$7B Periodic Labs (from $1.3B in 6mo)
$1.5B Granola (notetaker)
41% of all 2025 VC = AI
In the last week of March 2026, the AI funding market produced three data points that tell you more about the current investment climate than any quarterly report: a startup called Reflection AI — with no public model and no publicly released product — is in talks to raise $2.5 billion at a $25 billion pre-money valuation. Periodic Labs, a materials science automation company, is negotiating at approximately $7 billion — up from $1.3 billion just six months ago. And Granola, a London-based meeting productivity startup, quietly raised $125 million Series C at a $1.5 billion valuation.
These aren’t outliers. They’re the leading edge of a pattern that has been building throughout early 2026. AI startups took 41% of all venture capital in 2025. The global market raised $220 billion in the first two months of 2026. The VC market is bifurcating — mega-rounds for AI infrastructure leaders, disciplined funding for everyone else — and understanding how the bifurcation works is essential for anyone watching where capital is actually flowing.
The Wall Street Journal and Reuters reported on March 25 that Reflection AI is in talks to raise $2.5 billion at a $25 billion pre-money valuation — implying a post-money valuation of approximately $27.5 billion. The company is NVIDIA-backed. It is positioned explicitly as “seeking to counter Chinese AI” per the WSJ framing. And it has no publicly released model.
To understand this number, context helps. At $25 billion, Reflection AI would be valued more than ElevenLabs ($11B), larger than Wayve ($8.6B), and roughly comparable to Perplexity ($24B) — which actually has a product being used by tens of millions of people. Reflection AI’s valuation is entirely a bet on future capability, not current product-market fit.
1. Team capability: The founders have demonstrated ability to build frontier AI systems. In a field where top talent is genuinely scarce, the team is the product — at least at the pre-launch stage.
2. Compute access: NVIDIA backing isn’t just capital — it’s priority access to Vera Rubin hardware, early NemoClaw integration, and OpenClaw optimization. In a world where compute is the binding constraint, having NVIDIA as a backer is a structural advantage.
3. Geopolitical tailwind: “Seeking to counter Chinese AI” is a positioning statement that appeals directly to US government, corporate governance boards, and investors who want a domestic AI frontier lab alternative. That political framing is worth money in the current climate.
The honest analysis: Reflection AI’s $25 billion valuation is a bet that the next frontier model matters more than the current one — and that having a NVIDIA-backed, talent-dense, compute-prioritized lab building “against” Chinese AI is a uniquely fundable thesis in early 2026. Whether that thesis generates a product that justifies the valuation is a question the next 12-24 months will answer.
Bloomberg reported on March 24-25 that Periodic Labs, an AI-driven materials science and autonomous laboratory startup, is in talks at approximately $7 billion — raising “at least hundreds of millions.” That headline number becomes extraordinary when you add the timeline: in September 2025, Periodic Labs closed its seed round at $300 million and a $1.3 billion valuation. That’s a 5.4x step-up in under six months.
Periodic Labs represents the “Physical AI + scientific automation” investment thesis: AI as an autonomous laboratory that can perform experiments, discover new materials, identify drug compounds, and develop energy storage solutions — not through human scientists running experiments, but through AI systems that design, execute, and analyze experiments autonomously.
The investor thesis here is distinct from pure AI software: materials science breakthroughs have physical-world monetization paths. A new battery material discovered by Periodic Labs’s AI can be licensed to every EV manufacturer on Earth. A new semiconductor material can be sold to TSMC and Samsung. A new pharmaceutical compound can become a drug. The returns on scientific discovery, when they happen, are not software-scale — they are physical-world-infrastructure-scale.
Materials Science AI: Systems that discover new physical materials → revenue from licensing, manufacturing, and pharma → addressable market = everything physical (batteries, chips, drugs, energy)
The reason Andreessen Horowitz and Jeff Bezos backed Periodic Labs is that the upside scenario isn’t “better meeting notes” — it’s “AI discovers the material that makes solid-state batteries economically viable,” which is worth trillions across the energy and automotive supply chain.
Granola AI closed a $125 million Series C at a $1.5 billion valuation, led by Index Ventures and Kleiner Perkins, as reported by TechCrunch on March 25. The company is based in London and is genuinely popular in the Silicon Valley startup community — which creates a useful case study for understanding AI funding at the “non-mega-round” end of the spectrum.
Granola started as a meeting notetaker — AI that joins your video calls, transcribes them, and produces structured summaries. That’s now table stakes in a category where Otter.ai, Fireflies, and dozens of other tools exist. Granola’s differentiation is its evolution: it has expanded from meeting notetaker to “enterprise AI platform with persistent context, team collaboration Spaces, and developer APIs.”
The Granola raise is notable for what it represents: a disciplined, product-led funding round at a reasonable multiple, backed by top-tier investors, for a company that has expanded beyond its initial product category. This is the “non-AI-infrastructure” end of the bifurcated market — not $25B on a thesis, but $1.5B on product traction and a clear expansion narrative. Both ends of the market are funding simultaneously. They just have very different logic.
The Reflection AI valuation crystallizes something important about the current moment in AI investing: the market has developed a bifurcated logic. For infrastructure-layer AI companies — those building foundation models, compute infrastructure, or agentic OS layers — the valuation methodology is entirely forward-looking. Product is not the signal. Team + compute access + market positioning + geopolitical tailwinds are the signal.
This isn’t irrationality — it’s a specific theory of value creation. The argument is: if you believe the frontier model that will define 2027-2028 is going to be worth more than any current model, and if you believe NVIDIA hardware access is the binding constraint on who can build that model, then the team that has both the talent to build it and the compute priority to run it is worth betting on before they have a product. The option value on being the company that ships the dominant next-generation frontier model is enormous.
Tier 2 (Application + Platform): $1-5B valuations on product traction + expansion narrative. Examples: Granola ($1.5B). Logic: proven PMF with clear expansion path.
Tier 3 (Non-AI): Traditional VC multiples apply, with heightened scrutiny. The bar for non-AI companies has effectively risen because the opportunity cost of not being in AI infrastructure is now measurable.
Carta data shows AI startups took 41% of all VC invested in 2025 — out of a total $128 billion on platform. That’s not just a large share; it’s a structural shift in where the venture capital industry believes value is being created. For context: in 2020, AI was roughly 20% of VC. In 2022, it was about 30%. The acceleration to 41% in 2025 reflects the post-ChatGPT consensus crystallizing into capital allocation.
| Company | Valuation | Latest Raised | Product Stage | Thesis |
|---|---|---|---|---|
| Reflection AI | $25B | $2.5B (in talks) | No public product | Counter-China frontier AI |
| Periodic Labs | $7B | $300M+ (in talks) | Early / research phase | AI autonomous laboratory |
| Granola AI | $1.5B | $125M (Series C) | Live enterprise product | Meeting AI → enterprise platform |
| Perplexity | $24B | $500M+ (2025) | Widely used AI search | AI-native Google challenger |
| ElevenLabs | $11B | $180M (Series C) | Voice AI platform | AI voice infrastructure leader |
| Wayve | $8.6B | $1.05B (Series C) | Testing / deployment phase | Embodied AI for autonomous vehicles |
The March 2026 numbers show a slowdown from February’s extraordinary $189 billion US alone — March is tracking at approximately $13 billion US. But this is misleading: March lacks the mega-rounds that inflated February’s numbers. The underlying market is healthy: early-stage funding is up 47% year-over-year. The base is being built.
The honest answer, based on current capital allocation data: it’s harder. When AI startups take 41% of VC, the remaining 59% is competed for by everything else — fintech, biotech, climate, enterprise SaaS, consumer. The opportunity cost of not being in AI is now tangible to most LPs, which creates pressure on non-AI GPs to justify their thesis.
But there’s a more nuanced reading: the 41% figure is skewed heavily toward the top of the market. Most of the AI capital is concentrated in fewer than 100 companies. The long tail of AI startups faces the same market dynamics as any other category — product-market fit, defensible distribution, clear monetization. The “AI gets funded” narrative is really “AI infrastructure leaders get funded at extraordinary multiples.” For everyone else, the fundamentals still apply.
The $25 billion bet on Reflection AI without a product, the $7 billion bet on Periodic Labs after six months, and the $1.5 billion bet on Granola’s meeting platform expansion — these three data points together map the full topology of the current AI funding market. Understanding the logic of each isn’t just interesting. It’s the prerequisite for navigating what comes next.
💰 February 2026 Record VC Funding: $189 Billion Analysis
🦄 AI Infrastructure Startups and Unicorns 2026
🔍 Perplexity $24 Billion and the Google Chrome Bid: Explained