Artificial intelligence is often touted as a cost-saving technology. Automation of tasks, streamlining of operations, productivity gains, with the dominant narrative being the idea of a company capable of producing more with fewer resources.
Yet as AI agents move out of labs and into operational processes, another reality emerges. One where each decision taken by an agent, each interaction between models, each call to an external tool generates an increasing quantity of data that must be monitored, stored, analyzed and governed. AI not only creates new value streams, but also new operational debt.
It is in this context that Tsuga announces a fundraising of 35 million dollars, or nearly 30 million euros, led by General Catalyst and Singular, with the participation of DST Global Partners and Quantumlight. Founded in Paris in 2024, the company defends a simple thesis: the architecture on which modern observability is based is no longer adapted to the era of autonomous agents.
An industry built for the cloud era
Observability is one of the most discrete but also most critical layers of modern software. Its role is to enable technical teams to understand what is happening in their infrastructures through the collection of logs, traces and metrics.
For fifteen years, the market has been structured around players like Datadog, Splunk, Dynatrace, New Relic and Elastic. Their model is relatively simple: customers send their data to the provider’s infrastructure, which stores, indexes and analyzes it. The more the volumes increase, the more the bill increases.
This logic perfectly accompanied the rise of cloud computing. As businesses adopted distributed architectures, microservices, and multi-cloud environments, the need for visibility increased. Revenues from observability platforms grew at the same pace.
For a long time, the interests of suppliers and their customers seemed aligned. The growth of infrastructure automatically created more data and therefore more value. The arrival of artificial intelligence, however, profoundly modifies this equation.
When every agent becomes a telemetry factory
An AI agent does not behave like a traditional application. When a user queries a classic system, a few events are generally generated: a request, a response, a few service calls. When an agent intervenes, the chain becomes much more complex.
The system can request several models, call external tools, query different databases, generate chains of reasoning, trigger other specialized agents and then produce a final response.
Each step produces its own telemetry. Prompts, tokens, API calls, execution graphs, trust metrics, intermediate decisions: observability is no longer just a question of infrastructure. It becomes a question of understanding decision-making mechanisms.
This transformation creates a paradox where artificial intelligence is expected to reduce operational costs, and simultaneously increases supervision needs.
In some organizations, spending on monitoring is now growing almost as quickly as spending on the models themselves.
AI may not be the real culprit
Attributing this cost inflation solely to artificial intelligence would, however, be simplistic. Companies have already been facing an explosion in their infrastructure spending for several years. Storage becomes more expensive, architectures become more complex, data flows multiply and distributed systems generate more signals to monitor.
As AI acts more as an accelerator than a single cause, observability could be the visible symptom of a larger phenomenon: the continued growth of digital complexity.
This distinction is important because it determines the nature of the solutions to be provided. Is this an AI-specific problem or a structural problem of the cloud economy?
The emergence of a new market
One certainty remains, traditional metrics are no longer enough. Companies no longer just want to know if an application works correctly. They want to understand why an agent made a decision, what tools he used, what models were involved and what level of confidence can be placed in the result.
This development is giving rise to a new software category, the notions of AI Observability, Agent Observability, AI Governance and even AI Traceability are beginning to converge. Behind sometimes different terminologies, the same need appears: making AI systems auditable.
As companies deploy agents in finance, HR, legal or industrial areas, the question of liability becomes central. Understanding how a decision was made is no longer an option. This is an operational and soon regulatory requirement.
An already very crowded market
Tsuga, however, is not entering virgin territory; the major historical players quickly identified the opportunity. Datadog is already developing advanced model and agent monitoring functions. Dynatrace is pushing its Davis AI offering. New Relic, Splunk and Elastic are gradually enriching their platforms with capabilities specific to AI workloads.
At the same time, a new generation of specialists has emerged. Arize AI, Langfuse, Helicone or WhyLabs focus on model observability, prompt analysis, hallucination detection or performance monitoring of generative systems.
The subject is therefore no longer the existence of a market, but differentiation.
The real bet: architecture
This is precisely where Tsuga tries to distinguish itself, it does not present its innovation as an additional functionality, and directly attacks the dominant architecture of the sector.
Instead of centralizing data in its own infrastructures, the platform is deployed directly in the customer’s cloud environment. Data remains in AWS, Azure, Google Cloud or a sovereign cloud. They do not pass through Tsuga’s systems.
The argument is twofold, on the one hand, this approach reduces costs associated with data duplication and transfer, and on the other hand, it responds to growing sovereignty and governance concerns.
This strategy reveals a strong intuition: observability data itself becomes strategic assets. Traces now contain prompts, decisions made by agents, business information and sometimes sensitive data. For some companies, outsourcing them becomes as problematic as entrusting their customer data.
The question is then no longer just technical and becomes regulatory and economic.
The unexpected return of software accompanied by services
Another notable element: Tsuga does not only sell a platform, and also highlights teams of engineers responsible for supporting customers in the continuous optimization of their observability environment.
This approach is reminiscent of some recent developments in the AI market. After two decades of standardized SaaS, several software categories are returning to hybrid models combining product and expertise.
Tsuga’s real competitors
The most dangerous competition may not be found at Datadog or Splunk, and is probably at Microsoft, AWS and Google. Hyperscalers already control infrastructure, data, observability tools and, increasingly, artificial intelligence models.
They therefore have all the necessary elements to natively integrate these functions into their platforms. This is the main strategic risk for any startup in this category.
If AI observability becomes a standard feature of cloud environments, differentiation will need to be based on more than just technical capabilities.
A battle that has only just begun
The tech industry has spent the last three years building models, co-pilots and agents. She is now discovering that the value lies not only in the ability to automate, but also in the ability to control that automation.
The history of the cloud has produced its monitoring champions, the history of artificial intelligence could produce its governance champions.
By raising nearly 30 million euros, Tsuga is betting that this new layer of infrastructure will become as essential tomorrow as observability became yesterday. The success of this thesis will depend less on the company’s ability to monitor agents than on its ability to answer a question that all organizations will soon have to ask themselves: who monitors the systems that make decisions for us?
