"There is a 10 per cent probability that AI will have wiped out humanity within ten years."

We now read statements like this surprisingly often. They frequently come from American AI companies that are themselves at the forefront of development: OpenAI, Meta, Anthropic and others warn of an apocalypse. AI as a machine for ending the world.

It sounds dramatic. It attracts attention and clicks. It also gives companies that claim to control the most dangerous technology in human history an almost mythical importance.

The claim is about as plausible as the idea that my miniature poodle Lotte will one day attack me according to a strategic plan.

Media outlets still like to repeat such scenarios, partly because the actual capabilities and limits of today's AI are often misunderstood. The key clarification is this: AI cannot think.

What we call artificial intelligence is still based largely on statistical methods. Large language models recognise patterns and predict words, sentences and structures, often remarkably well. They do not think, understand or act consciously.

Consider the beginning of the saying "The early bird catches the W...". "Worm" is highly likely to follow because it is a familiar expression. If the bird outside my window is eating a grain, however, "worm" is wrong in that context.

That is the problem. AI often lacks the relevant context. It does not automatically know what is happening outside the window, nor which processes, rules, data, documents, roles and dependencies apply in a company. Put plainly: AI without context is stupid.

Why AI providers fuel the alarmism

Why this alarmism? Because it benefits American AI providers.

Anyone claiming that a technology could wipe out humanity is also implying that it is immensely powerful. Whoever controls it controls the future. That is a perfect narrative for capital markets, investors, politics and the media.

For companies such as OpenAI, Anthropic and other AI providers, the issue is enormous valuations, planned or anticipated flotations, huge funding rounds and future revenues worth billions. The greater the claimed historical importance of the technology, the easier it is to justify valuations that often have little to do with current revenues.

The message is clear: we are burning billions today on data centres, chips, models, talent and infrastructure, but tomorrow we will control a market larger than software, cloud, search, advertising and office applications combined.

At the same time, established business models are under pressure. Search engines, conventional software licences, advertising platforms, cloud services, consulting models and productivity tools are being reshaped by AI. Anyone claiming loudly enough that they are working on the world’s most dangerous and important technology positions themselves as an innovator and an indispensable gatekeeper.

The alarmism serves several purposes at once. It attracts attention, justifies high levels of investment, raises barriers for smaller competitors, influences regulation and increases the provider’s own importance. If AI supposedly carries existential risks, only the largest, best-funded and supposedly most responsible providers should be allowed to operate it. A warning about the apocalypse becomes a business model.

That does not mean that all risks are invented. There are real dangers: disinformation, discrimination, data leakage, cyberattacks, concentration of power, changes to employment and loss of control in automated processes. These are concrete, present and testable risks. They have little to do with Hollywood scenarios involving a superintelligence that suddenly wakes up.

The real question: who controls the token economy?

More important than the apocalypse debate is the question of who will control data, models, platforms, infrastructure and value creation in the future. AI has produced a new basic economic unit: the token.

AI models process knowledge in tokens. Every request, document, analysis, summary and agent step is broken down into tokens and charged accordingly. Anyone who consumes tokens without building their own token infrastructure, models or orchestration layers exports value creation, knowledge and economic control.

This is the real token economy. It sounds technical, but it is highly political and brutally concrete from a business perspective. When German and European companies conduct most of their AI activity through American hyperscalers, they pay more than software fees. They import the central computing unit of the coming knowledge economy.

The analysis underlying this article describes a scenario in which global token consumption rises sharply by 2030, driven above all by agentic AI systems. For Germany, it cites a possible token import of up to 157 billion euros per year by 2030, around 3.51 per cent of GDP. For comparison, that would be on a scale comparable to today’s major structural import dependencies.

Whether this forecast materialises exactly is less important than the direction of travel. If AI becomes the operating system of value creation in companies, token costs will become a new location factor. They will determine whether productivity gains remain within the company, Germany and Europe, or whether a significant share of the margin flows to foreign platforms.

This makes one thing clear: companies do not pay with money alone. They often pay twice. First through usage fees, subscriptions and token costs. Second through their knowledge, processes, documents, questions, working methods and context. Yet this knowledge is precisely the raw material from which future competitive advantages in AI will emerge.

Bigger is not automatically better

American hyperscalers largely follow the logic of larger models, more data, more compute and more parameters. That sounds like progress. For companies, however, it is often not the right answer. Companies do not necessarily need the largest model in the world. They need AI systems that know their own context.

This context is not openly available on the internet. It is deeply embedded in business IT systems: ERP and SAP environments, CRM and DMS systems, PPS systems, databases, spreadsheets, company wikis, SharePoint structures, office documents, process descriptions and proprietary specialist applications.

That is where the actual knowledge of the company resides. That is precisely where AI must connect.

Simply giving employees a generic chat window is therefore insufficient. Companies that want to make AI genuinely productive need a platform that integrates AI deeply into existing systems, respects permissions, connects data sources intelligently, orchestrates processes, controls costs and produces traceable results.

Why enterprise AI orchestration is becoming indispensable

A new software category is emerging for this purpose: enterprise AI management and orchestration platforms. These platforms are not built as an addition to CRM and ERP. They are the intelligence layer placed above existing systems, making them genuinely AI-capable.

Analysts are already forecasting a market volume for this category from 2025 onwards that could be larger than the combined ERP and CRM markets. This is plausible because AI will not simply be another application. It will become the new foundation layer of business value creation.

Platforms such as the neuland.ai HUB address precisely this point. The HUB is not designed as a standalone AI tool, but as an enterprise AI operating platform with an SDK, open APIs, a UI library, connectors and orchestration logic. Its aim is not to replace existing IT systems, but to make them AI-capable.

This is technologically decisive. Companies have invested heavily in SAP, ERP, CRM, DMS, databases and bespoke specialist systems over many years or decades. A sound AI strategy must not devalue these investments. It must protect them and make them usable. The neuland.ai HUB can operate as a controlling layer above existing systems and connect to in-house developments or third-party systems. AI therefore becomes an intelligent process and knowledge layer above existing IT, rather than a break with it.

The handling of large volumes of data is particularly relevant. Many AI projects fail because they treat corporate knowledge as one enormous pile of documents. Everything is searched, indexed or sent directly to a large language model. This is expensive, slow and often unsatisfactory in terms of quality.

The neuland.ai HUB follows a different approach. It combines small language models, knowledge graphs and ontologies with agentic retrieval-augmented generation. Corporate data is not merely searched. It is structured semantically. Entities, relationships and contexts are mapped explicitly. The result is not a flat document index, but a navigable corporate memory.

That is the decisive difference: AI should not simply guess. It should know where relevant information is located, how it is connected and which sources actually matter for a task.

Token costs are not a side issue

The token economy makes this architectural difference directly measurable. Conventional retrieval-augmented generation solutions often send large amounts of document context to an LLM. This creates high reading costs for queries and high writing costs when knowledge is added. At the same time, every unnecessary token increases the risk of transferring sensitive information to external systems.

A clear architecture layer determines which information reaches a model. Image: AI generated with neuland.ai HUB

A context layer made up of a knowledge graph, ontology, small language models and agentic orchestration reduces this effort. Instead of sending everything to a large model, the system first determines which information is genuinely relevant. The large LLM is then used only where it is actually needed.

The available documents on the neuland.ai HUB cite specific cost advantages. Query reading costs can fall by around a factor of 10 compared with conventional retrieval-augmented generation approaches. Writing costs for adding knowledge can fall by around a factor of 100. Elsewhere, internal measurements are described in which input tokens could be reduced by a factor of 50 to 100 and output tokens by a factor of 5 to 10.

This is not a technical detail. It is a business argument. Fewer tokens mean lower costs, faster responses, better scalability and less dependence on expensive frontier models. Above all, fewer sensitive details need to be transferred to external models.

In an economy in which tokens become the billing unit for knowledge, token efficiency is a competitive factor. Intelligent token orchestration reduces costs and preserves greater control over data, processes and value creation.

Independence from language models and hyperscalers

Another key advantage is independence from individual LLMs and providers. Companies should not tie their AI strategy to a particular model. Today it may be GPT, tomorrow Claude, Gemini, Mistral, Llama or a specialised internal model. Prices, performance, data protection conditions and availability are constantly changing.

The neuland.ai HUB is designed for a multi-model strategy. Frontier models, open-source models and proprietary models can be combined. This makes the company less dependent on individual providers and reduces the risk of another form of lock-in.

This is particularly important for Europe. Companies that base their AI infrastructure entirely on American or Chinese hyperscalers do not simply surrender data. In the long term, they also surrender control, negotiating power and digital sovereignty.

The operating model is another consideration. The neuland.ai HUB is designed for on-premise operation, sovereign cloud and conventional cloud environments. The platform is based on open-source technologies and deliberately avoids proprietary services from individual hyperscalers. This matters because data sovereignty cannot be merely a contractual promise. It must be possible at an architectural level.

Governance, guardrails and real process automation

For regulated industries, public administration, critical infrastructure and the legal and financial sectors, this is not a luxury. It is a prerequisite. AI must be operated securely, audibly, controllably and in compliance with applicable rules. This requires governance, budget control, audit logs, role and permission models, and system-level guardrails.

Corporate AI must not operate on the principle that the user asks and the model does something. Professional AI needs direction. Which data may an agent access? Which action may it perform? When is human approval required? Which model may be used for which data? What costs may be incurred? Which regulatory requirements apply?

That is precisely why an orchestration platform is needed. Generative capabilities and deterministic process steps must be brought together under unified control. AI then becomes more than an assistant for individual employees. It becomes a controllable process layer for entire organisations.

Europe's opportunity is not the largest model

Europe is unlikely to win the race for the largest foundation models. The race for the orchestration layer, however, has only just begun. This is where European strengths lie: domain knowledge, industrial process expertise, data protection, regulation, quality, security and depth of integration.

Germany has some of the deepest industrial domain knowledge in the world, in manufacturing, logistics, quality assurance, administration, law, compliance and the Mittelstand. An American or Chinese hyperscaler cannot simply replicate this knowledge. It must be made accessible through technology. That requires sovereign AI platforms.

The greatest danger posed by today’s AI is therefore not that it will wipe out humanity in ten years. The greater danger is that companies and economies will lose control over their knowledge, data, processes, token costs and digital value creation.

The problem is not a science-fiction apocalypse. It is the very real dependence on external platforms.

AI cannot think. It can still be highly useful if it receives the right context, is securely integrated, costs are controlled and authority over data and models remains with the company.

AI must therefore become the new core of companies. Organisations must reorganise around it. Europe also needs its own AI platforms that operate securely and reliably, protect data sovereignty, use tokens efficiently and remain economically independent of American and Chinese hyperscalers.

That is why the neuland.ai HUB was developed.