self-reported and highly path-dependent. A model trained on yesterday’s winners is going to be excellent at finding companies that resemble yesterday’s winners — potentially a death sentence in an industry where contrarian bets made early are half the winning strategy.

While portfolio management hasn't moved as rapidly, it may eventually prove just as consequential. Standard Metrics, Visible and Rundit automate the collection of revenue, burn, runway and ownership data. The next step is not another dashboard; it is an agent that notices a weakening sales pipeline, compares it with the rest of the portfolio and recommends an intervention. SignalFire already uses Beacon across recruiting and go-to-market as well as sourcing. In other words, the same data layer that helps select a company can help it compound afterwards.

Where, Then, the Next Moat?

The trouble with AI is that every fund has AI and then some. Much of the moat moves upstream into proprietary data and downstream into relationships, judgment and execution.

The best systems will learn from information other firms do not have: which signals led to a meeting, why a deal was rejected, how the company performed, which partner helped, and where the original thesis was wrong. That creates a feedback loop between sourcing, selection and portfolio outcomes. The models matter, but the real edge may well be in institutional memory and learning velocity.

There are risks. Automated sourcing can replace one narrow network with a set of invisible proxies for the same network. Synthetic memos can give weak evidence the appearance of completeness. If every investor uses similar datasets and similar models, the industry may converge on the same companies faster, increasing both prices and blind spots. Speed is an advantage only if it does not compress disagreement out of the process. If everyone runs the same stack, then it's right back to level playing field.

Where changes will certainly follow is in the org chart. If autonomous systems can sift through signals and canvass for interesting companies the world over, even analysts will have to shift far more towards effective judgement, conviction and presentation. We may well see leaner, meaner AI-augmented VC orgs where everyone has a thesis and a bet. We are also likely to see extensions into new areas beyond capital alone: platform support similar to a16z's new media team, employee sourcing, more emphasis on helping the startups get critical customers, and more. If so, then the value proposition of VC itself will start leaning from funder to partner.

What Venture Looks Like Next

By 2030, a venture firm may continuously screen most of its addressable market, generate an initial diligence pack before the first meeting and run portfolio support through a set of specialised agents. A small partnership will be able to cover far more ground; a large platform will behave increasingly like a growth partner with an investment fund attached.

But the fully autonomous VC is probably the least interesting end state. Venture returns come from exceptions, and exceptions are difficult to infer from historical patterns, no matter how extensive your agentic tooling. Founders also choose their investors. Board decisions carry accountability. Price, timing, trust and the ability to help in a crisis remain relational.

The defining firm of this cycle, then, will not necessarily be the one that scores the highest on AI adoption. It will be the one that turns every search, conversation, decision and portfolio outcome into a compounding intelligence system — and then retains enough independent judgment to make the final call itself.