FDE war moves east
rented labor, missing proof

For a year, the AI industry answered “enterprises cannot deploy” with one unit of measurement: engineers. Five deployment legions, and headcount was the promise. On September 10, 2026, the unit changed to channels. Three deals in a single day turned forward-deployed engineering (FDE) from a Western arms race of in-house legions into a rentable distribution business in Asia: Moonshot rented five Chinese system integrators, Palantir made Fujitsu a Global FDE Partner, and Sakana-SCSK-Sumitomo built a channel through a trading house’s roughly 100,000 client companies. Deployment labor is now purchasable on the open market. What agents do after install — and who proves it — is still a blank in all three markets.

One day, three deals: FDE becomes a distribution business

The first deal is in China. Moonshot announced its Kimi Enterprise Partner Program on September 10, pairing with five system integrators — Beijing Teamsun, Kingsoft Cloud, Beijing Asiacom, AsiaInfo Technologies and Chinasoft International — to form joint FDE teams at client sites, the first time a Chinese model maker formally adopted the FDE model (source: Caixin Global, Sep 11 — https://www.caixinglobal.com/2026-09-11/moonshot-ai-expands-enterprise-push-through-it-services-partnerships-102483755.html ). Seoul Economic Daily adds the commercial detail: revenue is split per contract, covering token usage fees and system implementation charges, with a possible IPO as early as next year (source: Seoul Economic Daily, Sep 12 — https://en.sedaily.com/international/2026/09/12/chinas-moonshot-ai-adopts-palantir-style-model-ahead-of-ipo ).

The second is in Japan. Fujitsu signed a new agreement with Palantir Technologies Japan KK for Palantir AIP and Foundry and became a Global FDE Partner, bringing its own Takane LLM and Uvance offerings to sovereign, production-grade architectures. Fujitsu’s COO put it plainly: “Palantir pioneered and established the FDE concept.” The official release cites a supply chain case — data integrated across more than 3,000 suppliers and 18 factories, over $10 million in savings within one year, doubled operational productivity. Those are business-outcome numbers, not independently verified execution evidence (source: JCN Newswire, Sep 10 — https://www.jcnnewswire.com/pressrelease/109983/3/ ).

The third is a homegrown triple alliance. Sakana AI, SCSK and Sumitomo Corporation agreed on September 10 to a comprehensive partnership: Sakana supplies the models, SCSK handles implementation, and Sumitomo routes through its network of roughly 100,000 client companies, starting with finance and manufacturing, then cybersecurity and social infrastructure, with overseas expansion planned. The three companies wrote a sentence worth quoting in full: successful use in day-to-day operations “requires more than model selection alone” — integrating AI with business processes, data and legacy systems while addressing quality, information security, governance and post-deployment operations (source: IBTimes JP, Sep 10 — https://jp.ibtimes.com/sakana-ai-scsk-sumitomo-corp-forge-japan-ai-partnership-104333 ; Nikkei Asia — https://asia.nikkei.com/business/technology/artificial-intelligence/sumitomo-partners-with-sakana-ai-to-push-japan-built-ai-model ).

Three channel structures, one move: nobody builds deployment in-house anymore

Line the three deals up and the structure is clean: a Chinese model maker rents an SI network (supply comes from the integrators’ client relationships and industry knowledge), a US platform rents a national champion (Fujitsu’s industry depth plus sovereign architecture), and a Japanese lab builds a triple alliance (lab plus implementer plus trading-house channel). What all three share is the interesting part — none chose to build a large in-house FDE legion.

That reads as a market correction of the first wave. Microsoft’s Frontier Co., AWS’s FDE organization, OpenAI’s Deployment Co., Anthropic’s Ode, Google’s rented Accenture legion — the first wave was build or joint-venture; the September 10 Asia triple is all rent. Deployment labor has moved from a differentiator to a commodity: FDE supply just went from five legions to a global distribution channel.

Commoditized is the labor, not the proof

All three agreements answer “who installs.” None answers “then what?” After the FDEs wire the systems and leave: who independently verifies the agent does what was contracted? Who proves outcomes? Does the evidence survive after the vendor or partner departs? Moonshot’s joint teams, Fujitsu’s Global FDE Partner status, Sakana’s triple alliance — every promise is an on-site action, none is a post-departure evidence obligation.

We asked the same question of the five legions on September 10 (see “FDE arms race: 1,000 engineers in, who leaves the evidence?”). Three new markets now confirm it: regionalization does not close the evidence gap, it copies it into more markets. What has been commoditized is deployment labor — channels can be rented, headcount can be bought. Trust cannot be rented. The post-deployment proof is the one part of this business that has not been commoditized, and it is exactly the part buyers should be paying for.

Our judgment

First, FDE commoditization is a net win for buyers. Supply has gone from scarce to abundant, and negotiating power has shifted; any quote that still prices “FDE headcount / certification badges” as a premium should be challenged. When one SI serves several vendors at once, a badge is a distribution signal, not a quality signal.

Second, channels can be rented; trust cannot. The three deals prove deployment labor has been commoditized, and they also prove the post-deployment evidence layer has not followed — the blank is identical across regions, and that is the most expensive information asymmetry buyers face right now: vendors are selling entry, while most failures happen after departure.

Third, the Asian players’ own statements admit this. Sakana lists post-deployment operations as a necessary condition for success; Fujitsu sells a business-outcome story (over $10M saved in a year), not an execution-evidence one. The vendors know the hard part starts after install — none of them has written the proof obligation into a contract.

Fourth, the buyer implication: put post-deployment evidence into contract language (who verifies, where the verification data lives, whether evidence survives vendor departure, how acceptance is defined) before choosing which FDE partner to use. FDE supply will keep over-supplying; the evidence gap will not close by itself.

Buyer’s action list

1) When evaluating FDE partners, drop “engineer count” and “certification badges” from the decision weights and replace them with three questions: who independently verifies agent actions after install? Where does the verification data live? Does the evidence survive the engagement?

2) Change acceptance clauses from “what was delivered” to “what the agent keeps doing, who proves it, and who holds the proof data.”

3) Use the channel structure as leverage: the Chinese SI channel, the Japanese national-champion channel and the local alliance channel are all competing for customers — buyers are in a position to demand evidence clauses rather than accept headcount promises.

4) Keep an independent, third-party action-level audit as an acceptance condition, and put “audit data must be exportable and portable” in the contract.