When Carta published its 2025 venture capital data in March 2026, the headline number was striking: AI companies received 41% of all venture capital tracked on the platform in 2025. That’s $52+ billion of an estimated $128 billion total flowing to one category of company.

Then came January and February 2026 — and 41% started to look restrained. OpenAI closed a $110 billion round. Anthropic raised $30 billion. Waymo took $16 billion. In February alone, $189 billion in venture capital was deployed globally, with AI-related companies capturing the overwhelming majority. Three companies took $156 billion — 83% of a record month.

In March 2026, the headline numbers dropped to ~$13 billion in the US as of March 23. The media narrative immediately jumped to “VC winter” and “correction.” That narrative is wrong. This article explains why — and what the data actually says about the state of venture capital, AI concentration, and what founders should do next.

Forty-one percent sounds like an extraordinary concentration. To understand whether it is, it helps to look at how previous technology platform transitions shaped VC allocation patterns.

During the peak of the internet bubble (1999–2000), internet-related companies captured an estimated 60–70% of all venture capital in the United States. The mobile transition (2010–2014) saw mobile-related startups take roughly 35–40% of VC at peak. The cloud transition (2015–2019) similarly concentrated capital in SaaS and cloud infrastructure — though more distributed because “cloud” was a delivery model rather than a discrete industry category.

By that historical lens, AI at 41% of 2025 VC is steep — but it is not yet at the dangerous concentration levels of the 1999–2000 internet peak. What makes AI different from those prior transitions is the speed: AI went from near-zero VC concentration to 41% in approximately three years (2022–2025), which is a faster ramp than either mobile or cloud.

February 2026’s $189 billion in global VC is the most capital deployed in any single month in venture capital history. It will likely hold that record for years. But to understand what it means, you have to disaggregate three deals from the rest:

$156 billion out of $189 billion. These three deals alone represent 83% of February’s total. Strip them out, and February 2026’s venture market — $33 billion for all other companies globally — looks approximately normal for a healthy market recovering from the 2022–2023 correction.

That context is critical. The “AI is eating all the money” narrative is accurate at the mega-round level. It is substantially overstated when applied to the broader venture market. The early-stage and mid-stage companies outside of the AI infrastructure tier are raising in an environment that is, if anything, improving.

By March 23, 2026, US venture activity for the month had reached approximately $13 billion — a 93% month-over-month decline from February’s $174 billion US total. This number has generated “VC winter” headlines in several publications.

The “winter” narrative makes a statistical error: it is comparing a month with three historically anomalous mega-rounds to a month without them. The correct comparison is February 2026 (ex mega-rounds) vs March 2026 — and on that basis, March looks entirely normal. A $13 billion month in the US, annualized, implies $156 billion annually, which is above 2023’s total and roughly consistent with a healthy mid-cycle market.

New unicorns being minted in March 2026 also do not suggest a frozen market:

Three unicorn or near-unicorn events in one week of March 2026. This is not a frozen market. It is a market bifurcated by tier: mega-rounds at the infrastructure layer (Reflection AI, OpenAI, Anthropic) and normal-sized rounds at the application layer (Granola, most B2B SaaS companies).

Reflection AI has no public model. No product page. No announced roadmap. It has: a team, NVIDIA’s backing, and a thesis centered on countering Chinese AI development. That combination is apparently worth $25 billion pre-money to investors participating in the round WSJ and Reuters reported on March 25, 2026.

The $25B-no-product signal has two plausible interpretations. The bearish reading: this is exactly the kind of irrational exuberance that precedes a market correction. Capital is flowing to teams with slide decks and no demonstrated capability. The late-stage parallel to 1999 internet IPOs.

The bullish reading — and the one most consistent with sophisticated investors’ behavior — is more nuanced: at the AI infrastructure layer, what you’re buying is not product-market fit; it’s future compute access, key talent density, and strategic positioning. Reflection AI’s NVIDIA backing means guaranteed access to next-generation GPU allocations when they matter most. In a world where compute access is the binding constraint on frontier AI development, that access is worth billions before a single line of production code is written.

The “Chinese AI competitive pressure” thesis adds a national security premium. Investors and limited partners who would not normally tolerate pre-product risk are accepting it when the investment can be framed as participating in the US AI competitive defense. That framing — explicitly present in Reflection AI’s reported pitch — unlocks pools of capital that pure commercial AI startups cannot access.

Carta’s 2025 data includes what is perhaps the most underreported number in the entire VC cycle: early-stage investment is up 47% year-over-year in 2025. Not late-stage. Not mega-rounds. The seed and Series A companies — the foundational layer of the startup ecosystem — are raising more capital than at any point since 2021.

This matters because early-stage VC is the leading indicator of the innovation pipeline. The late-stage mega-rounds you read about in 2026 represent companies that raised seed in 2021–2023. The early-stage companies raising in 2025 will become the late-stage stories in 2028–2030. A 47% YoY increase in early-stage investment means the innovation pipeline is filling faster than at any point in recent history.

The headlines focus on late-stage because the dollar amounts are larger and the company names are more recognizable. But for the health of the venture ecosystem, early-stage data is the signal that matters — and that signal is unambiguously positive.

Non-AI B2B SaaS is also not dead — it’s just not getting the attention. Carta’s data shows that enterprise software, fintech, and healthcare technology startups are raising at roughly 2019–2020 levels: down from the 2021 peak, but stable and improving. The narrative that “non-AI founders can’t raise” is a product of press coverage, not capital market reality.

The most accurate description of the 2026 venture market is not “AI bubble” or “VC winter.” It’s bifurcation: a two-tier market where AI infrastructure companies raise at unprecedented valuations and speeds, while non-AI and application-layer companies raise in a normal — even improving — environment.

The bifurcation is primarily visible at three dimensions:

1. Dollar concentration vs. deal concentration: AI commands a disproportionate share of VC dollars but not a proportionate share of deal count. Most VC deals in 2025–2026 were not AI companies. The dollar concentration comes from a small number of extremely large rounds, not from AI winning the majority of funding competitions.

2. Geographic concentration: US companies took 92% of February 2026’s global VC. This represents a reversal of the 2015–2021 trend toward globalization of venture capital, driven partly by regulatory uncertainty in China and Europe and partly by the concentration of frontier AI talent and compute access in the US.

3. Stage concentration: The bifurcation between infrastructure mega-rounds and application-layer normal rounds creates a valuation gap. Infrastructure AI companies (OpenAI, Anthropic, Reflection AI) are priced on potential market capture of the entire AI economy. Application AI companies (Granola, and thousands of B2B SaaS AI features) are priced on revenue multiples — much more grounded.

The current concentration of VC in AI infrastructure is not permanent. It will end when the infrastructure layer IPOs — and the LP liquidity events from those IPOs generate a new wave of deployable capital.

OpenAI, Anthropic, and xAI are all tracking toward public markets at different timelines — OpenAI reportedly targeting 2026–2027, others slightly later. When these companies IPO at valuations of $300B, $50B, and $80B+ respectively, the limited partners who participated in their funding rounds will receive substantial returns. Those returns, historically, are reinvested into the next generation of venture funds — which then deploy into the next wave of startups.

The historical pattern: Google’s 2004 IPO generated LP liquidity that funded the Web 2.0 wave (YouTube, Facebook, Twitter). Facebook’s 2012 IPO funded the mobile app economy. Uber and Lyft’s 2019 IPOs funded the delivery and gig economy wave. OpenAI’s IPO will likely fund whatever comes after the current AI infrastructure buildout.

The founders who survive the concentration period — who have extended runway past the mega-round era, built real products with real customers, and positioned themselves for the post-IPO reinvestment cycle — will have the wind at their backs. The founders who burned capital chasing AI hype without demonstrable revenue will not.

February 2026: The Record $189 Billion VC Month — Full Analysis →
Reflection AI $25B, Periodic Labs $7B: The March 2026 AI Unicorn Wave →
AI Infrastructure Startups and Unicorns: The 2026 Landscape →