ERP's Second Act: How 30-Year-Old Systems Became AI's Critical Infrastructure

The spotlight in AI has shifted from model builders like OpenAI and Anthropic to enterprise software vendors that have been overlooked for thirty years — Kingdee, Yonyou, and SAP. The article uses a human-body analogy: large language models are the brain, agents are the senses and limbs, but the missing link is the circulatory system — the blood vessels that carry signals between brain and body. In enterprises, that circulatory system is the ERP.

One: Brains Are Becoming Cheap

History shows every transformative technology — steam, electricity, internet — eventually becomes a utility. Open-source models are improving monthly and inference costs are dropping yearly; the lead time for "best model" advantage has shrunk from three years to six months. When intelligence itself becomes cheap, value moves to where that intelligence is applied: factory scheduling, hospital records, port manifests, balance sheets.

Two: Labs Are Moving to Customer Sites

Model companies now hire enterprise solution engineers and embed teams in client meeting rooms. Their roadmap is explicit: model → agent → workflow → enterprise production. An API is only an entry ticket; the real challenge is integrating with financial systems, respecting internal permissions, and actually processing expense reports, tracking orders, and collecting receivables. The cost to make a model "operational" can be ten times the model license fee, and whoever controls that integration point captures the value share.

Three: The Paradox of Commoditization

The cheaper and more open models become, the less lock-in they create. The electricity analogy applies: power plants earn revenue, but the vast wealth created by electricity accrues to factories. Enterprises that cut headcount by 100, reduce inventory by 20%, or halve bad debt keep those savings on their own balance sheets — none flows back to the model provider. The largest value pool in AI will likely sit at the production layer, where the "blood vessels" reside.

Four: Thirty Years of Building Blood Vessels

Kingdee and peers spent three decades digitizing procurement, finance, inventory, production, and sales. In doing so they accumulated two irreplaceable assets: (1) a living vascular network — every yuan's flow, every approval path, every handoff mapped and continuously updated; (2) blood-type trust — enterprises hand over their core assets because of thirty years of reliable service. Previously this infrastructure merely recorded transactions; now it is the only conduit through which an AI brain can command enterprise limbs.

Past: Human → ERP<br/>Future: Human → AI → ERP → Business

AI handles understanding and decision-making; ERP handles connecting data, rules, and execution. Without ERP, AI is a brain trapped in a skull — a vegetable unable to move limbs.

Five: Silicon Valley Wants to Transplant Vessels, But Blood Types Don't Match

Why don't model giants build their own ERP? The rejection reaction. An enterprise's vascular layout is shaped by thirty years of business processes; the blood is proprietary data; the vessel walls carry compliance, audit, and cultural immune responses. Foreign blood triggers immediate rejection. The ERP moat is not technology but process knowledge, industry know-how, and customer trust — assets that cannot be bought or replicated in three years. ERP vendors are not transplanting vessels; they are the vessels. AI brains only need a tube connecting to this existing network.

Six: The Optimal Position — Don't Build Brains, Build Hands and Keep the Vessels

Smart ERP vendors are doing three things: (1) avoid building foundation models — that plays to competitors' strength; (2) build agents (the hands and feet) to execute tasks; (3) guard the vascular network and plug the best brains into it. This combination is especially valuable in China: if models commoditize rapidly, Chinese firms should not wage a capital-intensive arms race at the model layer. Instead, adopt a Tian Ji horse-racing strategy — stay current on models, but dig deep into the application layer, embedding AI into China's massive real economy. Compute can be rented, models can be licensed, but customers, processes, domain knowledge, data, and trust cannot be rented — and those are exactly the thirty-year strongholds of Kingdee, Yonyou, Taxfriend, and countless industrial software companies. If they embed AI into business operations, the greatest value created by AI in China will remain in factories, ports, supply chains, and rewritten balance sheets.

Epilogue: The Old Supporting Actor at Center Stage

ERP finally understands its role: it need not be the brain or the strongest muscle. It simply holds the only vascular system that can connect brain to limb. The question is no longer "whose model is strongest?" but "which company can turn the cheapest AI into the most expensive real-world productivity?" Models provide intelligence; systems deliver execution. Maximum value arises at their intersection — not on launch stages or leaderboards, but in factory schedules, hospital wards, port manifests, and every rewritten balance sheet. After thirty years in the audience, ERP steps into the spotlight. It is not the protagonist; it is the stage that lets every protagonist come alive.