Offline is the normal case
A round trip to a datacenter is not always on the table.
Pre-seed · investors and cloud partners
The release and policy layer for every device fleet that ships AI.
Models are now good enough to run on industrial silicon. What does not exist is the layer that lets a fleet operator ship, update, audit, and roll back those models like firmware — across NXP, NVIDIA, and Android hardware, with the cloud used only by policy. ByteHub AI is building that layer. This round buys the design-partner proof that turns a technical advantage into a category position.
01 · The problem
AI has reached the device. Production operations have not caught up.
Buyer: industrial OEMs, equipment makers, and fleet operators running Linux-based devices in the field. Today they assemble local inference, cloud APIs, scripts, and OTA tooling themselves — until a failed update, data-policy block, or field incident makes AI unusable in production.
The data cannot leave
Cloud-only AI is blocked in procurement before a pilot starts.
Updates need proof, not prose
Signed artifacts, policy checks, staged rollout, observability, rollback. EU CRA raises the bar for covered products.
02 · Why now
Three reasons.
01
Edge models are now useful
Quantized vision and language models run on industrial edge hardware, enabling real-time and offline workflows.
02
Frontier models create a hybrid necessity
Devices need local autonomy but still benefit from cloud reasoning. The missing layer is policy-controlled routing and auditable context boundaries.
03
Model releases are becoming production risk
As AI reaches regulated and operationally critical products, model artifacts need the same release discipline as firmware.
| Category | What it provides | What it does not provide |
|---|---|---|
| Cloud AI platforms | Model APIs, tools, cloud-scale reasoning | Device-specific offline runtime and model release lifecycle |
| Silicon / edge SDKs | Accelerated inference and hardware tooling | Cross-fleet model governance and release control |
| OTA / device-management tools | OS / device updates and telemetry | Model-aware evaluation, artifact policy, AI routing |
| AI / agent wrappers | Chat and workflow automation | Device safety boundary, signed edge artifacts, rollback |
ByteHub is the model lifecycle and policy layer between AI capability and real-world device operations.
Market direction
The deployment layer is being bought. The release layer is unowned.
In 2026 the market told us where it is going. Silicon vendors acquired edge-deployment tooling for their own chips. Vendor runtimes started shipping OpenAI-compatible LLM servers on the device. Industrial operators put small language models on HMIs and air-gapped appliances. Every one of those moves creates more model artifacts on more devices — and none of them ships the cross-silicon release, policy, and audit layer a fleet operator needs.
Consolidation signal
Silicon vendors buy deployment, per chip
Deployment tooling is being pulled into single-silicon stacks. Industrial fleets are not single-silicon. The buyer needs one artifact and one policy across i.MX, Jetson, Android, and x86 — which no chip vendor will build for its competitors' parts.
Capability signal
Small models are production-usable on the device
1–9B models at INT4 run with usable quality on NPUs and edge GPUs; vendor runtimes make local inference routine. The bottleneck has moved from "can it run" to "can we ship, govern, and roll it back on 1,000 devices".
Regulatory signal
Model releases become compliance events
As AI reaches regulated, operationally critical products (EU Cyber Resilience Act for covered products), model artifacts need traceable, signed, recoverable releases — the discipline firmware already has, applied to models.
Our position
Cross-silicon, provider-aware, evidence-first. ByteHub sits above the silicon SDKs and below the cloud platforms, owning the boundary both need but neither will build for the other: the signed artifact, the execution profile, the policy gate, and the audit trail. Per-device runtime licences and a per-tenant control plane compound with fleet size; usage-metered cloud intelligence rides on the same gate.
03 · What we build
Small work stays on the chip. Big work goes through a gate.
01
EdgeRuntime
Quantized local inference for low latency, privacy, offline operation, device-specific execution.
02
ModelOps control plane
Build, evaluate, sign, deploy, observe, roll back model artifacts like firmware.
03
Policy-gated cloud intelligence
Escalate only approved, minimized context to a frontier model.
04
Business sandbox
Tenant-isolated context, tools, and audit boundaries per deployment.
04 · Initial beachhead
Land with fleet operations. Expand across the fleet.
Primary: private on-prem inspection AI for industrial operators and equipment makers running vision-language inspection within the plant.
Durable layer: firmware and fleet-operations copilot — grounded diagnosis across Yocto/BitBake logs, RAUC or SWUpdate outcomes, device telemetry, and runbooks, released with firmware discipline.
| Segment | Buyer | Budget driver |
|---|---|---|
| Industrial equipment OEMs | VP Engineering / Head of Platform | Faster, safer release lifecycle |
| Fleet operators | Director of device operations / reliability | Lower incident response and field-support burden |
| Robotics / inspection integrators | CTO / Product engineering | Private, offline AI without building the full lifecycle stack |
05 · Market and business model
A software-first control plane for the installed base of Linux edge devices.
Initial ICP hypothesis: 200 target OEMs / operators × 1,000 managed devices × annual runtime + platform value. To be validated in customer discovery; not a market-size claim.
Edge runtime license
Per-device or per-gateway annual subscription: local runtime, compatibility profiles, policy enforcement, updates.
Enterprise control plane
Per-tenant subscription: artifact registry, release management, audit, fleet observability, governance.
Usage + services
Metered cloud intelligence and paid integration / pilot services designed to become reusable product modules.
06 · Why us · what is real today
Engineering-led. Evidence before claims.
Most edge AI pitches lead with tokens per second. We lead with what happens when a release fails on a device in the field — because that is where fleets stop trusting AI. Our proof standard: measured on our own bench, then validated on a customer workflow, then published. Every number on this site will be earned that way.
In hand / underway
- i.MX95-class, Jetson Orin, and Jetson Thor hardware on the bench
- Quantized inference validation across the three targets
- Embedded Linux, Yocto, BSP, OTA, RAUC / SWUpdate foundation
- EdgeRuntime, signed artifact, policy-gate, and rollback architecture defined
- NVIDIA, Google Cloud, AWS, and Microsoft startup-program applications / ecosystem engagement
| Next milestone | Evidence |
|---|---|
| Local runtime | Measured latency, memory, power, and target compatibility |
| Model lifecycle | Signed artifact, staged deployment, device verification, rollback |
| Hybrid plane | Auditable policy-authorized cloud escalation with minimized context |
| Customer proof | 2–3 design-partner evaluations; one paid or committed pilot |
07 · Team and unfair advantage
Built by an engineer who has operated the layers AI products usually ignore.
Jerry — Founder / CEO. Embedded systems engineer with hands-on experience in NXP i.MX95 bring-up, Yocto, Linux kernel, U-Boot, and BSP development; RAUC firmware updates, partitioning, secure release workflows, and device debugging; DDS, messaging, fleet / device communications, and distributed systems; edge AI integration, model deployment, and hardware-aware optimization; embedded systems work in energy-storage and real-world device environments. The signed-artifact, staged-rollout, and rollback discipline ByteHub applies to models is the discipline he has shipped for firmware.
The founding team this round assembles
Three mandates, each tied to a milestone the market can check.
01 · Runtime
Head of Edge Runtime
Mandate: ship the signed EdgeRuntime across i.MX95, Orin, Thor, and Android with measured execution profiles — the compatibility matrix competitors cannot publish. Owns quantization evidence and per-SoC performance under thermal load.
02 · Control plane
Head of Control Plane & Cloud
Mandate: make the ModelOps control plane the system of record for models on devices — artifact registry, staged rollout, audit — and stand up the provider-aware policy gate to Vertex, Bedrock, and Foundry with per-tenant isolation.
03 · Customers
Design-Partner Engineering Lead
Mandate: convert 2–3 industrial design partners into a repeatable deployment package and the first paid pilot; turn every pilot integration into a reusable product module, never recurring custom work.
Senior, hands-on hires with production embedded-Linux or ML-systems records; recruited through the silicon and cloud ecosystems we already work in. Sequenced to the 18-month plan below so each hire lands against a milestone, not ahead of one.
08 · 18-month plan
Land with one workflow, prove repeatability, expand across the fleet.
| Time | Product milestone | Commercial milestone |
|---|---|---|
| 0–3 months | EdgeRuntime demo; signed artifact; local verification | 25+ customer discovery interviews; ICP selected |
| 3–6 months | Staged deployment, rollback, policy gate, benchmark pack | 2–3 design-partner LOIs / evaluation agreements |
| 6–12 months | Cloud escalation, audit trail, productized control plane | First paid or committed pilot |
| 12–18 months | Repeatable deployment package | Seed-ready evidence: pilot outcomes, pipeline, early recurring revenue |
09 · Raise and use of funds
Pre-seed SAFE to reach design-partner proof.
The goal of this round is not broad platform scale. It is proof that ByteHub becomes the repeatable operating layer for governed edge AI fleets — the evidence a priced seed round is built on.
$1.0Mtarget raise
18 motarget runway
$400Ktarget initial close
$12–15Millustrative post-money cap
Final financing terms subject to investor discussion and legal review.
| Use | Target allocation |
|---|---|
| Product and engineering | 45% |
| Customer pilots and GTM | 15% |
| Hardware and lab | 12% |
| Legal, finance, and G&A | 10% |
| Cloud, data, and infrastructure | 10% |
| Contingency | 8% |
This capital delivers
- Signed, rollback-capable EdgeRuntime on selected hardware
- Policy-controlled cloud escalation and audit trail
- 2–3 industrial Linux-device design partners
- One paid or committed pilot
- Measured technical and commercial evidence for a priced seed round
10 · Cloud and silicon partners
Governed frontier intelligence and device operations for edge fleets.
ByteHub remains provider-aware for customer deployment, residency, and procurement requirements. Names below are the platforms we build on and are engaging through their startup programs — not endorsements.
| Partner | What we build on | What the partner gains | What we ask |
|---|---|---|---|
| Google Cloud | Vertex AI / Gemini for policy-authorized long-context reasoning, multimodal analysis, technical retrieval, evaluation | A differentiated edge-native workload that drives governed Gemini / Vertex usage from real device fleets; reference path for hybrid edge + cloud in industrial Linux environments | AI-first startup credits, Vertex AI technical guidance, startup ecosystem |
| NVIDIA | JetPack, CUDA, TensorRT / TensorRT-LLM / Edge-LLM; JetPack OTA extended with signed model artifacts; Orin and Thor on the bench | Jetson becomes a deployable, auditable AI product platform | Technical feedback on execution profiles, packaging, OTA / rollback; preferred Jetson pricing; DLI; Inception introductions to OEMs and integrators |
| AWS | Amazon Bedrock and Guardrails; AWS IoT identity and connectivity | Governed Bedrock and AWS IoT usage from edge fleets | Activate credits, technical guidance, partner benefits |
| Microsoft | Microsoft Foundry (Azure-billed models), Entra ID, Key Vault, Azure Monitor; optional Arc / IoT Operations | Governed Foundry and Azure usage; enterprise hybrid reference | Founders Hub credits (enrolled), Marketplace and co-sell path |
Next step
Request the deck, or start a design-partner conversation.
We share the full deck, the technical brief, and the evaluation format under a short conversation. No customer or NDA material is published here.