Where Venture Capital is Flowing in 2025: A Deep Tech Investment Map

The VC market in 2025 looks nothing like 2021, which is healthy. The frothy multi-stage deals that drove valuations up across the board are gone. What's left is more interesting: genuine deep tech investment, defense tech finally having its moment, and climate tech finding real revenue models. If you're a public market investor trying to understand where the next decade's interesting companies are coming from, paying attention to where VC capital flows now is one of the few actual information edges available.

The 2025 VC landscape is bifurcated in a way that's historically unusual. Capital is extremely concentrated in AI-adjacent categories while simultaneously contracting for companies that can't credibly connect to the AI narrative. Understanding both sides of that split is essential context for allocating across public and private markets.

Total VC Deployment: The Numbers in Context

Global VC deployment peaked around $680B in 2021 and collapsed to roughly $285B in 2023. The recovery since has been real enough that 2025 landed as the third-largest year on record. Read the headline and you'd think the winter was over. It isn't, and the composition is the reason.

The concentration is the whole story. AI companies raised about $211B globally in 2025, up from roughly $114B in 2024, and that's close to half of every venture dollar deployed on the planet. OpenAI and Anthropic by themselves took something like 14% of global venture investment. Two companies. Fourteen percent. This isn't purely enthusiasm, and the revenue is genuinely showing up. But it also means the rest of the startup world, the majority of companies by count, is competing for a much smaller pool than the headline number suggests.

The AI Infrastructure Gold Rush: Which Layers Are Getting Funded

Not all AI investment is equal. The VC community has developed a rough mental model of the AI stack that drives allocation decisions:

• Foundation models: OpenAI, Anthropic, Mistral, xAI, and Cohere. Foundation model companies pulled in roughly $80B in 2025, about 40% of all AI venture funding. This layer is now purely a scale game. Frontier training runs cost enough that competition is limited to a handful of very well-capitalized players, and new entrants mostly target open-weight model development rather than trying to fight at the frontier.
• AI infrastructure: compute orchestration (CoreWeave, Lambda Labs), inference optimization (Together AI, Fireworks AI, Groq), observability (Weights & Biases, Arize AI, Langsmith), and vector databases (Pinecone, Weaviate, Qdrant). This layer is well-capitalized and increasingly competitive. Margins will compress as the market matures.
• AI applications: vertical-specific AI tools (Harvey for legal, Abridge for healthcare, Glean for enterprise search, Cursor for engineering) are the current high-velocity layer. These companies can reach $10M+ ARR faster than any prior software generation because they sell productivity ROI that is immediate and measurable. Valuations remain elevated, at 20-30x ARR for the fastest-growing vertical AI companies.
• Enablement infrastructure: data labeling, synthetic data generation (Scale AI, Gretel), fine-tuning platforms (Modal, Replicate), and AI safety/red-teaming tools. This is the "picks and shovels" layer that wins regardless of which foundation model ultimately dominates.

The application layer will create more durable public companies than the foundation model layer, because applications can build distribution and vertical data moats that pure model capabilities cannot replicate. The infrastructure layer will consolidate rapidly as margin compression forces exits.

Defense Tech: The Most Counterintuitive Growth Sector Right Now

Defense tech is the most counterintuitive growth sector in VC right now, and I mean that in a specific way: a lot of great engineers who would have joined a consumer startup five years ago are now working on defense problems. That's a meaningful shift, and it has compounding effects on what's possible in the category.

Until 2020, the major venture firms were culturally and contractually reluctant to fund defense applications. Partner conflicts, university technology licensing restrictions, and ESG-oriented LP mandates created real friction. That posture has reversed sharply. The Russia-Ukraine conflict, Taiwan Strait tensions, and the demonstrated performance gap between commercial technology and legacy defense procurement have created political and economic urgency. Andreessen Horowitz launched a dedicated American Dynamism fund. Founders Fund has been in defense since Palantir. General Catalyst, Lux Capital, and Shield Capital are writing checks that would have been culturally controversial five years ago.

The companies attracting capital fall into three categories: autonomous systems (Anduril's autonomous defense platforms, Shield AI's pilot AI), AI-enabled command and control (Palantir's AI Platform, C3.ai's defense applications), and cybersecurity for critical infrastructure. Anduril is the proof. It raised $1.5B at a $14B valuation in 2024, then $2.5B at $30.5B in mid-2025, then $5B at $61B in May 2026. Roughly 4x in under two years. Whatever you think of the multiple, that trajectory settled the argument about whether defense can produce venture-scale returns, and it pulled in LP capital that had sat the sector out for a decade.

The structural tailwind is procurement reform. The DIU (Defense Innovation Unit) and AFWERX programs provide faster acquisition pathways for commercial technology companies, bypassing the traditional DARPA/prime contractor model that made government revenue inaccessible for startups. A company like Anduril can now generate meaningful DoD revenue within 18 months of a contract award rather than the 5-7 year traditional procurement cycle. For public market investors who want exposure to these trends without waiting for an IPO, Palantir and L3Harris are the most direct proxies right now.

Climate Tech: What's Getting Funded and What Isn't

Climate tech VC investment has been more resilient than the broader market correction, sustained by IRA incentives (production tax credits and loan guarantees for clean energy infrastructure) and genuine urgency around decarbonization timelines. But the funding is highly selective.

Fusion energy has attracted more serious capital than at any prior point: Commonwealth Fusion Systems, Helion (backed by Sam Altman and Microsoft), TAE Technologies, and Zap Energy have collectively raised over $5B. The DOE's National Ignition Facility achieving scientific breakeven in 2022 was a genuine technical milestone that shifted investor perception from "speculative" to "engineering problem." Commercial fusion power by 2035 remains optimistic. By 2040 is plausible for the best-capitalized teams.

Geothermal is underappreciated relative to its potential. Enhanced Geothermal Systems (EGS) use oil and gas drilling technology to create heat exchange loops in hot dry rock, providing dispatchable baseload power with near-zero emissions. Fervo Energy, which uses horizontal drilling techniques borrowed from fracking, is the leading VC-backed company in this space. The sector benefits from the fact that it can hire directly from the contracting oil and gas workforce, which solves one of the biggest bottlenecks in energy transition.

Grid software, covering transmission planning, demand response, virtual power plants, and energy storage optimization, is the highest-margin, fastest-payback segment of climate tech. Companies like Leap, AutoGrid, and Voltus operate software businesses with SaaS economics in a regulated infrastructure market that's being forced to modernize rapidly by renewable intermittency.

What's not getting funded: direct air capture at scale (LCOE remains uneconomic without sustained carbon credit pricing), long-duration storage without a contracted off-taker, and hardware-heavy climate solutions in markets where Chinese manufacturers have already established dominant cost positions (solar panels, lithium-ion batteries).

Biotech AI Convergence: The Most Patient Capital in the Market

The application of foundation models to drug discovery, protein structure prediction (AlphaFold's impact on structure-guided drug design), and clinical trial optimization represents a genuine scientific inflection point. Recursion Pharmaceuticals, Insilico Medicine, and Isomorphic Labs (the DeepMind spinout) are the names to watch. Worth noting that Recursion absorbed Exscientia in late 2024, which tells you something about the category: consolidation is arriving before the first approval does.

The investment thesis requires patience measured in years, not quarters. The fastest a drug can move from AI-identified target to Phase II readout is 5-7 years. The capital requirements are enormous: a Phase III oncology trial costs $300-500M. Investors with a 10-year time horizon and tolerance for binary outcomes (trial success or failure) can access exceptional upside if the AI-discovered molecule pipeline begins generating Phase II data. The first AI-native drug to reach commercial approval will be a landmark event that re-rates the entire category.

The Series A and B Crunch for Non-AI Companies

The most actionable insight for understanding the current venture environment: Series A and B funding for companies that can't credibly attach to the AI narrative has contracted dramatically. Median time from seed to Series A has extended from 18 months (2021) to 28 months (2024). Conversion rates from seed to Series A have dropped from approximately 40% to 25%. Many seed-stage companies that raised in 2022-2023 on 2021 multiples are unable to raise a Series A at any valuation that doesn't represent a down round.

This dynamic has positive long-term implications that are counterintuitive in the short run. The companies that survive this crunch will have demonstrably better unit economics, lower burn rates, and higher revenue quality than the cohort that raised in the flood of 2021 capital. The 2025-2026 Series A and B vintage will likely produce better returns than the 2021-2022 vintage, for the same reason that the 2009-2013 vintage outperformed the 2000-2001 vintage.

Geographic Shifts: Middle East and Southeast Asia

Two geographic shifts in LP capital deserve attention. Gulf sovereign wealth funds (Mubadala, Saudi Arabia's Public Investment Fund, ADQ) have become major LPs in top-tier US venture funds and are increasingly co-investing directly in late-stage AI companies rather than just writing fund commitments. PIF's direct technology portfolio and Abu Dhabi's concentrated bets on AI infrastructure represent a large pool of genuinely patient capital, and patient capital changes deal dynamics at the growth stage. It means a company can stay private longer, which is exactly what has been happening.

Southeast Asia's venture ecosystem has matured significantly. Singapore, Indonesia, and Vietnam have produced second-generation founders who have successfully exited once and are now building more sophisticated companies. Grab, Sea Group, and GoTo's trajectories have demonstrated that the region can produce public market scale outcomes, attracting more serious allocations from global fund managers.

Reading VC Signals as a Public Market Investor

VC activity is public information with a lag. The most useful signals to monitor: Crunchbase and PitchBook funding announcements for category-level trends, GP commentary in limited partner letters (which sometimes become public), and the IPO pipeline itself. The companies registering S-1s today were Series B and C investments 3-5 years ago.

The practical framework: identify the VC-funded companies in a category that are approaching the scale at which they could go public ($200M+ ARR for SaaS, $1B+ GMV for marketplaces). Build a research position before the IPO. Evaluate the S-1 rigorously. The investors who develop conviction about a company's unit economics before the roadshow will make better allocation decisions than those reacting to the IPO narrative.

The 2025 vintage of VC investment is deploying into a more disciplined environment than 2021, at better entry valuations, with higher bars for demonstrating product-market fit. The cohort of companies that emerge from this environment, battle-tested on lean budgets with real revenue and clear economics, will likely produce the defining public market technology companies of the late 2020s. Following the capital now is how you find them early.