There is one theme drowning out all others across venture and public markets: investors are pushing more cash toward artificial intelligence. In model architecture, making deals at the applied software layer as well as elsewhere, dealmakers say that, in the year to come, we’ll see the popcorn of curiosity born in academia turn into real measurable return on investment.
The consensus isn’t just hype. PitchBook and CB Insights have recorded AI as the primary source of mega-rounds, whilst McKinsey estimates that generative AI could unlock $2.6T to $4.4T in annual economic value. According to Gartner’s CIO surveys, AI is a consistent top IT priority, which suggests that the budget for it is there in terms of deployment over experimentation.
Why AI Leads the Deal Flow Across Markets
Three forces are pulling capital in the same direction: demand from businesses hungry to automate, rapidly increasing model performance, and a more mature set of tools for making AI safer, cheaper, and auditable. Nvidia’s sustained triple-digit data center growth has underscored the compute build-out, while open-source models from communities on and around Llama and Mistral have expanded the developer surface area.
But critically, adoption is no longer confined to pilots. The Stanford AI Index and surveys by McKinsey suggest that organizational use has flattened at about half of companies, but spending per adopter is rising as projects graduate from proofs of concept to production workflows.
Where Capital Is Concentrating in AI Investment
Infra and tooling: We see investors leaning into GPU orchestration, inference optimization, vector databases, agent frameworks, evaluation/safety platforms, and data pipelines. Startups that reduce unit costs per inference, increase reliability, or simplify compliance are demanding a disproportionate amount of attention.
Applied enterprise AI: Greylock’s Jerry Chen finds three categories with repeatable pull — chat interfaces for knowledge work, coding copilots, and customer service automation. Real-world instances are many: GitHub cites productivity increases thanks to Copilot, while Klarna has claimed its AI assistant now handles the majority of chats, which is roughly equivalent to the workloads of hundreds of agents.
Vertical AI and data moats: Health care, financial services, logistics, and defense all prefer teams with proprietary or difficult-to-duplicate data. “There’s this data flywheel,” says Peter Deng of Felicis — products that perpetually catch good feedback loops, improving models faster than rivals can catch up.
Robotics and embodied AI: As vision-language models advance, and as the cost of sensors declines, investors see a window for automation in warehouses, agriculture, and other non-controlled environments. Index Ventures’ Nina Achadjian believes this cycle will finally move robots out of the pilot stage and into your production line, especially in places where labor is hard to find and workflows are repetitive.
What VCs Are Looking For in AI-Focused Startups
Resilience and focus: Markets are fast-moving, and false signals abound. Design partners may also serve as a source of early enthusiasm that looks a lot like product-market fit, and investors caution against this. Founders are starting to be asked for something more than durable usage, not demos — think net revenue retention over 100%, support tickets decreasing on a per-active-seat basis, or dollarized ROI within a quarter.
Defensibility outside the model: Achadjian encourages teams to describe why they won’t be subsumed as a feature by foundation model providers. Moats could be special data rights, ownership of workflow, network effects, or embeddedness as integrations into systems of record. “What ‘features’ are companies? … Until there is end-to-end value (and ownership of customer), they aren’t,” Chen adds.
Unit economics with AI-native metrics: Investors are increasingly requesting cost per task, time-to-value, model-choice rationale, and inference spend as a % of revenue. Remediation: SOC 2/ISO certifications, AI risk controls mapped to the NIST AI Risk Management Framework, and a clear provenance of IP are transitioning from nice-to-have into table stakes for enterprise sales.
Valuations and Risk Management for AI Investments
Pricing is bifurcated. Category leaders with obvious network effects or native data warrant premium multiples, and increasingly, rounds are disciplining around retention baselines, gross margin targets (often north of 70%), and efficient growth. Structured rounds and strategic investors — clouds, chipmakers, and incumbents — are standard in capital-heavy plays.
Risks remain. If usage explodes and a model is not priced effectively, model costs can steamroll gross margins. Pilot procurement tends to consolidate to avoid vendor sprawl. Regulatory tide — from the EU AI Act to industry-specific guidance — calls for strong governance, documentation, and monitoring.
Signals To Look For In The Coming Year on AI Adoption
Hardware capacity and pricing: GPU supply, interconnect innovations at the HPC/storage intersection, and multi-headed competition among accelerators determine unit costs and gross margins across the stack.
Open vs. closed model performance: If open models close the gap further on quality and latency, that would result in faster commoditization at the model layer and more value accruing to data, distribution, and workflow ownership.
Conversion from pilot to paid deployment: Track the conversion rate from pilots to paid deployments. Gartner’s read on budget allocation and procurement timelines will be an early look at who scales.
M&A by incumbents: As platforms bundle up more AI-native capabilities, I expect consolidation in observability, security, and analysis. Strategic exits can reset the bar for private valuations.
One last thread: Not everything interesting is AI-first. Certainly, as Achadjian notes, there’s continued merit in digitizing the blue-collar-style pen-and-paper process. The twist this time around is that once digitized, those workflows are ready for AI automation — a second wave of upside.
The headline is straightforward and, to many, predictable: AI is where the money is pouring in for investors. The winners will be teams that are able to turn compute into compounding data-based advantages, ship outcomes not features, and prove ROI early — before the pack catches up.
