Standing watch since 2026

The Watch

News, new models, capability drops, and commentary on what is happening in maritime, AI, and the technology around them, filtered through a single question: does it matter if you lose the link? If the answer is yes, we cover it here. For long-form pillars, visit Field Notes.

  1. № 02Edge AI

    Nemotron 3 Nano and the Vessel AI Stack

    NVIDIA's open-weight Nemotron 3 Nano activates 3B of 31B parameters at inference. Here is what that means for on-vessel deployment.

    By Ethan Marsh·5 min read

  2. № 03Security & Compliance

    CISA's Three-Day Clock Hits the Fleet

    CISA BOD 26-04 requires three-day patching for critical vulnerabilities. At sea, that timeline is fiction without local vulnerability management.

    By Marcus Hale·5 min read

  3. № 04Infrastructure

    Rotterdam's AI Barge Needs No Cloud

    A 135-meter vessel just navigated Rotterdam's busiest waterways with zero human input. Every decision ran on local hardware.

    By James Calder·4 min read

  4. № 05Resilience

    When the Ocean Data Feeds Go Dark

    NSF is dismantling 900 deep-sea monitoring instruments. Vessels that keep their own data on local hardware will not notice.

    By James Calder·4 min read

  5. № 06Networking

    SpaceX IPO Changes the Starlink Math

    SpaceX is now a public company answering to Wall Street. For vessel operators who depend on Starlink, the pricing calculus just shifted.

    By James Calder·4 min read

  6. № 07Strategy

    Anthropic Says Pause. Plan Accordingly.

    Anthropic called for a global AI development pause. For vessel operators running cloud-dependent AI, this is a version control problem.

    By James Calder·4 min read

  7. № 08Infrastructure

    JetPack 7.2: The Vessel Agent Stack Arrives

    NVIDIA's JetPack 7.2 lands NemoClaw on Jetson Thor with multi-instance GPU. Here is what that means for running AI agents on a vessel.

    By Ethan Marsh·5 min read

  8. № 09Edge AI

    NVIDIA AI Grid and the Vessel Math

    NVIDIA's AI Grid cuts edge inference costs by 76 percent. The math for sovereign vessel AI just got a lot harder to argue against.

    By James Calder·4 min read

  9. № 10Security & Compliance

    What 10,000 Zero-Days Mean for Fleets

    Anthropic expanded Project Glasswing to 150 organizations. Over 10,000 zero-days confirmed. The math on vessel cybersecurity just shifted.

    By Marcus Hale·5 min read

  10. № 11Industry Insights

    Posidonia Has 30 AI Demos. One Question.

    Posidonia 2026 has 30+ AI exhibitors. Most of their demos need a satellite link. That tells you everything about their architecture.

    By James Calder·4 min read

  11. № 12Networking

    Starlink V3: Faster Pipe, Same Problem

    SpaceX's V3 satellites promise 1 Tbps per bird. For vessel AI, the throughput was never the bottleneck. The architecture stays local.

    By Ethan Marsh·5 min read

  12. № 13Security & Compliance

    USCG Cyber Rule Meets Vessel AI

    The USCG cybersecurity rule hits its July milestone in six weeks. The IT/OT segmentation mandate validates sovereign vessel AI architecture.

    By Ethan Marsh·4 min read

  13. № 14Industry Insights

    Cruise AI Fails on Data, Not Models

    Seatrade says cruise AI fails because the data layer is broken. They are right. But fixing it means fixing where the data lives, not just how it connects.

    By Marcus Hale·4 min read

  14. № 15Security & Compliance

    npm Worms and the Case for Air Gaps

    A worm compromised 633 npm packages in an hour. On-vessel software needs air-gapped dependency pipelines, not live registry pulls.

    By Ethan Marsh·5 min read

  15. № 16Edge AI

    MSC's AI Concierge: One Missing Answer

    MSC's AI concierge processed a million guest messages at 93% satisfaction. Nobody asked where those conversations live or what happens offline.

    By Marcus Hale·5 min read

  16. № 17Security & Compliance

    When Cyber Risk Meets Voyage Planning

    ABB and Cydome are merging cyber risk and voyage optimization into one platform. The real question is where that platform runs.

    By Marcus Hale·4 min read

  17. № 18Security & Compliance

    AI Found the OT Gateway First

    Dragos found attackers using Claude to identify OT targets inside a water utility. Vessel operators should assume the same capability applies at sea.

    By Marcus Hale·5 min read

  18. № 19Edge AI

    Why MSC's Robot Dogs Need Local AI

    MSC Cruises is putting AI robot dogs and digital avatars on its ships. Every one of those systems proves why compute has to live on the vessel.

    By James Calder·4 min read

  19. № 20Strategy

    The MASS Code Endorses Sovereign AI

    The IMO's autonomous vessel code requires ships to operate safely without shore connections. That is a regulatory endorsement of AI that lives on board.

    By James Calder·4 min read

  20. № 21Strategy

    Hormuz Convoys: Cloud AI Goes Dark First

    The US military is escorting stranded vessels through the Strait of Hormuz. Cloud-dependent AI stops working the moment a ship joins the convoy.

    By James Calder·4 min read

  21. № 22Strategy

    Royal Caribbean's AI Blind Spot

    Royal Caribbean told Wall Street that AI drives its $4.5 billion revenue engine. The question nobody asked: what happens when the satellite link drops?

    By James Calder·4 min read

  22. № 23Security & Compliance

    AI Gateways Are a Single Point of Failure

    A critical LiteLLM SQL injection hands attackers every API key the proxy stores. Vessel operators relying on cloud AI gateways take note.

    By Marcus Hale·5 min read

  23. № 24Security & Compliance

    APT28 Targets Maritime Supply Chains

    Russia's APT28 is deploying PRISMEX malware against NATO maritime logistics. Here is what vessel operators need to know about this active campaign.

    By Marcus Hale·5 min read

  24. № 25Security & Compliance

    Three CVEs in Your Vessel Data Gateway

    Cydome found authentication and path-traversal flaws in the NAVTOR NavBox. The fix matters less than what the bugs reveal about maritime OT security.

    By Marcus Hale·5 min read

  25. № 26Resilience

    ChatGPT Went Down for 90 Minutes

    ChatGPT went down for 90 minutes on April 20. For desk workers, that is an inconvenience. For a vessel at sea, it is an architecture problem.

    By James Calder·4 min read

  26. № 27Edge AI

    Kimi K2.6: 1T Parameters, 8 H200s Required

    Moonshot AI's Kimi K2.6 ships natively in INT4 at 1T parameters. Open weights, world-class agent benchmarks, and a hardware bill that starts at 8 H200s.

    By Ethan Marsh·6 min read

  27. № 28Security & Compliance

    When Ransomware Reaches the Engine Room

    Maritime cyber incidents doubled in 2025, with ransomware now targeting engine and ballast systems. Local-first AI reduces the blast radius.

    By Ethan Marsh·5 min read

  28. № 29Resilience

    What the Starlink Outage Actually Proved

    A Starlink outage grounded Pentagon drone tests this week. The same single-provider risk threatens every vessel running cloud AI.

    By James Calder·4 min read

  29. № 30Edge AI

    First Yacht OEM Ships Local-First AI

    Next Yacht Group is shipping on-vessel AI as a standard feature across its entire range. Here is why that changes the conversation for every yacht owner.

    By James Calder·4 min read

  30. № 31Strategy

    1,500 Cloud Agents, One Satellite Link

    Virgin Voyages grew to 1,500 cloud AI agents in four months. The results are real. So is the connectivity risk nobody is discussing.

    By James Calder·4 min read

  31. № 32Security & Compliance

    Mythos Zero-Days and Fleet Security

    Anthropic's Mythos found thousands of zero-days in partner infrastructure. Here is what that means for vessel fleet cybersecurity.

    By Marcus Hale·6 min read

  32. № 33Edge AI

    KV Cache Quantization at Sea (3/3)

    Google's TurboQuant compresses the KV cache to 3 bits with no accuracy loss. Here is what that means for a vessel GPU rack.

    By Ethan Marsh·17 min read

  33. № 34Resilience

    Gigabit Starlink Won't Fix Vessel AI

    SpaceX is talking gigabit speeds for Starlink in 2026. Here is why throughput is the wrong thing to focus on for vessel AI.

    By James Calder·5 min read

  34. № 35Infrastructure

    1M Context: The GPU Math (2/3)

    VRAM and KV cache math for models from 7B to 405B at 1M tokens. Per-user scaling for crew and guests. The numbers are ugly.

    By James Calder·14 min read

  35. № 36Edge AI

    Context vs. RAG on Vessels (1/3)

    Vercel ran the eval: a compressed doc file beat on-demand retrieval 100% to 53%. What that means for AI deployments on vessels.

    By Ethan Marsh·10 min read

Subscribe

Stand the watch with us

Short commentary delivered when news breaks, not on a schedule. Sovereign AI, maritime tech, and what still works when the link drops.

No spam. Unsubscribe any time.

Pillar essays live in Field Notes; commentary and reactions live here. Subscribers get both in their inbox when they ship.