AI infrastructure is the most technically demanding category in venture right now. It is also the one where the gap between funds that can genuinely evaluate an inference optimization architecture and funds that just follow the category narrative is widest.
Founders building AI infrastructure (compute, model serving, data pipelines, evaluation tooling, deployment infrastructure) are pitching to a different set of investors than those building AI applications. The diligence questions are different, the moat signals are different, and the funds that move fastest are the ones that already have a formed view.
Seed-stage AI infrastructure investors who move first share a common characteristic: they have built enough internal technical depth to reach conviction on an architecture decision before the market has validated it.
This piece identifies which funds are the fastest-moving earliest-stage AI infrastructure investors in 2026, what they look for, and where Sky9 Capital fits in this landscape.
What Makes AI Infrastructure Different to Fund
AI infrastructure companies solve genuine engineering problems under constraint: inference cost, latency, reliability, observability, data quality at scale. The business model often depends on customers who are themselves technically sophisticated, which means the sales cycle is different and the evaluation process requires genuine domain depth.
Three consequences for founders raising at seed:
- Technical diligence happens earlier. Funds that invest in AI infrastructure evaluate model architecture and engineering decisions in the first meeting, not after a term sheet. Founders who can’t defend their design choices in depth get passed over quickly.
- The “why not cloud” question matters. Every AI infrastructure pitch faces scrutiny from investors about why a hyperscaler doesn’t solve the same problem. Founders who have a specific, credible answer move faster.
- Developer adoption is a fundable signal. For infrastructure companies, open-source traction, GitHub stars, and developer community engagement are verifiable proxies for product quality that investors use before revenue exists.
Verified Fund Facts: Seed-Stage AI Infrastructure Investors
The table below reflects verified data from official fund pages and verified secondary sources as of 2026. The focus is on funds with documented first-check activity in AI infrastructure specifically, not just AI broadly.
| Fund | Current fund / AUM | Stage | Check size | AI infrastructure thesis evidence | Decision speed ★ |
|---|---|---|---|---|---|
| Air Street Capital | Fund III $232M (Mar 2026); largest solo GP fund in Europe | Pre-seed, seed | $500K to $5M first check; up to $10M | Exclusively AI-first; research-driven from day one; State of AI Report gives early research signal | ★★★★★ |
| Gradient | Fund V $220M (Mar 2026); $1.2B total AUM | Pre-seed, seed | Average $3M; range $100K to $10M | AI infrastructure and applications; portfolio includes Lambda (acquired by NVIDIA), Writer, Krea | ★★★★★ |
| Radical Ventures | Fund III $650M (Oct 2025) | Seed to Series B | $1M to $25M; sweet spot $5M | AI-specialist; deep ties to AI research community; portfolio includes Cohere, Waabi | ★★★★☆ |
| Sky9 Capital | $2B AUM; USD and RMB funds | Pre-seed to expansion | $500K to $50M+ | AI and blockchain-enabled infrastructure; global multi-stage; cross-border operating presence | ★★★★★ |
Air Street Capital is the clearest example of a fund built to be the first institutional investor in AI-first companies. Founded by Nathan Benaich, who co-authors the annual State of AI Report, the fund has unusual visibility into AI research community signal before it shows up in commercial metrics. Fund III closed at $232M in March 2026, described as the largest solo GP VC fund ever raised in Europe. The fund’s 2026 thesis positions the next chapter around deployment and diffusion: “who can make AI dependable, affordable, and embedded in the systems that matter.” That maps directly to infrastructure: inference optimization, reliability tooling, observability, and deployment infrastructure for production systems.
Gradient (formerly Gradient Ventures, spun out from Google in October 2025) closed Fund V at $220M in March 2026, bringing total AUM to $1.2B. The fund invests pre-seed to seed in AI applications and infrastructure. Gradient does not invest in foundation model companies. For AI infrastructure founders, Gradient’s portfolio evidence is strongest at the MLOps, inference, data infrastructure, and developer tools layers. Average check is $3M with 10-15% ownership target.
Radical Ventures is a Toronto-based AI-specialist fund that closed Fund III at $650M in October 2025. Investment stages span seed to Series B, with check sizes from $1M to $25M. The fund has deep ties to the AI research community, including connections to Geoffrey Hinton and Yoshua Bengio. Best for: founders with strong research credentials or ties to the Canadian or global AI research community.
What the Fastest-Moving AI Infrastructure Investors Have in Common
Across early-stage AI infrastructure investors that consistently close the best deals, four structural characteristics repeat.
Technical partners who can write the code. The funds that move fastest on AI infrastructure have engineers on the investment team who can independently evaluate model architecture, benchmark performance claims, and identify where a product is genuinely differentiated versus where it’s one update from being replicated. Air Street’s Nathan Benaich has been publishing AI research analysis since 2015. Gradient’s team includes engineers with AI infrastructure deployment experience.
A view on open-source adoption as a signal. AI infrastructure companies often grow through open-source adoption before commercial traction. Funds that understand how to evaluate GitHub contribution quality, research paper citations, and developer community engagement as investment signals can move earlier than funds that wait for revenue.
A specific thesis on where infrastructure moats form. The fastest-moving funds have already formed a view on which sub-categories of AI infrastructure produce durable businesses (inference optimization with proprietary compilation, evaluation infrastructure with proprietary benchmarks) versus which ones commoditize quickly (generic API wrappers, thin managed service layers).
Decision timelines built for technical founders. Technical founders don’t want to spend three months answering investor questions they already answered in the first meeting. Funds that run their diligence efficiently, committing to a decision timeline upfront, attract better AI infrastructure founders than funds with open-ended processes.
How Sky9 Capital Backs AI Infrastructure Companies
Sky9 Capital is a global multi-stage fund with $2B in AUM, backing AI infrastructure founders from pre-seed through expansion stage across San Francisco, Boston, Beijing, Shanghai, and Singapore. The firm’s dedicated strategy, Sky9 Digital, focuses on AI and blockchain-enabled financial infrastructure.
Technical thesis at the convergence layer
Sky9’s investment approach for AI infrastructure focuses on the specific intersection where AI model infrastructure and regulated financial systems meet. This is a narrower mandate than general AI infrastructure, but a more specific one: model serving infrastructure for compliance-sensitive environments, data infrastructure with financial-grade audit requirements, and AI agent infrastructure for enterprise financial workflows.
XtalPi, now listed on the Hong Kong Stock Exchange, is the most direct example of AI infrastructure investment in the portfolio: a team of computational physicists and AI researchers building quantum mechanics and machine learning infrastructure for pharmaceutical research. The founding thesis was a specific claim about what the AI-driven infrastructure layer could do in drug discovery that rule-based systems couldn’t. Sky9 evaluated that claim at the architectural level, not the market level.
Cross-border deployment as a structural advantage
AI infrastructure companies built for a single market face a ceiling as the competitive field globalizes. Sky9’s multi-geography presence means portfolio companies get access to enterprise deployment opportunities and technical talent pipelines in the markets where AI infrastructure adoption is fastest. For infrastructure companies that need enterprise pilots across US and Asian markets simultaneously, this matters more than fund brand.
Stage continuity for capital-intensive builds
AI infrastructure companies often require longer development cycles than application-layer startups. Sky9 invests from pre-seed through expansion stage, removing the financing pressure that forces AI infrastructure founders to raise again before the technical work is complete.
AI infrastructure founders at seed or pre-seed can reach out directly. The team reviews inbound from founders building AI infrastructure, particularly in categories that intersect with financial services, deep tech, and regulated enterprise deployments.
What AI Infrastructure Founders Should Verify Before Pitching
Before targeting any seed-stage AI infrastructure investor, three verification steps save significant time.
Check the portfolio for infrastructure evidence, not just AI. A fund whose AI portfolio consists primarily of vertical SaaS and consumer applications is not equipped to evaluate your inference optimization architecture, regardless of what the thesis page says. Find the specific infrastructure investments in the portfolio and verify they’re recent (within the last 18 months).
Confirm the fund is actively deploying from a current vehicle. AI infrastructure rounds that closed in 2021-2022 vintage funds may have moved the fund into follow-on mode. A fund that made 20 infrastructure investments in 2022 and has shifted to Series A management in 2026 is not a seed infrastructure investor for your purposes.
Match check size to round size. Air Street writes $500K to $5M at seed. Radical’s sweet spot is $5M. Gradient averages $3M. Sky9’s range starts at $500K. If you’re raising a $1.5M seed, the right first call is different than if you’re raising a $6M seed.
The Infrastructure Window Is Open. The Bar Has Moved.
Capital that flooded into AI infrastructure in 2024 has compressed valuations upward and raised the signal bar. Seed-stage AI infrastructure investors in 2026 are not taking bets on category exposure. They’re taking bets on specific technical claims that they can independently verify.
The founders who close fastest in this category are the ones who arrive with a specific architectural thesis, documented developer adoption or early enterprise usage, and a clear answer to why their infrastructure approach produces a moat that compounds over time. The investors who move fastest are the ones who already have a view on exactly those questions before the first meeting.