netzstrategen AI Operations.
Insights Glossary Resources Deutsch Self-Check
← All posts
Strategy

Expense or Asset: What's Left of the AI Budget at Year-End

Published on 6/23/2026 · André Hellmann

CAPEX AI Operations ROI Accounting

At year-end, management looks at the AI line item. The budget is gone. Nothing sits on the balance sheet. The pilot projects produced presentations — not value. That is how AI looks in most companies: a cost block that evaporates anew every year. It does not have to be that way. What matters is how AI is set up — and how it is booked.

Positioning

Discuss the next step in a free diagnostic call. Book a call →

Contents

Two ways to book AI

For every digital expenditure, there are two ways.

The first: immediately as an expense. The money flows out. The result is written off in the same year. Classic OPEX.

The second: capitalize it. The expenditure lands on the balance sheet as an asset. It is depreciated over its useful life. That is CAPEX — the logic companies have always used for machines and buildings.

The difference is not just cosmetic. Capitalized investments appear as value. They spread the cost across years. They show: something was created here that stays.

What accounting rules actually allow

Now the uncomfortable truth. For AI, capitalization is narrower than many assume. The following reflects German accounting and tax law.

Purchased digital assets can be capitalized. Acquired software, licenses, commissioned system development — under German commercial law this is even mandatory (§ 246 HGB). These assets sit on the balance sheet and are depreciated.

Self-built software is a special case. Commercial law grants an option to capitalize (§ 248 (2) HGB). Tax law says the opposite: § 5 (2) EStG requires intangible fixed assets to be acquired for consideration. Self-created AI therefore stays an immediate expense in the tax balance sheet.

Purchased software is effectively written off at once. Since the directive of 22.02.2022, Germany’s Federal Ministry of Finance allows a one-year useful life for computer hardware and software. This rule still applies (2025 and 2026). In practice: full depreciation in the year of acquisition.

Cloud and SaaS subscriptions are running expense. A monthly subscription for an AI tool is not an asset. It is consumed, not capitalized.

In short: tax law pushes AI spending toward expense. “Parking it on the balance sheet” to spread costs does not work.

Not tax advice, but orientation

The specific treatment depends on the individual case. It belongs in the hands of a tax advisor or auditor. netzstrategen advises on AI Operations — not on taxes.

The real question is a different one

If the balance sheet helps only so much — why this whole article?

Because “capitalize” has two meanings. The accounting one is settled. The more important one is economic.

To capitalize means to turn an expenditure into a lasting value. Something that stays. Something that compounds.

And this is where everything is decided. Not in the chart of accounts, but in the build.

Why most AI budgets evaporate

Most AI budgets create no lasting value. Three patterns repeat:

  • Pilot projects that end in a presentation.
  • One-off prompts that nobody reuses.
  • Tool subscriptions with no system behind them.

The money is spent. Nothing remains. That is the Pilot Graveyard — and it is expensive. At least 30% of GenAI projects are abandoned after the proof of concept (Source: Gartner, 2024). 95% of pilots deliver no measurable effect in the profit and loss statement (Source: MIT NANDA, 2025).

The numbers explain why 60% of companies see no material value despite continuous investment (Source: BCG, 2025). The technology is not what is missing. What is missing is lasting value behind the spend.

AI Operations builds value that lasts

There is another way. Set up correctly, AI creates real, lasting value:

  • Structured, governed data. The data history becomes a usable asset — clean, anonymized, retrievable.
  • Reusable building blocks. Agents, Skills, and Workflows that every team uses again — not once, but daily.
  • Documented systems and Cockpits. Work surfaces that stay in operation instead of disappearing after the pilot.

These values do not vanish at year-end. They grow. Each further step builds on the previous one.

The purchased or commissioned part — a platform license, a commissioned system — can even appear as an investment on the balance sheet. The rest is at minimum an economic value. Either way: substance is built, instead of budget burned.

This is exactly where AI Operations comes in: the permanent, measured operation of AI in daily business. The corresponding building principle is described in Production from Day One — every initiative targets operations from day one, not the demo.

What this means in practice

The question is not “OPEX or CAPEX”. The question is: does the money evaporate — or does a value remain?

A pilot that leaves only slides is an expense. An AI Operations setup that structures data and builds systems is an investment. Same amount, entirely different outcome.

AI should be treated like infrastructure, not like a project. Infrastructure is built once, properly. Then it runs. And it pays back over years.

That is exactly how AI is built at netzstrategen: designed for operations from day one, not for the next pilot. What gets built stays — as structured data, as reusable systems, as value the company owns.

In a free diagnostic call, netzstrategen shows which parts of the AI spend build substance — and which just evaporate.

Frequently asked questions about capitalizing AI costs

Can AI costs be capitalized on the balance sheet?

Partly. Under German law, software, licenses, and commissioned development acquired for consideration must be capitalized (§ 246 HGB). Self-created software is subject to an option under commercial law (§ 248 (2) HGB) but a prohibition under tax law (§ 5 (2) EStG). Cloud and SaaS subscriptions are running expense. The specific treatment belongs in the hands of a tax advisor.

Why is economic capitalization more important than the accounting kind?

Because tax law pushes AI spending toward expense anyway. The difference is made not by the chart of accounts, but by the build: does the AI spend create a lasting value — structured data, reusable systems, documented processes — or does it evaporate in the next pilot? Same amount, entirely different outcome.

How does an AI budget become a lasting asset?

By treating AI like infrastructure, not like a project. Structured and governed data, reusable Agents and Workflows, and documented systems stay in operation and grow. That is the core of AI Operations.

Sources

What's next