KKR paid $4B for a Japanese System Integrator (SI) that is built to resist change. The headline multiple looks aggressive. Strip out $2B in prime Tokyo real estate and the operating business traded at 8x EBITDA — below peer group for a firm with long-standing corporate partnerships and recurring T&M revenue.

FujiSoft builds the software inside things: the firmware for Toyota’s engine, the embedded logic for an MRI machine, the control software for assembly line robots. 10,000 engineers, almost $2B in annual revenue, decades-long relationships inside Japan’s largest conglomerates. Operating margins run 3x below global SI peers — but stable, protected by Japan’s closed enterprise procurement culture.

The deal was triggered by an activist investor pushing to revalue FujiSoft’s ~$2B in prime Tokyo real estate at market prices. KKR emerged as white knight.

The net net. FujiSoft is a below-market SI acquisition financed by fixing the balance sheet. The real estate monetization funds the transformation; the transformation is the multiple expansion story. The hold is just long enough to get a few AI advisory wins, improve margins through AI tooling, and sell at a higher multiple with proof points in hand. Full transformation belongs to the next owner.

The strategy.

• Sell/leaseback the Tokyo real estate first — the proceeds fund everything
• Segment the engineering workforce into three cohorts: AI transformation leads, empowered legacy maintainers, then manage the rest through attrition and voluntary retirement
• Start with the evals partnership before building anything — partner with R&D labs needing embedded safety-critical evals; it requires no new capability and generates the training data for proprietary models
• Structure the AI capability gap as a JV, not an acquisition — keep the partner entity legally and culturally separate; full acquisition loses the talent within 18 months

The risks.

• The patcher trap. Japan’s enterprises know how to buy more engineers. They do not know how to transform. FujiSoft’s near-term incentive is to become the patcher of last resort — good for T&M, corrosive to the exit multiple.
• The repricing deadlock. Clients never initiate outcome-based pricing. FujiSoft has to — against procurement teams that have bought T&M for 30 years and a consensus culture that makes repricing conversations slow. This doesn’t fail loudly; it just never starts.

The T&M trap

The paradox is common to every SI. The Japanese context makes it worse. AI productivity improvement means fewer billable hours for the same output — revenue compression until the pricing model changes. The transition from “we bill you for engineers” to “we charge you for outcomes” means a renegotiation of every client contract.

FujiSoft’s clients are Japanese mega-corporations whose procurement teams have been buying T&M for decades. Outcome-based pricing requires agreeing on what the outcome should be and how much it is worth — a negotiation the Japanese consensus culture is not ready for.

The transformation goal and the business model are in direct conflict. FujiSoft has to initiate the repricing conversation. Clients never ask to be repriced.

The unfirable army

FujiSoft has over 10k engineers. Under Japanese employment norms, with severance reaching two years of salary, a major workforce reduction would eat up a large chunk of the real estate proceeds.

Asking a 52-year-old embedded software engineer with 25 years at FujiSoft to become an AI prompt engineer is a confrontation with identity, not a reskilling program.

The workforce cannot be fired, so it must be reformed. Even modest AI productivity gains of 20–30% applied across 10k engineers creates real capacity without new hiring. The play is revenue expansion through capacity, not headcount reduction — with natural turnover gradually lowering costs.

The playbook: Segment the workforce into three cohorts.

Retrain a small group as AI transformation leads. These are the people sent into Toyota or Sumitomo to run AI advisory engagements. The new margin engine.

The second, and largest, cohort stays on legacy system maintenance — the embedded code, the aging banking infrastructure, the factory automation systems that require human oversight. This is FujiSoft’s moat. AI tooling raises output without changing headcount.

The third cohort — sunset cases managed through natural attrition, voluntary retirement incentives, and redeployment to lower-intensity roles. This runs on attrition time, not PE hold time.

The evals army

The threat story is: AI replaces engineers, headcount must fall, margin improves. Leading AI models are strong at writing new software from scratch in mainstream languages. They are significantly worse at embedded domains: C for automotive safety, medical device firmware, real-time OS for robotics. Safety-critical, heavily constrained, regulatory-governed — little data to learn from and zero tolerance for error.

AI productivity gains for FujiSoft’s engineers are higher than for generalist engineers. A skilled embedded engineer using AI to draft, then reviewing and validating output, goes faster. An offshore generalist running the same workflow is just the thing AI tools replace.

Evaluating AI-generated embedded code at scale requires exactly the kind of deep domain expertise FujiSoft has accumulated over decades. AI research labs need it to make their models better in safety-critical domains. Automotive OEMs need it to validate AI-generated firmware.

FujiSoft has the world’s largest concentrated pool of embedded software evaluators. That pool is a data and IP asset: partner with model developers, or build proprietary fine-tuned models for automotive and industrial domains, and sell them back to the same Japanese clients who already trust FujiSoft. The sequencing matters: start with the evals partnership — it requires no new capability and generates the training data for everything else.

The patcher trap

METI’s Digital Cliff — aging systems built on COBOL and custom software that nobody else runs, costing tens of billions from inaction — is the macro driver KKR cited publicly. But the enterprises most exposed are also the hardest to transform. Large, consensus-driven, with procurement processes that move slowly by design. Their instinct is to patch, not to modernize. FujiSoft is at risk of becoming the patcher of last resort: good for near-term T&M revenue, not great for exit multiples.

Comparable Japanese IT modernization programs take around a decade — longer than a standard PE hold. True repositioning from SI to AI advisory, from T&M to outcome pricing, from staffing firm to product, belongs to the next owner.

The honest exit story is: stabilized the balance sheet, improved margins through AI tooling, get a few AI advisory wins as proof points, and sold to a strategic acquirer or secondary at a higher multiple. Full transformation is the next PE firm’s problem.

The acqui-hire trap

FujiSoft holds every input that AI-native challengers lack: trusted relationships inside Japan’s largest enterprises, regulatory know-how, engineers inside client organizations, and a brand that mega-corp procurement will actually approve. What it lacks is AI capability.

Japan’s AI-native firms have the capability but not the distribution. PKSHA Technology has strong NLP and ML platform work for enterprise clients; Sakana AI, founded by ex-Google Brain researchers, is building foundation models with a Japan-specific angle; Recursive AI is targeting enterprise AI deployment. None of them can walk into Toyota’s procurement office and win a decade-long engagement on relationship credibility alone. Japanese mega-corps will not buy transformation from a three-year-old startup.

The obvious move is acquisition. Buy the capability, inject it into FujiSoft’s distribution network. But there’s a known failure mode in Japanese M&A: great talent does not stay post-acquisition inside legacy culture. The cultural gap is enormous. FujiSoft could buy the capability and lose it within 18 months, retaining the brand but not the people who made it valuable.

Structure it as a JV instead: keep the AI talent entity legally and culturally separate, connect it to FujiSoft’s distribution pipeline, and give it the autonomy that retains top engineers. Messier to model, but the talent stays.

Sakana AI is the template. Technically world-class, Japan-anchored, unburdened by legacy employment structures. A FujiSoft joint venture with a Sakana-type partner combines the distribution the challengers lack with the capability the incumbent cannot build fast enough.