What We Learned This Year From 6 Active AI Investors Backing Startups Across The Stack
AI dealmaking hit overdrive in 2025. By the third quarter, nearly half of all global startup funding flowed to AI companies, and overall venture investment rose more than a third year over year, driven by record megadeals. AI startups raised roughly $100 billion in the first half alone, nearly matching all of 2024, according to Crunchbase data. We spoke with six active AI investors—Accel, Foundation Capital, Dell Technologies Capital, Sierra Ventures, AI Fund, and GV—about where the market is going, what’s scarce, and where startups can still win.
Accel: Competing in a world of AI superpowers
The “Super Six”—Nvidia, Microsoft, Apple, Alphabet, Amazon and Meta—are plowing vast operating cash into AI infrastructure, redrawing the power map. Accel partner Philippe Botteri argues that while incumbents command scale, focused AI-native startups can still carve out durable categories or reinvent legacy ones, especially where speed, product depth and distribution create wedges.
Accel has been among the most active on the Crunchbase Unicorn Board, backing model and application players including Anthropic, H Co. (small models), Nebius Group (infrastructure), and app-layer teams such as Anysphere (Cursor), Perplexity, Synthesia and Cyera. Botteri’s thesis: if generative AI doesn’t materially lift global productivity—on the order of low single-digit GDP gains—then the current level of investment wouldn’t make sense. Accel is betting it will.
Foundation Capital: AI meets physics, power and people
As models scale, constraints are increasingly physical: chips, power and data centers. Botteri’s analysis flags a potential 117-gigawatt energy shortfall over five years to meet projected AI demand—akin to powering several large European economies. Foundation Capital’s Steve Vassallo has long focused on that physical substrate. The firm incubated Cerebras Systems in 2016, anticipating that traditional architectures would hit a wall under AI workloads. That call looks prescient as Nvidia’s market cap tops $4 trillion and Cerebras signals IPO plans.
Vassallo says the standout companies will pair technical excellence with products designed around human behavior. Reinforcement learning with human feedback is one example: not only does it shape model behavior, it also trains people to collaborate more effectively with AI systems. Foundation’s portfolio includes Tennr (automating authorization workflows in healthcare), Jasper (writing assistance on top of GPT-3), and PlayerZero (predicting and debugging failures in AI-authored code pre-deploy). The throughline: founders operating at the frontier where AI, infrastructure and human factors intersect.
Dell Technologies Capital: Betting at the silicon layer—and beyond
Dell Technologies Capital (DTC) sits where enterprise demand meets infrastructure supply. Dell expects around $20 billion in AI server shipments by fiscal 2026, and DTC has logged a brisk streak of exits since June despite a sluggish broader market. With Dell a major GPU server provider, the venture arm sees most serious enterprise AI buyers and builders up close—and the fundraising tempo has been blistering.
DTC invests at the silicon level because that’s where ecosystem disruption can be profound. Its bets include Rivos (AI chips—now slated to be acquired by Meta, pending approvals), SiMa.ai (edge inference for autos, drones and robotics), and Runpod (developer layer with on-demand GPU access). On the application side, DTC has backed Maven AGI (complex, high-compliance customer support) and Series Entertainment (a GenAI platform to accelerate game development). The thesis: own critical layers where performance, cost and availability shape everything above.
Sierra Ventures: The data-first, layered-cake thesis
If compute is the bottleneck, data is the moat. As DTC partner Elana Lian notes, “AI is almost a data problem”: improvements increasingly hinge on high-quality, domain-specific data. Sierra Ventures managing partner Tim Guleri applies a “layered cake” framework—spanning infrastructure, applied infrastructure on top of foundation models, horizontal apps, vertical apps, and net-new products only possible with AI.
Sierra avoids the most capital-intensive base layers, leaning into applied infrastructure and applications where proprietary data and crafty distribution can create defensibility. The firm looks for teams attacking painful workflows with 10x productivity gains and privileged data access. With global GDP around $110 trillion—and more than $100 trillion beyond agriculture—Guleri expects AI efficiency gains to compound across services, software-driven industries and regulated verticals for years to come.
AI Fund: Building with hard-to-access data via corporate partners
Andrew Ng’s AI Fund takes a venture studio approach: source ideas with corporate LPs, validate markets, and recruit CEOs to co-found companies. Partners including AES, HP, Mitsui & Co. and Mitsubishi help the studio enter complex sectors—renewables, industrials, insurance—where internal data is both critical and difficult to unlock. Many startup concepts originate directly from these partners’ pain points in vast, under-digitized markets.
Unlike traditional VC, AI Fund’s core activity isn’t winning competitive deal flow—it’s creating it. The team identifies opportunities, pressure-tests customer demand, and assembles founding teams with privileged data access from day one. Ng sees ongoing openings in visual and voice AI and emphasizes that “AI” is not a monolith but a suite of techniques spawning distinct market categories. The edge comes from deep vertical focus plus defensible data pipelines.
GV: Flexible capital, stack-wide, even at premium prices
GV, Alphabet’s independent venture arm, has emerged as one of the most active corporate investors in AI, backing everything from chips and compilers to applications at both early and late stages. Despite Alphabet being its sole LP, GV is comfortable funding startups that may compete with Google’s efforts—much as it once invested in Slack.
In today’s environment, GV partners Dave Munichiello and Tom Hulme are ready to pay up when growth justifies it. Many AI startups are posting revenue run rates and acceleration curves that outpace prior cycles, making non-AI comparisons feel quaint. The message: when usage, retention and monetization bend the curve, price discipline should account for velocity—and for the chance to back platform-defining companies.
The takeaway
Across these perspectives, three themes recur: physical constraints (power, chips, data centers) will shape the ceiling; proprietary, high-signal data will define moats; and company creation models are evolving, from corporate-backed studios to layer-specific plays. The incumbents have scale, but the door remains open for AI-native startups that move fast, own critical layers or datasets, and solve hairy, high-value workflows. The wave is here—now it’s about where, and how, to surf it.
