The aviation industry is affected by the crisis in the world. The current war with Iran is adding economic uncertainty, increasing geopolitical tensions, and impacting airlines in the region. Like the previous crisis, it is testing the resilience of the sector. Once again, aviation has helped repatriating travelers blocked in the Gulf and will return stronger. Better days will come.

Next week, on 25-26 March 2026, aviation leaders will gather again in Singapore for the next edition of the Aviation Festival Asia. The conversations will inevitably touch on the current crisis but also focus on the central technology theme of the past three years: Artificial Intelligence (AI).

In this article, I take a step back from the daily use of AI assistants and will look into the “models” underpinning AI, to better understand where the next developments will take us. I’ve picked three types of models:

  • Market models, such as the “large market model” developed by Fetcherr, an airline pricing software company
  • Language models, such as Claude[1] and other Large Language Models, used more and more by travelers and airline staff for daily requests
  • World models, an emerging category explored by companies like AMI Labs[2], a “frontier AI lab” that has just raised a $1B seed round

The market model

The first keynote at the upcoming Aviation Festival will be delivered by Fetcherr, a tech company that attracted attention in the airline industry with its new system. Their machine learning application simulates the air travel marketplace, whereas traditional revenue management systems rely on forecasting demand and optimizing fares based on historical booking patterns.

At the core of the system lies the “Large Market Model”[3]. This dynamic model learns from multiple signals such as booking curves, competitor pricing, seasonality, and customer willingness to pay. The objective is to “understand” the market (e.g. consumer responses, competitors’ reactions) and to anticipate how the market will behave to the next airline pricing decision.

If an airline raises or lowers a fare, what will be the impact on the booking curve in the next hours or days? It is likelihood forecast adapted to economic dynamics of a marketplace. For airlines used to optimization compared to predictive curves, this approach represents a mindset shift. Pricing becomes the instantaneous match between supply and demand, like in a stock market.

The language model

ChatGPT was released on 30 November 2022. I remember the date. A few months later Anthropic released Claude widely and it has since become my favorite Large Language Model (LLM) and AI assistant for work. These language models are trained on massive text datasets to understand human language by statistically predicting the next character, letter or word.

They perform various knowledge tasks and roles. They began as dialogue partners, like a chat interface to Wikipedia, or to a more productive extent, like being my personal editor for this article (I keep writing myself). They evolved quickly into executing tasks, the most common use case being programming. Nvidia for example argued[4] that 100% of their software engineers use Claude Code, or similar coding assistants.

Over time, LLMs have become so powerful that some wonder if language models have become conscious. At least they give this illusion. The AI industry (OpenAI, Anthropic, etc.) is currently debating how LLMs should behave, not to mislead users into emotional dependency or misplaced trust.

Interestingly these language models do not simulate markets or do not understand the physical world. Their knowledge and understanding are entirely derived from learning human language. In aviation, language models already support a wide range of applications, from customer service reps relying on AI assistants to retrieve relevant information during calls to operational teams using AI tools to summarize operational reports.

The latest development of LLMs is called “agentic”, whereby AI assistants work towards a goal by executing tasks autonomously. In agentic mode, the AI agent may not only enhance productivity of human staff but eventually fulfil an entire job description.

The world model

A third category of AI models is emerging, sometimes called “world models”. AMI Labs, where Yann LeCun is an Executive Chairman, has just raised a US$1.03 billion seed round to build the next generation of world models.

These models aim to represent how the physical world works. Rather than predicting words or economic reactions, they attempt to learn the physical relationships between objects, actions, and consequences in real environments. The learning process is similar to children observing the real world and figuring out the underpinning laws of nature. For humans, learning the world (e.g. if I drop an egg on the floor it breaks) happens in parallel to learning a language. Both learning experiences help build human’s awareness and consciousness.

In an aviation world, the ambition of world models is to have machines answer questions such as: How will weather patterns affect this flight trajectory? Or what are the knock-on effects of a disruption across an airline network? The answer comes from advanced simulations based on granular observations, not from language reasoning. This type of modelling moves beyond pattern recognition and forecasting toward causal understanding of systems and laws of nature.

Combined with airline operational data and digital twins, such models could provide powerful tools for decision support in airline operations, as mentioned in a recent article. Unlike with language models, humans are not ready to delegate authority to world models for critical operational decisions. Human beings remain accountable, for now.

What models do you need?

I gave these three examples of AI models because it is tempting to use one (for example language) and to apply it to any use case (where typically another model would work better). It is fair to represent AI models as a family of tools with their respective objectives, learning sets and architectures.

In the aviation domain, each model captures a different dimension of the airline ecosystem. They help interact with customers through language, optimise offer creation with market knowledge and operate efficiently within complex physical systems involving weather, aircraft, crews, airports, and air traffic management.

With the recent progress of AI and its implementation, the question is no longer whether to adopt artificial intelligence, but to prioritise topics and to choose the right approach, models and partners. The most advanced organisations may eventually combine multiple AI models that interact between themselves, under human supervision, for some time.

Conclusion: Next steps for your AI project

For companies (airlines, airports, etc.) starting their journey with AI models, the path could begin with three usual practical steps, from ideation to action.

  1. Set a clear goal and select the relevant AI model. Pick a concept to prove, something not done before. For example, predicting a type of situation and showing how to avoid it, or delivering a type of service and automating it completely. Then select the type of model – market, language or world – that makes sense.
  2. Collect and organise your data. Any AI project depends on reliable data. Airlines already generate massive volumes of operational and commercial data. This data needs to be collected, cleaned, and structured to power AI models. Data preparation is the foundation for future AI capabilities, and each type of model requires their own data and training.
  3. Build. Once the goal and data are in place, build the AI model, test it with your data, and learn from the results. Improve through iterations. Eventually showcase how each model works better for each use cases.

Artificial Intelligence has entered the era of large models and is writing the latest chapter of airlines’ journey to make travel simpler.

There’s still time to join us next week at Aviation Festival Asia 2026!

For more insight from Eric, see:

–

[1] Claude: https://claude.ai

[2] AMI Labs: https://amilabs.xyz

[3] Large Market Model : https://www.fetcherr.io/technology

[4] Nvidia GTC 16-19 March 2026: https://www.nvidia.com/gtc