Sovereignty is now better on
both economics and security.

Leaving Microsoft 365 is no longer ideology — on both economics and security, it has become the rational default.

Restate the premise the previous chapter (3-01) set: companies bought office as a package because it was cheaper and safer than standing it up themselves. That was the right call. The US cloud was, for a long time, cheap, safe, and stable. Choosing Microsoft 365 was the default that won on both cost and security, beyond dispute.

That premise has inverted from two directions at once. OSS plus sovereign AI is now better on both economics and security. This chapter reads the inversion as two problems — the Microsoft problem (the dependence itself) and the Trump problem (that the US government holding that dependence can no longer be trusted).

Microsoft 365 Used to Be the Cheap and Safe Default

Begin without hostility. Buying Microsoft 365 was rational at the time. Rather than standing up your own mail server, running a document platform, and managing authentication, bundling it all into one vendor's suite was cheaper, safer, and more reliable.

• Economics — the per-seat monthly charge cost less than the labor of running it yourself.
• Security — Microsoft's data centers genuinely beat your own server on availability, encryption, and patching.

This is the same structure 3-01 described as "buying was cheaper." Generic office work gave no reason to build in-house — buying won on both cost and security.

The problem is that these two premises — cheap and safe — have both inverted. Take them in turn.

The Microsoft Problem — The Convenience IS the Dependence

The first inversion comes from the dependence itself. The convenience of Microsoft 365 is, in the same stroke, the structure of dependence. 2-01 wrote that "the convenience and the hostage are two sides of the same chain"; here we read that chain as a security exposure.

On economics — the per-seat charge only rises. Once you're on it, the vendor sets the price. A price hike lands on a hostage with nowhere to flee. Then Copilot stacks on top — another few thousand yen per seat. You rent forever, and the other party sets the price. That is the structure of rent.

On control — this is the heart of the security case.

• Your data lives on a US company's cloud — documents, mail, calendar, all in a place beyond your reach. And under the US CLOUD Act, that data can be reached by US government warrant — even when the data sits outside the US.
• The telemetry is opaque — you cannot audit what is sent, when, or where. It runs inside a black box.
• Copilot routes your content through Microsoft's models — your business documents and conversations pass through the vendor's models, with no verification layer in between.

None of this is a weakness of feature. It is the underside of the convenience. It is convenient because every layer is wired through one account, and you are dependent because that one point is held by another. The same structure 2-01 named — "a closed bundle, hanging from someone else's key — that is the lock-in" — is read here as a security exposure.

The Trump Problem — The US Government Cannot Be Trusted

This is the heart of the chapter. Depending on US Big Tech means depending on the US government's goodwill. And that goodwill can no longer be assumed.

The logic is simple. Your data, your mail, your AI all sit on a US company's infrastructure. That US company falls under US government jurisdiction. So when the US government acts — sanctions, export controls, cutting off service to a targeted entity, company, or country — your "own" services can be severed regardless of your intent.

This is not an abstract worry. It happened, in reality, in 2025.

The ICC Case — "Your" Services Can Be Severed by a Foreign Government

In February 2025, the Trump administration sanctioned the International Criminal Court's chief prosecutor, Karim Khan, by executive order. He then lost access to his Microsoft account and moved to Switzerland's Proton Mail. (Microsoft denies it "in any way involved the cessation of services to the ICC" — but as a matter of fact, following US sanctions, the targeted party lost his US-cloud email.) That October, the ICC decided to move from Microsoft Office to OpenDesk, the OSS developed by Germany's center for digital sovereignty.

Read structurally, the lesson is one. A single act by the US government can sever "your" services on a US cloud. Even an international institution was not spared. Set aside where the responsibility lies (the vendor or the government) — it does not change the structure: as long as you depend, a third party's single act can cut you off.

Trump Is the Era's Upheaval-Side Figure

Widen the view by a step. The Trump administration has already shown it acts coercively and transactionally. Tariffs, sanctions, budgets, staffing — all swing on the spot. So you cannot assume continuity of access. Critical infrastructure must not be built on trust in a foreign government that has shown it will weaponize dependence.

This is not partisan denunciation — it is a matter of structure. In the frame that runs alongside the parent series: just as the Renaissance was an age of creation and, at the same time, an age of upheaval, this era has two sides. The side of creation under AI, and the side of upheaval, where the old order collapses and the new one has not yet stood up. Trump is the canonical figure on that upheaval side — "I decide everything alone" governance, the old era's logic of judgment-concentration pushed to its limit at the scale of the state.

It is precisely that unpredictability that makes sovereignty urgent now. Under a calm and stable hegemon, dependence could stay a convenience. But when the hegemon swings transactionally, dependence stops being convenient and becomes a liability.

That is why, for any organization outside the US (Japan, and the EU — which is the reason for the EU's "digital sovereignty" push), and even for US entities that fall out of favor, dependence on a US vendor is now not a convenience but a security liability.

OSS Plus Sovereign AI Resolves Both Problems

The two problems — the Microsoft problem (dependence) and the Trump problem (distrust of the government that holds it) — dissolve with one single solution: stop depending. The OSS foundation the Independence part stood up one layer at a time is itself the answer.

What the OSS foundation (Independence part) removes —

• Data lives on your own infrastructure — not a US company's cloud. Neither a CLOUD Act warrant nor foreign-government jurisdiction reaches it.
• It is auditable — settings, logs, and telemetry are all in your hands. No black box.
• No per-seat rent — a single server's fixed cost. No lever for a price hike.
• No foreign-government access lever. No remote kill-switch — there is no third party holding the act that cuts you off in the first place.

Sovereign AI — and the last layer is the AI. Run the local open-weight model you stood up in 2-11 on your own hardware.

• Data never leaves — documents and conversations never pass through a US AI API.
• It can be air-gapped — it runs even cut off from the network.
• It cannot be severed or repriced — a US AI API can be cut off and can be raised in price. Local weights can be neither.

And the decisive point: this is now viable. Open models have reached the "good enough" level, and hardware has come within affordable reach. Sovereignty is no longer a sacrifice of cost or quality — it is the cheaper AND safer choice.

Structurally — Both Premises Inverted at Once

Step back and restate it in the Shift part's logic. The Microsoft-dependent structure was rational as long as the US cloud was cheap, safe, and stable — the same as 3-01's "buying was cheaper." At the time, dependence was a convenience.

The two pillars that held that premise have both broken.

• The economic pillar — the AI cost inversion. One person plus AI can run the OSS foundation, and local weights have become "good enough" (Independence part). In-house became cheaper.
• The security pillar — the geopolitical risk of the Trump era. The hegemon swings transactionally, and dependence has been shown to be weaponizable. In-house became safer.

With both broken at once, the conclusion is one. Leaving Microsoft is not ideology — it is the new economic-and-security rational default. The same rationality that once steered companies to Microsoft 365 now steers them away from it. The premise inverted, so of course the conclusion inverts.

To the Next Chapter

This chapter set the premise of the Shift part's office (Microsoft) side — sovereignty is now the cheaper AND safer default, and the Microsoft problem and the Trump problem together make leaving Microsoft not an optional choice but the rational default.

That covers the office side of the two parallel worlds (3-01). From the next chapter, we turn to the other — the core (SIer) side. The next chapter asks why the SIer-commissioned model becomes structurally uneconomic — examining the structure in which the overhead of the outsourcing process itself exceeds the AI-native build cost (3-03).

Related articles

• 2-01: Stand Free of Microsoft and Google — Overview and Mapping
• 3-01: Companies Don't Write Their Own Code — Office and Core, Two Parallel Worlds
• 3-03: The Structural Uneconomy of the SIer Model