Shipyard AI Valuation Isn't About Order Numbers

I spent two days trying to use AI for due diligence in the shipbuilding industry, sparked by a Huxiu article claiming Chinese shipyards hold 80% of global orders, with headlines suggesting AI crushes Japan and South Korea. My first reaction was: what's the ceiling here, and what's the valuation logic? So I threw several reports, industry data screenshots, and shipyard order tables into a large model to build a mini due diligence workflow. Preparation involved defining the problem: Can order share translate to profit? Where does AI truly shorten delivery? Terms needed explanation: Deadweight Tonnage (DWT) is how much cargo a ship carries; Compensated Gross Tonnage (CGT) adjusts for construction labor intensity; LNG carriers transport liquefied natural gas; intelligent navigation systems are roughly perception and decision aids for ships.

Getting started, I inputted facts: In H1 2026, Chinese shipyards' new orders accounted for 82.3% of global DWT, up 173.1% YoY, asking it to map the competitive landscape. It quickly generated timelines and source cards on the right, lining up different numbers: reports saying Q1 share was 84.9%, data showing Feb China new orders at 80% vs Korea 11%, and articles citing ~70% for H1. I was pleasantly surprised; it flagged potential metric inconsistencies—some looked at new orders, others backlog; DWT vs CGT causes differences. I asked it to append applicable metrics to each number and export a table. Results were clean, though I had to manually merge shipyard aliases.

But I hit pitfalls fast. It conflated "90% of global orders" with "98% of mid-to-high-end orders," making it sound like all ship types were covered. Checking back, share and ship-type structure are different. Chinese yards dominate bulkers and containerships with cost/scale/delivery speed advantages, but Japan holds niches in special gas/engineering ships and marine electronics; Korea guards core positions in LNG carriers and ultra-large containerships, building green tech barriers in dual-fuel engines and smart navigation. Second snag: PDF drawings. Class society symbols, fuel types, engine models—I uploaded a scanned page, and it misidentified ammonia fuel as methanol. Shipowner financing, insurance, and compliance are tied to fuel/specs; such errors are unacceptable. Third: time metrics. Some firms say orders booked until 2028, others 2029. AI initially wrote "orders fully booked," failing to distinguish new orders, backlog, and delivery cycles.

Later, I dumb down the process: let the model only extract, not judge. Force source metrics next to every number; require counter-examples for every conclusion. E.g., for "Can AI crush Japan/Korea?", I didn't let it answer, but asked it to list evidence in three categories: design scheduling, welding QC, and supply chain coordination. It wasn't complete, but helped frame the argument. The real value was translating share into valuation questions. If a yard has many orders but thin margins, poor cash flow, and volatile prices, it's cyclical manufacturing. If AI enters scheduling, reduces steel waste, shortens block assembly time, and improves first-pass inspection rates, it shifts from a cost tool to a profit tool.

From an investment angle, China's shipbuilding strengths are scale, supply chain, and delivery certainty; weaknesses lie in not fully securing the highest-value segments. AI's role here: short-term cost reduction, mid-term learning curve compression for high-end ships, long-term data accumulation. Projects that just summarize news into reports? Not interested. Projects integrating design, procurement, welding, painting, and sea trial data? Worth serious discussion. Valuation logic depends on proving reduced rework, saved man-hours, and improved yield per delivery. Industry size alone isn't enough. Without trackable data, it's still just a story.

It's suitable for industry DD folks to quickly organize metrics and for investors to stress-test BP drafts. Directly using AI conclusions to invest in shipyards, industrial software, or robotics? Not recommended. Biggest pro: breaks scattered news into verifiable hypotheses. Biggest con: blends different metrics into smooth-sounding conclusions. Before using, you must watch the numerator and denominator yourself. Only projects connecting to shipyard scheduling and quality data layers are worth serious talk; AI that just writes industry reports isn't enough.