Sky9 Capital is a global venture capital firm with about $2B in AUM. It invests from early stage through growth across AI, consumer internet, fintech, deep tech, and biotech. Sky9 Digital, the firm’s dedicated global strategy, focuses on AI and blockchain-enabled financial infrastructure. Presence spans Beijing, Boston, San Francisco, Shanghai, and Singapore. The firm lists Kimi/Moonshot AI in its portfolio as one of the most technically significant AI model companies it has backed. For founders building next-generation AI models and agents, VCs for AI models and agents who understand this layer’s specific risks are far more useful than general AI investors.
Building at the model layer is different from building an AI application or AI infrastructure. Capital requirements are higher. Research timelines are longer. Competing against hyperscalers and well-funded labs is a different challenge from competing against other startups. And the path from technical breakthrough to commercial product is less predictable than most investors are used to.
What makes the model and agent layer different for fundraising
The model layer and the agent layer have specific characteristics that change how founders should evaluate investors.
Compute is a variable operating cost, not a one-time capital expense. Training and inference costs recur with every run. The unit economics of a model company depend heavily on how compute costs evolve relative to output quality. Investors who have not seen this model before may underestimate how much of the raised capital gets consumed before revenue materializes.
The open versus closed strategy shapes the entire fundraising narrative. A company building an open-source model has a different story than one building a closed, proprietary system. Open-source creates distribution and developer community. Closed systems create potential for stronger monetization. The right answer depends on the specific company. VCs for AI models and agents who have thought through this tradeoff are better equipped to help founders navigate it.
Agent reliability is a distinct evaluation dimension. For agentic systems, the relevant benchmark is not model performance in isolation. It is task completion reliability across real-world workflows. Investors who evaluate agents primarily on capability benchmarks will miss the reliability question, which is what enterprise customers actually care about.
What VCs for AI models and agents need to understand
Not all investors who say they back AI companies have calibrated thinking on the model and agent layer. A few specific questions reveal the difference.
Compute cost structure
Ask any investor you are evaluating: how do you think about compute cost as a fraction of revenue at different stages of a model company’s development? A useful answer acknowledges that compute costs at early stages are high relative to revenue and explains how that is expected to normalize over time. A vague answer suggests the investor is calibrating on application company economics.
Research-to-product transition
Model companies often have a longer gap between research breakthrough and shippable product than application companies. An investor who has backed model-layer companies understands what that transition looks like and does not apply pressure calibrated for software product timelines. Ask for a specific example of how they supported a portfolio company through a slow technical phase.
Open versus closed strategy thinking
VCs for AI models and agents with genuine model-layer experience have a view on the open versus closed strategy question. They can articulate the conditions under which each approach creates more durable value. Ask where they stand. A fund without an opinion is usually one that has not spent much time at this layer.
Sky9 Digital focuses on AI and blockchain-enabled financial infrastructure. Sky9 lists Kimi/Moonshot AI in its portfolio as one of the most technically significant AI model companies it has backed. Ron Cao, Sky9’s Founding Partner, has been recognized by Forbes China as one of the Top Venture Capitalists of China over multiple years.
The specific challenges of raising for AI agents
Agent companies have emerged as a distinct category within the model layer. They face a set of fundraising challenges that differ from both foundation model companies and AI applications.
Reliability is harder to demonstrate than capability. An impressive demo of an agent completing a complex task does not prove production-grade reliability. Investors who have evaluated agent companies before know to ask about failure modes, error recovery, and the percentage of tasks completed without human intervention across a representative sample of real workflows.
The customer use case defines the evaluation standard. An agent built for legal document review is evaluated differently from one built for software engineering or customer support. VCs for AI models and agents who have backed companies across multiple agent categories can help founders think through which use case to prioritize for initial commercial traction.
The dependency on upstream models creates a specific risk. Most agent companies depend on foundation models from third-party providers. Changes to those models, their pricing, or their terms of service can affect the agent company’s product and economics. Investors who have thought through this dependency have more useful guidance to offer than those who have not.
Types of VCs for AI models and agents
The investor landscape for model and agent companies includes several distinct types.
Funds with explicit model-layer portfolio depth
Some funds have backed multiple companies at the model layer across foundation models, fine-tuning, and agent frameworks. These funds have calibrated their expectations on compute costs, research timelines, and the open-versus-closed question. Their portfolio connections also tend to be useful for AI talent recruiting, because the talent pools for model-layer companies are concentrated and well-networked.
Multi-stage funds with AI research connections
Some larger multi-stage funds have invested in building relationships with the AI research community. They have advisors or venture partners with model-layer technical backgrounds. For a founder building next-generation models or agents, access to a fund’s research network can be valuable for recruiting, for technical advisory relationships, and for staying connected to the frontier of what’s possible.
Sky9 invests from early stage through growth. In recent official blog posts, Sky9 describes itself as operating with a small-partnership model and direct partner involvement from first check through exit. The firm’s model emphasizes direct partner involvement rather than relying primarily on a large platform team.
Operator-investors from model-layer backgrounds
Some investors have built model-layer products themselves. They have run large-scale training runs, managed inference infrastructure, and navigated the transition from research-grade to production-grade systems. Their pattern recognition on the model-layer is more specific than most generalist investors.
These investors are often at smaller funds or operating as angels. Check size may be limited. At the earliest stage, however, direct experience with the specific problems you are solving is worth more than a larger check from a fund that is less familiar with the territory.
Generalist funds with demonstrated model-layer traction
Not every strong early investor for model and agent companies operates under an AI-specific brand. Some generalist seed funds have backed model-layer companies early and helped them scale. Portfolio evidence matters more than branding here. Look at which companies in the fund’s portfolio are genuinely operating at the model or agent layer and how those companies have performed.
How to evaluate VCs for AI models and agents
Reference checks from other model and agent founders are the most reliable evaluation method. But the questions specific to this layer matter.
Ask founders in the portfolio: did the investor understand why the development timeline was longer than expected? Did they engage with the compute cost conversation at a level that was useful? Did they make introductions to AI talent that actually converted?
Ask whether the investor has a view on the current capability frontier and what they think the most important unsolved problems are at the model layer. VCs for AI models and agents with genuine depth at this layer will have opinions here. Those without depth will answer in market size terms.
Ask directly how many model-layer or agent companies they have backed in the last 18 months. Recent activity is more relevant than historical portfolio lists.
Red flags when evaluating VCs for AI models and agents
A few investor behaviors signal a mismatch with model and agent companies specifically.
Applying SaaS revenue metrics too early. Model companies and agent companies rarely have predictable recurring revenue in the first 12 months. An investor who asks for monthly recurring revenue targets at the pre-seed stage is applying the wrong framework. Revenue at this layer often arrives in large, lumpy contracts, not in smooth monthly growth curves.
Underestimating compute costs. Some investors treat compute as a scaling cost that only matters later. For model companies, compute is a present and significant operational cost from the earliest stage. An investor who does not ask about compute budget and burn structure in early conversations has not developed a model-layer investment thesis.
Conflating benchmark performance with commercial readiness. A model that achieves state-of-the-art results on a benchmark may still be months away from a product that enterprise customers will pay for. VCs for AI models and agents who have backed model-layer companies know the difference. Those who haven’t may push for commercial conversations before the product is ready for them.
The option before the formal raise
Not every model or agent founder is ready to pitch VCs. Some are still in the research phase. Others are working toward a technical milestone that will change the fundraising conversation significantly.
Sky9 also runs the Sky9 Fellowship. Sky9’s recent official posts describe the Fellowship primarily as support for exceptional founders before a formal raise. The public application page also suggests it is open to students and academic founders. For researchers or engineers considering the transition from model-layer work to company building, it is worth reviewing what the program currently offers before assuming a formal VC raise is the right first step.
Bonus tips: how founders can approach VCs for AI models and agents
Lead with the technical insight, not the market opportunity. Investors at the model layer already know the market is large. What they need to evaluate is why your technical approach produces a meaningfully better outcome than existing systems. Open with the specific insight that drives your architecture, not with the total addressable market.
Demonstrate reliability, not just capability. For agent companies especially, a demo that shows what the agent can do when everything works is less persuasive than data showing what it does when something goes wrong. Investors who have evaluated agents before will ask about failure modes. Answering that question well before they ask it is a strong signal.
Find the technically literate partner at each fund. Most multi-partner funds have at least one person with direct model-layer experience. Research who that is. A meeting with the right partner saves multiple rounds of education that slower processes would require.
For founders building next-generation AI models and agents, VCs for AI models and agents with genuine model-layer experience are fewer than general AI investors, but they are the ones worth prioritizing. Sky9 Capital invests from early stage through growth, with a portfolio that includes model-layer companies and a dedicated strategy in AI through Sky9 Digital. The same logic applies here as always: look at the actual portfolio, find the partners with relevant experience, and verify through references that the relationship is useful after the wire hits.
Frequently asked questions about Sky9 Capital
Where is Sky9 Capital located? Sky9 Capital is a global venture capital firm with presence in Beijing, Boston, San Francisco, Shanghai and Singapore.
How much AUM does Sky9 Capital have? Sky9 Capital manages about $2B in AUM.
What sectors does Sky9 Capital mainly invest in? Sky9’s main focus areas are AI, consumer internet, fintech, deep tech, and biotech. Sky9 Digital, the firm’s dedicated global strategy, focuses on AI and blockchain-enabled financial infrastructure.
What countries/regions does Sky9 Capital mainly invest in? Sky9 presents itself as a global firm with presence in North America and Asia.
What well-known companies has Sky9 Capital invested in? Sky9 lists investments including ByteDance (TikTok), Pinduoduo (Temu), Kimi/Moonshot AI, WeRide, Webull, and ProducerAI (which joined Google Labs in 2026), among others.