The 4 narratives shaping the future of AI
Vous souhaitez lire en français ? → Cliquez ici
AI Wars — Episode 1
Disclaimer: AI news moves faster than any commentary can follow. Investor sentiment now swings within hours. On November 21, markets opened betting that Nvidia’s exceptional earnings might delay — or dispel — the risk of a short-term correction. By afternoon, the Nasdaq had reversed sharply. A reminder of how fragile the equilibrium remains.
We have no idea what next week will bring. But the analysis that follows should hold regardless of where the next headline shifts the market.
Let’s get to the point.
Artificial intelligence has the potential to reshape the economic order. It also carries profound social, ecological, and geopolitical consequences. At Asterion, we find it impossible to approach this technology without acknowledging its dual nature. The philosopher Bernard Stiegler called such objects pharmaka — forces that heal and disrupt simultaneously.
This is why we’re launching AI Wars, a two-part exploration of where AI narratives come from, what they reveal, and how they influence investment, regulation, and public imagination.
In this first episode, we examine the four major stories battling to define what AI “is.” Next month, Episode 2 will explore how a new generation of European AI startups focused on real-world impact is emerging.
The moment narratives outrun technology
Over the past three years, AI has stopped being a single field. It has become a cultural arena. What we call “AI” is no longer a set of models or techniques: it’s a cluster of competing interpretations, each offering a different future and a different risk profile.
A pattern becomes obvious as we review AI startups across life sciences, energy, mobility, and infrastructure:
Technology moves forward in steady increments. Expectations shift abruptly.
A breakthrough paper changes the mood of an entire industry. A disappointing demo triggers a wave of skepticism. A new moonshot reframes the horizon overnight.
The last few weeks delivered all three.
Jeff Bezos unveiled Project Prometheus, an ambitious bet on machine-based scientific reasoning.
Google’s Gemini 3 outperformed ChatGPT-5 across key benchmarks, triggering the first serious “OpenAI might actually be in trouble” vibes.
And Yann LeCun left Meta to work independently on world models, arguing that today’s LLMs will never produce real intelligence.
His verdict was blunt:
A cat can remember, understand the physical world, plan complex actions, and do some level of reasoning — actually much better than the biggest LLMs.
Technology evolves. But the stories surrounding it mutate faster — and they pull capital, regulation, and perception with them.
To make sense of this turbulence, we’ve mapped four narrative families.
Narrative landscape 2022–2025: how we got here
2022 — The year of enchantment
ChatGPT’s release turned a research breakthrough into a global cultural event. AGI, long a marginal concept, suddenly felt plausible. Investors rotated portfolios around the promise of a cognitive revolution. It was peak techno-euphoria.
2023–2024 — The rise of fear
The pause letter marked a turning point. Geoffrey Hinton publicly voiced concerns. Regulators scrambled to anticipate risks that didn’t yet exist. In Brussels, safety became the dominant lens. In Washington, intelligence agencies entered the picture. The tone shifted — excitement gave way to caution, even dread.
2024–2025 — The skeptical counterwave
Beneath the headlines, harder realities accumulated. Compute costs ballooned. Margins evaporated. Many “AI wrappers” revealed fragile business models. Analysts like Jerry Neumann warned that value would not accrue to model builders, but to industries that use AI to transform logistics, energy, healthcare, or defense. Venture capital quietly became more selective.
Late 2025 — A triangle of hype, threat, and realism
Today the equilibrium is unstable.
Prometheus reactivates the transhumanist imagination. Safety concerns still shape regulation. And markets show signs of doubt: Nvidia’s recent sell-off, despite record earnings, reminds everyone that financial gravity still applies.
We now live in a period where no single story dominates — and that makes the landscape harder to read.
Story 1
AI as the next cognitive species
This is the most ambitious narrative. It frames AI as a scientific and civilizational turning point — a new form of intelligence that will eventually surpass ours.
Vocabulary: AGI, self-improvement, world modeling, discovery engines.
Champions: Demis Hassabis, Sam Altman, OpenAI, DeepMind, Anthropic, and now Bezos via Prometheus.
Its influence is strongest on frontier research and long-horizon capital.
Its blind spot is material constraints: energy supply, semiconductors, thermodynamics. Even if intelligence scales, infrastructure may not.
Still, this story shapes ambition. It attracts talent, captures attention, and justifies multi-billion-dollar bets.
Story 2
AI as an existential threat
This narrative also believes AGI is coming — just not safely.
It’s driven by existential-risk researchers, many policymakers, and parts of the scientific community. In 2023–2024, it became the dominant force shaping regulation, from the EU AI Act to White House executive orders.
Its logic is precautionary: even a low-probability catastrophe requires intervention.
But it has a structural flaw: it often constrains smaller innovators more than the actors who actually possess frontier capabilities. Safety becomes a moat. And it frames the future as something to be controlled rather than built.
Still, it is a critical counterweight to uncritical optimism.
Story 3
AI as an industrial tool
This is the most grounded narrative.
It treats AI as a lever for productivity, not a metaphysical breakthrough. It’s the worldview of Satya Nadella, enterprise CIOs, and the teams integrating AI into logistics, healthcare operations, rail networks, energy grids, and industrial maintenance.
Most measurable value today comes from this story: automation, copilots, scientific workflows, optimization systems.
Its weakness: commoditization. Products depend on models they don’t control. Defensibility relies on data and integration, not hype.
But it is the story closest to the real economy — and the one we see daily in the startups we assess.
Story 4
AI as a bubble waiting for correction
This story argues that generative AI has reached a functional plateau — and that many business models in the field are economically fragile.
It draws strength from:
Analysts noting capital flows to model builders while value accrues to industries that use AI.
Energy bottlenecks: data-center expansion outpacing grid capacity.
Market volatility: Nvidia’s sharp swings despite exceptional fundamentals.
Cultural critique: the MIT Technology Review comparing AGI belief systems to “the most consequential conspiracy theory of our time.”
This narrative explains something the others don’t: why so many AI products impress but fail to generate durable revenue.
How these four stories shape markets, regulation, and perception
These narratives compete daily to define priorities, valuations, and regulation.
Prometheus pushes investors toward moonshots.
LeCun’s departure energizes researchers seeking alternatives to LLMs.
Model failures strengthen the doomer camp.
Adoption metrics reinforce the builder narrative.
Market volatility bolsters skepticism.
This constant drift of narrative gravity explains why AI feels simultaneously overvalued and undervalued, overhyped and underestimated.
Without these stories, the sector looks chaotic. With them, its movements become legible.
Where we stand at Asterion
We don’t subscribe fully to any of these four stories, but we take all of them seriously.
We don’t believe AGI/ASI is imminent, but we acknowledge rapid improvement. We see legitimate risks, but not an existential timeline. We see real productivity gains, but not infinite ones. We note speculative excess, but not a collapse of the underlying trend.
Our perspective is shaped by what we see in the field: companies solving real problems in energy, biology, transportation, and industrial systems. These domains don’t rely on speculative breakthroughs — they require robustness, data quality, regulatory alignment, and tight integration with the physical world.
That’s where the next durable value will emerge.
Conclusion — staying critical to stay rational
Bubble or not, AI is not going away. The dot-com crash didn’t extinguish the internet; it clarified it. The same pattern is likely here.
Clarity requires distance. It requires reading grand narratives without being absorbed by them. It requires distinguishing technological promise from metaphysical projection — especially as AGI rhetoric becomes more extravagant.
When the MIT Technology Review describes AGI discourse as a “conspiracy theory,” it reveals how culturally loaded these narratives have become. Understanding them is the only way to navigate them.
This is what AI Wars — Episode 1 set out to do: map the stories, not the code.
Next month: Episode 2 — AI for real world impact
In the next installment, we’ll explore the conditions under which a new generation of European AI startups delivering real-world impact can emerge — companies with measurable benefits, industrial relevance, and an energy footprint compatible with planetary limits.
Because if AI is a pharmakon, the question is no longer whether it will transform our societies, but how deliberately we choose to shape that transformation.
A quick look at three companies in our portfolio currently raising — all building real-world impact across critical infrastructure, mobility, and biology.
→ Futurail — building the next generation of autonomous, high-frequency rail operations to boost Europe’s train capacity.
→ .omics — using AI-driven trait discovery to engineer climate-resilient seeds at industrial scale.
→ Bubble Robotics — designing autonomous underwater robots to inspect, monitor, and protect critical ocean infrastructure.
If you’re interested in any of these rounds, contact me: antonin@asterionventures.com
Three questions for… Pierre-Louis Guhur, CEO, Argile.ai
Pierre-Louis Guhur holds a PhD in machine learning. He is also the co-founder and CEO of Argile.ai, a startup that uses AI to support professionals working on residential energy-efficiency renovation — and, incidentally, one of our most recent investments. Because in the end, AI may well be a story that is ultimately meant to be written in the physical world.
As a PhD in AI and an entrepreneur, what do you make of the incredible hype surrounding AI over the past three years?
Honestly, it’s been a fascinating ride! When I started my PhD back in 2019, several professors actually advised me not to go into AI. For them, the field had plateaued and nothing major was expected anytime soon.
And then, almost overnight, everything shifted. GPT-2 came out, then GPT-3. In 2020 its release shook the research world. Sure, it hallucinated and it got things wrong, but it was already doing things that felt out of reach just the day before. And in December 2022 the public discovered ChatGPT. Say what you want about it, but it was one of the best communication operations tech has ever seen.
Very quickly though, people assumed we’d already hit the ceiling again. “There is only one Internet,” as Ilya Sutskever famously put it, and the models seemed to have already consumed it. But progress didn’t stop. New models proved that AI could learn to solve new problems, not just repeat what it had seen. And that’s exactly what François Chollet calls intelligence: the ability to adapt to the unknown. That’s the shift. LLMs are no longer just fancy ZIP files of the web — they can reason.
So today, the real issue isn’t whether the models are good enough. They are. The bottleneck has moved. The real question is what we do with this capability and how we inject it into the real economy to reshape entire markets. In the end, what will matter are the use cases, not the model rankings.
Looking at things calmly, what applications can we realistically expect to scale in the short and medium term?
At this point, comparing one LLM to another isn’t really the point anymore. Gemini pulls ahead, then Claude takes the lead back the week after… it’s fun, but it’s not where the real momentum is.
Where things are getting exciting is elsewhere. Take 3D reconstruction. It opens up incredible possibilities: rebuilding spaces, generating environments, simulating physical constraints. Yann LeCun often says humans don’t think mainly in words, but in images. And once AI starts reasoning in 3D, it gains access to much richer data, better intuition about physics, and you can even let agents experiment in a world that looks like ours. You also have big movement in diffusion models applied to weather, and causality research has accelerated a lot recently.
But again, technology isn’t the judge here, usage is. The question is really how entire industries are going to shift.
There’s a simple way to know whether a sector is exposed to AI disruption. Imagine you suddenly have unlimited interns working for free. What happens? If everything changes, then AI will change it too. Customer support? Absolutely. Strategic consulting? Likely. Installing a heat pump, much less. Software development sits in the middle. Interns help, but you still need strong senior oversight.
From there, it becomes a strategic choice. Either you build an AI-powered SaaS and transform an industry from the outside, or you go full-stack AI, acquire a company and rebuild it from the inside with AI at the core. The second option works when consolidation is possible, like in legal tech. When a market is fragmented — like energy renovation — SaaS makes more sense. And that’s the path we chose at Argile.ai.
And how does all this translate into what you’re building with Argile.ai?
At Argile.ai we support professionals through the entire renovation chain: lead qualification, data capture, technical sizing, pricing and even project follow-up. It’s a difficult industry. Producing a quote can take a month, and a single mistake on a €100k project can put a small company at risk. We want to make renovation faster, more reliable and more ambitious. We’re a mission-driven company and we’ve set a clear target: saving one million tons of CO₂. To get there, we need contractors to aim higher than the regulatory minimum.
Our journey hasn’t been linear. In the beginning we tried the 100% LLM approach. We thought a chatbot could do everything. It didn’t work. So we built a strong product base instead, merging what already existed on the market, but improved. At that point AI was mostly helping us with the code. Shipping half a million lines with six engineers in two years simply wouldn’t have happened otherwise.
Later we brought AI back to the frontline — at the right place this time. 3D reconstruction to replace endless forms, automated note-taking during technical visits. On average, contractors save about an hour per visit and can focus on advice and sales instead of admin.
Now we are entering another phase. MCP, the Model Context Protocol, released late 2024, lets us build translators between LLMs and APIs. Thanks to that, Argile.ai connects to Anthropic’s Claude and soon users will have built-in AI applications directly in the interface. We are in beta today. 2026 will be the year we roll this out to smaller firms on the market.
In short, the race is on. Either we build our own user base and distribution, or we become just another icon inside a GenAI AppStore. The question is no longer whether AI will transform this industry, but who will be ready when it does. We know where we stand.
AGI as conspiracy theory? (MIT Technology Review)
MIT Tech Review argues that AGI has spread less like a technology milestone and more like a conspiracy narrative: vague, unfalsifiable, and endlessly adaptable. From early fringe thinkers to Yudkowsky, Thiel, DeepMind, and OpenAI, the idea gained legitimacy through a mix of utopia, doom, and strategic hype. With no clear definition and no evidence it’s imminent, AGI still drives massive investments, policy debates, and industry direction — a story powerful mainly because it’s always “almost here.”
→ Read on MIT Technology Review
Is the AI boom already a bubble? (Crazy Stupid Tech)
Fred Vogelstein argues the current AI frenzy mirrors the 1999 internet bubble — only larger, faster, and far more concentrated. AI capex and VC spending could exceed $1.5T in 2025, with Big Tech burning $70–100B a year on data centers despite unproven economics. Leverage and circular vendor-financing deals amplify the risk, while OpenAI’s cash burn and China’s advance add systemic pressure. The core point: AI will be transformative, but we’re years too early — and the bubble is likely to burst before the returns materialise.