AWS keeps the GPU + AMI Marketplace · Cloudflare adds the global edge for OTA + telemetry

Deep Connected Platform runs on AWS.
The global vehicle edge runs even better on Cloudflare.

The AWS blog from Oct 29, 2025 lays out exactly how Sibros + Panasonic + AWS reduce connected-vehicle integration friction. AWS hosts the vSkipGen GPU emulator on EC2 G5G and ships Deep Connected Platform via AWS Marketplace — both of which Cloudflare doesn't try to replace. What Cloudflare adds is the layer in front: OTA binary delivery at zero egress, vehicle telemetry triage at 330+ POPs, AI Gateway in front of Deep Insights, and Workers for Platforms for per-OEM tenancy — on the same edge that already serves sibros.tech today.

NS: carrera + merlin.ns.cloudflare.com · server: cloudflare on www · api.sibros.tech on AWS us-west-2 ELB · app.sibros.tech on CloudFront · Anthropic + OpenAI + Cursor + Lovable all verified on apex TXT

AWS for Industries Blog · Oct 29, 2025
"Sibros and Panasonic Automotive have collaborated to help automakers reduce connected service integration problems ahead of critical vehicle timelines. This new workflow consists of using Panasonic vSkipGen™ as a virtual replica of both the infotainment software and the vehicle's integrated cluster. Sibros has installed its Deep Logger on the vSkipGen environment and is directly integrated into the virtual CAN bus."
— Authors: Nick Weber (Sibros), Jerry Bonnah (AWS), Senthilnathan Subramanian (Panasonic), Stefano Marzani (AWS, WW Head of SDX), Vladimir Beloborodov (Panasonic). Read the full blog →

What's already running on Cloudflare today

DNS
sibros.tech on Cloudflare via carrera + merlin.ns.cloudflare.com
WEB EDGE
www.sibros.tech served via Cloudflare anycast (104.26.x / 172.67.x) with server: cloudflare
EXPANSION PATH
Same MSA, same SOC mapping — add R2 + Workers for Platforms + AI Gateway in front of the AWS stack
4
Deep products: Logger, Updater, Commander, Insights
CAN bus
VirtIO integration into virtual + physical bus
AAOS 14
Android Automotive support via Panasonic vSkipGen
2
AWS Marketplace listings: vSkipGen + DCP SaaS
Joint go-to-market partners (from the AWS blog)
Sibros· Panasonic Automotive· AWS Automotive Innovation Lab· NVIDIA GPUs (via EC2 G5G)· Android Automotive Operating System

A three-party architecture — with one role still un-filled.

The AWS blog post is admirably clear about who does what. Sibros writes the connected-vehicle software. Panasonic ships the digital cockpit. AWS provides the GPU instance and the Marketplace. The thing missing from the diagram — not yet a problem at demo scale, very much a problem at fleet scale — is the global edge layer that handles OTA delivery and telemetry triage. That's where Cloudflare slots in.

Sibros builds

The Deep Connected Platform & embedded firmware

Deep Logger for CAN-bus capture + analytics, Deep Updater for OTA, Deep Commander for remote diagnostics, Deep Insights for AI-powered root-cause analysis. Embedded firmware that integrates with any automotive stack.

  • Deep Logger (real-time fleet telemetry)
  • Deep Updater (multi-ECU OTA)
  • Deep Commander (remote feature activation)
  • Deep Insights (Anthropic + OpenAI verified)

AWS provides

GPU compute, AMI Marketplace, the lab

EC2 G5G with NVIDIA GPUs for vSkipGen virtual cockpit, AWS Marketplace as the OEM distribution channel, the Automotive Innovation Lab in Santa Clara, and generic cloud compute for the Sibros DCP SaaS backend.

  • EC2 G5G (NVIDIA GPU) — vSkipGen host
  • AWS Marketplace (vSkipGen + DCP)
  • AWS Automotive Innovation Lab
  • Generic SaaS hosting for DCP backend

Cloudflare adds

The global vehicle edge, OTA + telemetry + tenancy

R2 for OTA binary distribution at zero egress. Workers + Queues for telemetry triage at the POP closest to the vehicle. AI Gateway in front of Deep Insights' LLM calls. Workers for Platforms for per-OEM tenancy. All on the same edge already serving sibros.tech.

  • R2 + Workers for OTA at zero egress
  • Workers + Queues for global telemetry ingest
  • AI Gateway in front of Deep Insights
  • Workers for Platforms — per-OEM tenancy

Nine primitives, mapped to the AWS blog's actual architecture.

Each maps to a specific component in the AWS architecture diagram from the blog. Status tags show what's already on Cloudflare, what's an obvious next addition, and what should stay on AWS as the GPU-bound parts of the system.

PRIMITIVE 01 Live on CF

DNS + web edge for sibros.tech

Authoritative DNS via carrera + merlin.ns.cloudflare.com. The web edge serves with server: cloudflare. Procurement is in place, SOC mapping exists — expansion is a configuration change, not a vendor selection.

DNS Web edge Foundation
PRIMITIVE 02 Highest-leverage next

R2 for OTA binary delivery

Every connected vehicle pulls firmware updates — sometimes hundreds of MB per ECU per update, multiplied across millions of vehicles. R2's zero egress fees vs. S3 + CloudFront is the single biggest cost line in the entire Deep Updater workflow at fleet scale.

R2 Zero egress Deep Updater
PRIMITIVE 03 Telemetry triage

Workers + Queues for telemetry ingest

Deep Logger captures CAN signals from millions of vehicles. Workers at 330+ POPs catch the inbound telemetry at the POP closest to the vehicle — not centralized us-west-2 backhaul. Queues handle fan-out into ML pipelines without standing up SQS regionally.

Workers Queues Deep Logger
PRIMITIVE 04 AI Gateway wedge

AI Gateway for Deep Insights

Deep Insights is described as "AI-Powered analytics, including root cause diagnosis." Anthropic + OpenAI are both verified on your apex TXT records. AI Gateway sits in front: semantic cache for repeated root-cause queries, per-OEM attribution, budget caps, BYO keys for enterprise OEM customers.

AI Gateway Anthropic OpenAI Deep Insights
PRIMITIVE 05 Per-OEM tenancy

Workers for Platforms for OEM isolation

Sibros sells to many automakers via Panasonic, AWS Marketplace, and direct. Each OEM customer has its own SLA, its own data-residency requirements, its own audit cadence, its own brand surface. Workers for Platforms gives each OEM its own isolated Worker namespace inside one control plane.

Workers for Platforms Per-OEM Isolation
PRIMITIVE 06 Per-vehicle state

Durable Objects for per-vehicle session state

During an OTA push, each vehicle is a stateful session: which ECUs are updating, where the update is paused, whether cellular dropped mid-flight. Durable Objects give you a single-writer state holder per VIN at the edge with native WebSocket support — no per-region Redis cluster.

Durable Objects WebSockets Per-VIN
PRIMITIVE 07 CAN signal retrieval

Vectorize for CAN-signal scenario search

Deep Insights does "root cause diagnosis." That's a scenario-similarity problem: "find every CAN-bus event in the last 30 days that looks like this thermal anomaly." Vectorize indexes telemetry embeddings; the answer returns in single-digit milliseconds, not minutes.

Vectorize Scenario search Thermal Guardian
PRIMITIVE 08 Engineer access

Zero Trust for vehicle data plane access

Engineers, OEM partners, and validators all need access to per-OEM dashboards and telemetry pipelines. Cloudflare Access + Tunnel give identity-aware, audit-logged access without VPN sprawl — critical for the kind of SOC 2 / ISO 27001 / TISAX posture automotive customers expect.

Access Tunnel TISAX-ready
PRIMITIVE 09 Stays on AWS

AWS keeps the GPU + AMI Marketplace

The vSkipGen Android Automotive emulator legitimately needs an NVIDIA GPU instance (EC2 G5G), which Cloudflare doesn't provide. The AWS Marketplace AMI distribution is also AWS-native and stays there. This isn't a replacement pitch — the GPU-bound parts of the stack belong on AWS.

EC2 G5G AWS Marketplace Keep as-is

The architecture in the AWS blog, re-drawn with Cloudflare at the edge.

The diagram in the AWS blog shows vSkipGen running on AWS, the Sibros DCP SaaS running on AWS, and the in-vehicle Deep Logger talking to the cloud. The blog doesn't show a global edge in front — not because it isn't needed, but because at single-vehicle demo scale you can ignore it. At fleet scale, you can't.

The OTA + telemetry path, sketched with Cloudflare added

From "vehicle initiates OTA check" to "binary delivered + telemetry returned" — AWS hosts the control plane, Cloudflare carries the bits.
VEHICLE
Deep Updater + Deep Logger
vSkipGen or physical ECU
CF EDGE
Workers + R2 + Queues
closest POP, zero egress, triage
DCP SAAS
Deep Connected Platform
AWS us-west-2 ELB origin
AI ANALYTICS
AI Gateway + Vectorize
Deep Insights inference + RAG
What this changes: The OTA binary never leaves Cloudflare's network — R2 stores it, Workers serve it from the POP closest to the vehicle. Telemetry triage happens at the edge before anything backhaul-traffics to AWS, which means the DCP SaaS only sees the events it needs to see. Deep Insights' LLM calls go through AI Gateway with full per-OEM attribution. And the AWS side — the EC2 G5G vSkipGen and the DCP control plane — doesn't change at all.

The economics of connected vehicles at fleet scale.

The AWS-only architecture is the right starting point for demos and pilot OEMs. The architecture for 10M+ vehicles needs an edge layer that doesn't charge per-byte for OTA delivery, doesn't centralize telemetry through one region, and gives per-OEM cost attribution as a default not a future-roadmap feature.

A back-of-the-envelope, not a quote
Modeled across OTA delivery + telemetry ingest + Deep Insights inference at one-OEM-million-vehicle scale
OTA EGRESS SAVINGS
40–60%
R2's zero egress vs. S3 + CloudFront pricing across firmware + map updates delivered to millions of vehicles globally — the dominant cost line at fleet scale.
TELEMETRY TRIAGE AT EDGE
70–90%
Workers triage telemetry at the POP closest to the vehicle — only interesting CAN events backhaul to DCP, not raw firehose. Backhaul + storage cost falls accordingly.
PER-OEM ATTRIBUTION
100%
AI Gateway breaks Deep Insights spend down per-OEM, per-vehicle, per-feature — the data needed to defensibly price OEM contracts at the right tier instead of cost-plus.
The real win is per-OEM contract pricing defensibility. Today the Sibros AWS bill goes up as a customer's fleet grows. Tomorrow with Cloudflare in front, the marginal cost per vehicle drops sharply (zero egress on OTA), and per-OEM cost attribution becomes a default dashboard. When a new automaker asks for a 5-year contract for 2M vehicles, the answer to "what's our cost per vehicle?" is a chart, not an estimate.

Multiple OEM partners, one platform. Workers for Platforms is the boundary.

Sibros sells to automakers directly, to tier-1 suppliers like Panasonic, and to OEM joint ventures. Each customer has its own SLA, its own audit cadence, its own data residency requirements (EU GDPR, China, etc.), its own brand surface. That's not a feature flag — it's an isolation boundary.

Per-OEM tenancy, sketched

Each OEM customer gets its own Worker namespace inside Workers for Platforms. Same edge, same observability, region-bound data residency, isolated AI budget per Deep Insights workload.
🚚
Panasonic (tier-1)
🚚
OEM A
🚚
OEM B
🚚
OEM C
🚚
+ N more
Shared control plane — Workers for Platforms + AI Gateway + Vectorize + R2
one runtime · one observability surface · region-bound data residency enforced by binding, not by checkbox

Current stack, with Cloudflare overlaid on AWS.

Every row is sourced from public DNS records, the sibros.tech apex TXT, HTTP response headers, and the AWS blog post architecture diagram. Lime rows are already on Cloudflare. Orange-tinted rows belong on AWS and should stay there. The orange overlay column is the additive Cloudflare layer.

What's running today, and where Cloudflare slots in

Lime rows = already on Cloudflare. Amber-tinted rows = AWS-native and should stay there. Orange column = the additive Cloudflare overlay.
LAYER
SIBROS RUNS TODAY
CLOUDFLARE FIT
DNS
Cloudflare (carrera + merlin.ns.cloudflare.com)
✅ Live — the foundation everything else snaps onto
WEB EDGE
Cloudflare (104.26.x / 172.67.x, server: cloudflare)
✅ Live — expand to api + ota subdomains next
API PLANE
AWS us-west-2 ELB (api.sibros.tech, a0cbb584…)
+ Workers in front: rate-limit, per-OEM auth, edge cache
CUSTOMER DASHBOARD
CloudFront (app.sibros.tech, dqy8chis22to8…)
+ Cloudflare CDN + WAF in front for unified observability
VSKIPGEN EMULATOR
EC2 G5G (NVIDIA GPU) via Panasonic AMI on AWS Marketplace
Stays on AWS — Cloudflare doesn't offer GPU VM instances
DCP SAAS BACKEND
AWS-hosted, sold via AWS Marketplace
Stays on AWS for marketplace continuity; CF goes in front
OTA DELIVERY
Implied S3 + CloudFront for binaries to vehicles
+ R2 + Workers + Smart Placement — zero egress at fleet scale
TELEMETRY INGEST
Likely Kinesis / direct-to-DCP from Deep Logger
+ Workers + Queues at edge POPs for triage before backhaul
AI / DEEP INSIGHTS
Anthropic + OpenAI (both verified on apex), Cursor + Lovable in dev
+ AI Gateway in front: cache, attribution, BYO keys per OEM
PER-OEM TENANCY
Multi-tenant DCP with config-level isolation
+ Workers for Platforms — per-OEM namespace by construction
OBSERVABILITY
Dynatrace (verified ×2 on apex TXT)
No change — Dynatrace keeps watching; CF adds edge metrics
EMAIL / DEV TOOLS
Google Workspace + Microsoft 365 + Atlassian + Apple
+ Zero Trust SSO in front of admin consoles

The team behind your build tool just joined Cloudflare.

sibros.tech ships Vite in production. We can see it on the wire today: five /assets/[name]-[8-char-hash].js chunks (react-vendor, i18n-vendor, supabase-vendor, vimeo-vendor, plus the main bundle), modulepreload hints for each, and the __vite__mapDeps runtime symbol baked into the main chunk. That is the canonical Vite output signature.

On June 4, 2026, Cloudflare acquired VoidZero. The full team behind Vite, Vitest, Rolldown, Oxc, and Vite+ — Evan You and the maintainers — are now Cloudflare engineers. Vite stays open source and vendor-agnostic. The change is that the people building the toolchain Sibros engineers use every day now sit on the same team that operates the front door of sibros.tech.

What this opens up for Sibros. Direct line into the Vite roadmap on questions that matter to the Deep Connected Platform team: Rolldown-powered production builds (no Rollup/esbuild split), Oxc for type-checking and linting at native speed, and tighter integration between Vite's dev-server semantics and the same Workers runtime that already fronts sibros.tech. The combination of VoidZero engineers + Workers + R2 + AI Gateway on one platform is not available anywhere else.

The conversation-starter framing. “The bundler running sibros.tech right now is built by people on Cloudflare's team” is the opener. The substance underneath it is what is interesting: a single platform spanning the build tool, the edge runtime, the OTA delivery layer, and the AI inference path. That is the version of the developer-platform story that did not exist 60 days ago.

Why this is the right quarter to start the conversation

The AWS blog made the architecture public. Sibros + Panasonic + AWS just published a joint go-to-market piece on Oct 29, 2025 — with named co-authors from all three companies. The diagram is now public, the marketplace listings are live, and the demo runs at the AWS Automotive Innovation Lab in Santa Clara. The architecture conversation is open, which is the right moment to talk about what else fits in the picture.

You're already on Cloudflare. sibros.tech DNS is on us. The web edge serves through us. Anthropic, OpenAI, Cursor, and Lovable are all verified on the apex — you're an AI-tooling-heavy team. Expanding from DNS + web edge to R2 + Workers for Platforms + AI Gateway is the most natural roadmap conversation in the lineup.

VoidZero is now Cloudflare. The Vite/Vitest/Rolldown/Oxc team joined on June 4. Your production build is shipped by their toolchain today. Direct roadmap access — on the same platform that already fronts your front door — is a delta that did not exist last quarter.

The economics flip at fleet scale. At 100 demo vehicles, the AWS-only architecture is the cheapest, simplest path. At 1M+ vehicles, OTA egress is the single biggest cost line in the bill. R2's zero-egress profile is the most valuable thing Cloudflare can offer the Deep Updater roadmap — and the cheapest hour you can spend in front of that curve, before the math starts to matter at OEM-contract-pricing time.

Worth a 30-minute conversation with the connected-platform team?

The interesting conversation is which of these primitives is closest to your current sprint: R2 for OTA at zero egress, AI Gateway in front of Deep Insights, Workers for Platforms for per-OEM tenancy, or Workers + Queues for global telemetry triage. I'd rather hear what's actually on your roadmap than guess.

Matt Holscher Calendar  → Reply by email