[F] sourced fact · [B] attributed claim · [I] inference. Long version: wiki/organization/hitachi-digital.md.
The situation, in four facts
[F] The data layer exists, is named, and belongs to DSS: Jun Abe’s sector deck labels the Data Fabric band “DSS Sector” and gives its job as “integration and structuring of IT/OT data” and “enable cross-sector and cross-industry data utilization” (Abe, Investor Day, slide 16; slide numbers are the printed ones). [F] It has money: the ¥30bn FY2026 organic investment opens its technology column with “enhance data fabric capabilities” (slide 18). [F] It has no architect yet — the “HMAX-focused CTO office” is an FY26 deliverable on Anand Birje’s path-forward slide (Birje, 28 Oct 2026). [F] And neither deck contains confidentiality, provenance or attestation vocabulary: Abe’s Security pillar means “infrastructure that must never stop” (slide 14), and Birje’s stack has no security layer at all.
[I] SINE offers a property of a layer that is funded and owned, into an office with no occupant — and the opening is time-boxed: while an architecture is drawn, a mechanism is a design input; once it lands, a change request.
1. Which layer — Data Fabric, or AI Operations?
[I] Written in the Data Fabric, argued from AI Operations. An attestation is a property of a datum created at capture — the Data Fabric’s own stated job. “Preventing inappropriate AI behavior” (Birje) is a property of the model, and cannot answer the post-incident question: what did the system know, and was it true. [F] Not hypothetical — IWIM is justified by “large-scale power outages and railway accidents resulting from incorrect AI decisions”. [I] Provenance is a precondition for AI assurance, not a parallel concern.
[I] Do not claim the AI Operations box — NVIDIA, ServiceNow and the model vendors are on Birje’s partner slide, so claiming it makes SINE a competitor to a named partner in the first sentence; claiming a provenance property in the Data Fabric that makes AI Operations auditable makes SINE a component. The Data Fabric is contested too, but [F] Abe’s six sector-level partners are all compute or model layer (slide 19), [I] and lineage answers a different question: inside one company it suffices; across a company boundary, lineage is a promise and a proof is a proof.
2. Who to talk to while the counterpart does not exist
[I] Three beats, each repeatable internally by someone who need not own it: (i) the decision is already made — frontline data becomes a knowledge asset AI acts on; (ii) one property of that data has no owner — whether a figure derived from it can be checked by someone not allowed to see it, today carried by contracts, which stop at the company boundary; (iii) the office that would own it is being built this year, and the budget line exists. The ask is a question, not a pitch: who is drawing the HMAX Data Fabric architecture, and is anyone responsible for data provenance in it? Only an insider can answer, it costs them one sentence, and the answer is a name.
| Audience [F] | Role |
|---|---|
| HMAX-focused CTO office (DSS) | Eventual owner; unstaffed — target, not door |
| Physical AI FDE Team (Abe sl. 15) | DSS–sector crossing point, already inside Rail on Factory IQ; cheapest first call |
| CI Customer Zero — 175 sites | The site, if a run is needed; different sector and budget |
| Energy / EcoSpace | Only unit already shipping a sustainability-data product |
| Standardization Cttee / H-AIST CECRL | Separate door: Japan, slow, durable |
| Hitachi Rail CSO organisation | Sponsor who can route — not the buyer |
[I] Six is too many to open at once: the request to Dr. Shibasaki is to rank them.
The nuance on circular-economy and value-chain data
[F] The leading Data Fabric use case is “optimal production linked to green power supply and demand” — a value-chain sustainability case in Hitachi’s own hand. [F] But all three value examples run between Hitachi sectors; no arrow reaches a supplier, operator, customer or assurer (slide 16). [I] “Cross-industry” means across Hitachi’s own industries.
[F] The CSO’s binding constraint sits where the arrows stop: actual supplier emissions data covers 76 suppliers, 14.93% of Scope 3 Cat. 1; ~85% of the purchased-goods footprint is estimated; HMAX Mobility serves 2,500 trainsets, over 35% built by another OEM (question set). [I] Data Hitachi captures itself is a Data Fabric problem it has solved; data Hitachi must be given is drawn nowhere: Hitachi can integrate data it is given; for data it is not given, it has no architecture.
[I] The caveat to carry rather than hide: nothing says Hitachi feels this as a problem — [F] Abe’s risks are an IT-investment slowdown and component prices (slide 22). [I] So we carry the case to DSS, and the CSO’s demand signal — a tender requirement, an ISO 14067 boundary decision, a supplier that refuses — is the evidence they lack.
3. Becoming legible to the DSS CEO and CTO
[F] Their scoreboard: AI-related revenue ¥800bn at CAGR 20–25%, HMAX ¥300bn at CAGR 50–60%, Lumada ratio 62% → 70% plan, adj. EBITA 16–18%, benchmark “the gap to best-in-class multiples”. [I] Five translations:
- IP, not a project — a reusable component shipping inside the Data Fabric, identical in every vertical (“IP-led differentiation” is Birje’s phrase). Never a study, pilot or governance work: all three read as headcount, which the margin story is shedding.
- It opens the channel that is already growing. [F] GlobalLogic’s stand-alone revenue grew 2.6% in FY2025 while synergy revenue booked by other Hitachi units grew 73%, to $2,500m (slide 4). [I] The external book is flat — internal-first is their fastest motion, not our fallback.
- A falsifiable test, not a programme — one figure from one Hitachi site, checkable by a third party, raw data never leaving the site; pass or fail in weeks.
- Say what SINE is not, early: not a data platform, not a model vendor, not a competitor to anyone on the partner slide. And pick one vertical — Energy and Auto are where cross-boundary supplier data is already priced.
- Arrive protocol-shaped. [F] Birje’s integration row reads “APIs / MCP / Orchestration”. [I] An open mechanism travelling with a datum fits that row without a partner tier — delivered as a one-page component spec, not a memo.
What would refute this. [I] A CTO office already staffed with a frozen architecture; a Security remit that does cover provenance between parties; a cross-company arrow planned but undrawn. [B] And this rests on investor decks, where an absence is weak evidence.
Internal. Don’t name the FDE team to Rail as “the cheapest conversation” — ask who the FDEs on the Factory IQ engagement are. Four follow-up questions: sources/hitachi/2026-11-01-questions-hitachi-digital-addendum.md, §VIII.
Sources. Abe, 10 Jun 2026 · Birje, 28 Oct 2026 · question set, 15 Oct 2026.