Production is the thesis.
We back the engineers solving inference cost, latency, and GPU scheduling — the systems layer that determines which AI applications make it past the prototype.
$68M deployed across 12 infrastructure companies — Seed and Pre-Seed.The compute layer is not a commodity.
Every generation of AI applications runs on infrastructure assumptions that are wrong within 18 months. Inference cost per token, batching throughput, model routing overhead, cold-start latency — these are the constraints that actually determine whether a model ships to users or stays in a Jupyter notebook.
We invest when a technical insight about these problems is sharp enough to build a company around. Not a product vision. A systems insight — the kind you only form by having built the thing yourself.
12 companies building the production layer.
Former engineers. Not just capital.
Lukas built GPU scheduling layers at a hyperscale cloud provider. Sarah built the model serving platform at a European unicorn. After the term sheet closes, that experience stays available — at the architecture level, not just in quarterly board decks.
Architecture Office Hours
Sarah holds bi-weekly sessions with portfolio CTOs navigating inference scaling decisions. The agenda: batching strategy, hardware selection, KV cache sizing, serving architecture trade-offs. Founders bring real decisions; Sarah brings the context of having built production serving systems at scale. Not advisory theater — actual technical work.
GPU Capacity & Cloud Access
Lukas's network spans hyperscale cloud teams and GPU cloud operators built during six years shipping GPU scheduling infrastructure. Portfolio companies get introductions to capacity allocation contacts before those relationships are needed — relevant when your inference product depends on hardware that has a waitlist.
Enterprise Buyer Introductions
Niklas maps infrastructure buyers in financial services, healthcare, and manufacturing — sectors where AI inference is moving from pilot to procurement. Introductions go to platform engineering directors and CTO-level buyers who are actively building inference stacks, not generic LinkedIn outreach.
Follow-on Fundraising Preparation
Niklas starts fundraising preparation 9 months before portfolio companies need capital — financial model construction, investor targeting, and narrative positioning. Built with time to iterate, not assembled with two weeks of runway left.
Recruiting Network
The ETH Zurich ML engineering and distributed systems community runs through Firntal. We connect portfolio companies with senior ML engineers, infrastructure leads, and potential founding CPOs — from networks built as practitioners, not as investors.
LP-Backed Distribution Partnerships
Several Firntal LPs are corporate strategics with active AI infrastructure procurement programs. When portfolio companies reach distribution-readiness, we facilitate structured introductions — not warm emails into procurement black holes.