NetworkOcean

What do they actually do

NetworkOcean designs, builds, and operates self‑contained data center “capsules” that sit underwater and larger floating barges that can host multiple capsules. The company’s pitch is that seawater cooling can lower operating costs and avoid freshwater‑intensive evaporative cooling used on land NetworkOcean homepage and YC profile.

Today, they are marketing reserved cloud compute—specifically H100 GPU capacity—for customers to book. Their YC launch lists 2,048 H100s available to reserve with per‑GPU hourly pricing, and their Terms of Service describe standard cloud services (VMs, networking, storage) delivered on top of their infrastructure YC launch and ToS. A 1 MW capsule is being tested in San Francisco Bay, and reporting indicates local regulators scrutinized the planned underwater trial YC launch and WIRED. The current buyer flow is to contact the team or use the “reserve compute” link, place a reservation for GPUs/VMs, and get access as a cloud service; there are no public case studies or named customers yet NetworkOcean homepage and ToS.

Who are their target customer(s)

• AI model training teams at startups and research labs: They need large, predictable blocks of H100 time; public cloud GPU instances can be expensive, scarce, throttled, or preemptible. NetworkOcean offers reserved capacity as an alternative YC launch, ToS.
• ML/ops teams in mid‑to‑large companies running steady GPU workloads: They face variable cloud pricing and high cooling/operating costs. NetworkOcean positions seawater cooling and reservations to improve cost predictability NetworkOcean homepage, YC launch.
• Cloud resellers, colo operators, and GPU capacity brokers: They need to add coastal capacity quickly but face high capex and permitting timelines for new land builds. NetworkOcean’s modular capsules and barges aim to shorten deployment cycles NetworkOcean homepage.
• Sustainability and facilities teams with water‑use and energy targets: They must cut freshwater consumption and cooling energy amid regulatory pressure. Seawater cooling is the proposed benefit, though environmental review is a live constraint NetworkOcean homepage, WIRED.
• Media, gaming, and edge‑service providers near coastal population centers: They want dense compute close to users to reduce latency without paying high onshore land, power, and cooling costs. NetworkOcean markets coastal barges/capsules to address this NetworkOcean homepage.

How would they acquire their first 10, 50, and 100 customers

• First 10: Offer discounted, short paid pilots to 6–8 AI startups and research labs from YC/universities, with clear SLAs and hands‑on onboarding, using the SF Bay capsule test and telemetry as the technical proof point while disclosing regulatory status and risks YC launch, ToS, WIRED.
• First 50: Publish 2–3 case studies from pilots (cost, uptime, cooling metrics) and run targeted outbound to ML/ops teams and GPU brokers with 3–12 month reserved blocks and startup tiers; include environmental/regulatory documentation for procurement comfort NetworkOcean homepage, YC launch.
• First 100: Add channel partners (colo operators, cloud resellers, GPU brokers) with white‑label/reseller deals and standardized reservation contracts; simultaneously sell to mid/large enterprises with audited energy/water comparisons and volume discounts NetworkOcean homepage, WIRED.

What is the rough total addressable market

Top-down context:

AI data‑center spending is massive—hyperscaler AI capex alone is estimated around $527B in 2026—but most of that is internal and not directly addressable by new providers Goldman Sachs.

Bottom-up calculation:

The near‑term target is the third‑party GPU‑as‑a‑Service market, estimated at about $7.4B in 2026; capturing even 0.1–5% would translate to roughly single‑digit to high‑hundreds of millions in annual revenue Mordor Intelligence. Broader colocation spend in the tens of billions offers additional room but overlaps with GPU services Precedence Research.

Assumptions:

• NetworkOcean can secure permits and environmental approvals for some coastal deployments.
• Sustained demand for reserved H100‑class capacity at market‑comparable pricing persists through 2026+.
• Marine operations and network backhaul costs still allow a clear TCO advantage or unique placement/latency benefits vs. land‑based alternatives.

Who are some of their notable competitors

• CoreWeave: Specialized GPU cloud with large pools of H100/A100 capacity and reserved instances; competes directly for teams seeking fast access to high‑end GPUs.
• Lambda: GPU cloud and on‑prem solutions with reserved clusters for AI training; a common alternative for startups and labs needing H100/A100.
• OVHcloud: Established cloud provider offering GPU instances; offers a conventional land‑based option for price‑sensitive buyers.
• Equinix: Global leader in colocation and interconnection; customers can colocate their own GPU racks near coastal network hubs instead of using underwater infrastructure.
• Digital Realty: Major colocation provider with broad global footprint; competes for enterprises that prefer traditional facilities for GPU deployments.