Welcome to French Tech Updates! Your weekly source of startup, VC, and tech news and insights. I’m James, a startup-obsessed American living in Paris.

“At $500 billion OpenAI just became the world’s most valuable private company—valued higher than 96% of the Fortune 500” 

“Seven months ago Lovable didn’t exist, today they bring in $75M ARR.”

“ElevenLabs doubles valuation to $6.6B in nine months.”

These are just a few AI-related headlines from the last few weeks, the latest sign of an industry moving at a speed that can make your head spin. Even as a consumer it’s hard to keep up. For investors, whose jobs depend on correctly identifying the best of these companies before anyone else, the challenge is existential.

On Monday, two of these investors, Briac Lescure from Daphni and Cyril Bertrand, CEO and Managing Partner of XAnge, shared what’s changing for investors and founders, including:

• 🚀 Where (and where not) to build an AI company in 2025

•  🔎 What the AI bubble means for investors

• ⚙️ How they identify winning AI companies

In this special edition of French Tech Updates I’m sharing all of my notes from the session. There’s something here for founders, investors, and the generally curious-minded, so I hope you enjoy.

With that, let’s jump in!

10x speed and 10x valuations

Two ways to think of AI’s role are as an enabler 💁‍♂️ and an accelerator 🏎️. In its enabler role, AI cracks problems that were once impossible or uneconomical. Some of these, like predicting the structure of all known proteins, as AlphaFold did in 2022, were nearly unsolvable without AI assistance. Others were simply unprofitable with existing tech and therefore not pursued.

As an accelerator, AI is increasing the speed of everything—compressing the startup timeline and allowing founders to build, launch, and get customers in days instead of months. As a result, the minimum startup speed limit has been raised and moving faster has become the new requirement rather than a differentiator.

When AI’s enabler and accelerator roles combine, €1M in recurring revenue for a months-old startup isn’t as impressive as it once was.

We have entered a new paradigm for builders and investors alike—and seemingly everyone is eager to participate.

The AI Valuation Bubble

Cyril and Briac aren’t shy about the state of AI investment: we are in a bubble. 

But what kind of bubble matters. Cyril, who cut his teeth as an investor through some of the largest hype and bust cycles in tech history, stressed that the current bubble is affecting valuations, not underlying tech itself—which still delivers very real outcomes.

The results is a risk-laden paradox where three contradicting truths intersect:

1. Valuations are wild

2. There will be big winners

3. Opting out isn’t an option

In principle, during a bubble everyone should stop investing, but as Briac mentioned, the bigger mistake investors can make now is one of omission. Investors know they’re overpaying and still can’t afford to sit on the sideline—hence the larger and larger rounds for companies like Anthropic, OpenAI, and Mistral.

In many ways this is the job of a good VC, to differentiate between hype and reality and to bet big despite the noise. The power law nature of VC returns enables this dynamic, but when one big win can overshadow a dozen losses, sitting on the sidelines isn’t an available option.

Barbell startups and value creation concentration

The expanding valuation bubble in AI isn’t the only thing Briac and Cyril see changing. AI is also concentrating value creation and capital allocation patterns.

Across the market, value creation concentration is increasing, especially for giants like OpenAI who can use their own powerful tech to forcefully expand into additional markets. A supercharged whale like OpenAI entering your industry can spell overnight death for existing competitors, but more on that later.

Deployment of capital is also becoming more polarized. The number of deals has fallen year-over-year while, at the same time, more money is being funneled into the hottest rounds. Instead of eliminating the need for funding, the AI wave has changed where it flows—concentrating more capital into fewer hands.

On the other side of this barbell there are some lean teams who are raising less and still delivering interesting results thanks to AI accelerating their roadmaps and reducing resource demands.

No matter where a company sits on this spectrum, fiercer competition is increasing churn and compressing margins across industries, even as benchmarks for “impressive” growth are raised continually higher.

How to be a VC in 2025

Unsurprisingly, it’s not just the startups that have changed. The role of a VC has radically transformed in the last 10 years—especially in the last three.

The old inbound model where founders hunt for VC attention to look at their pitch deck is dead. Instead, top firms are now doing the hunting. Instead of sorting through decks they’re building outreach automations to build relationships with promising founders even before they’re actively raising.

According to Cyril, “It’s almost completely changed. We are selling constantly. You have to build a sales machine or you end up eating the crumbs of the deals other VCs are not willing to do.”

To avoid the crumbs and get a slice of the cake, speed has become key. Much like for the startups they’re funding, the minimum speed limit for investors has also been raised. To meet this pace, Briac and Cyril have a few tips:

• Most due diligence needs to happen before the first meeting. Deals close in weeks, and to win investors need build conviction early. This means deep diving on a sector or problem, establishing a solid knowledge base, and then going and finding likely winners in that space.

• Pre-empt founders and reach them before the hype hits. Sometimes, if they’re publicly raising it’s already too late. Keeping up with personal networks, former employees of prior portfolio companies, and proactive outreach can close a round before it even hits the broader market.

AI is also helping increase the pace internally at Daphni and XAnge with multiple AI agents automating well defined tasks like crawling data, performing deep research, and preparing investment memos.

Tips for a winning pitch

For founders, the new game is becoming visible. They need to make themselves easy to find and get on the radar of VCs instead of bothering with cold outreach. Once in the meeting, winning pitches still have a few common characteristics:

1. 🔎 Clarity. The best founders can explain what they are doing clearly—including their vision and their path to get there. For example, when payments unicorn Lydia was just getting started they knew that they could dominate their initial market (university campuses) by reaching 30-50% adoption on campus. From there, every action aligned towards that goal.

2. 💎 Initial proof. A vision and a path to get there are necessary but not sufficient. Successful founders also need to show their progress down that path—whether with product, customers, revenue, or other signs of traction.

3. 🎪 A sizeable market. Not all businesses are VC scale, capable of returning 100X or similar outlier returns. To be backable, the underlying market needs to support €1 billion outcomes and, ideally, be rapidly growing.

Other than these three criteria, early-stage VCs are mostly checking for any red flags not to invest. For pre-seed rounds, this due diligence is largely focused on the founders and consists of deep reference checks

As Briac put it, startups are too complex and too diverse to set too many “green flag” criteria. A fund only investing in repeat founders, for example, would have missed Meta, PayPal, Google, and many other multi-billion successes.

By forming a research-backed thesis, prioritizing founders that demonstrate clarity and initial proof in a large market, and avoiding red flags VCs can continue to pick out winning deals amidst the fog of AI hype.

What’s a real moat in the age of AI?

Jeff Bezos famously said “I almost never get the question: 'What's not going to change in the next 10 years?' And I submit to you that that second question is actually the more important of the two, because you can build a business strategy around the things that are stable in time.”

At a time when things are seemingly changing faster than ever, Cyril and Briac identified these five areas as the ones likely to remain stable in time:

1. People: great engineers and founders remain a very real moat—one that is hard to builder and even harder to maintain.

2. Culture: while invisible, strong culture leads to enduring advantages.

3. Brand: when all else fails, brand is a way to differentiate even commoditized products.

4. Distribution: while often ignored in France, unique and strong distribution is what led to Revolut and Lydia’s successes.

5. Friction: sticking with one set of sticky customers creates resilience that speed can’t match. For example, most accountants won’t change their tools, but once they do you can lock them in.

Where (and where not) to build an AI company

Looking at the AI ecosystem there are broadly four categories a company may fall into:

1. Applications

2. Middleware (agents)

3. Large Language Models

4. Hardware

This kind of structure exists in plenty of other sectors like mobile telecom and computing. What makes AI weird is that, as Cyril put it, “from time to time the LLMs jump to the surface and eat the apps.”

Previously, this behavior rarely happened. It would be like if Salesforce killed off an entire section of apps in their ecosystem overnight. With AI, the stratified splits between infrastructure and apps are not set in stone anymore—leading to overnight death for any application layer companies that find themselves building in the critical path of an LLM company.

Overnight death syndrome isn’t just scaring founders.

“VCs are practically freaking out whenever Sam Altman takes the stage,” Briac added.

As if to prove the point, as I was on my way home from the event I opened my phone to see OpenAI had just introduced app connections inside of ChatGPT to sites like Figma, Expedia, Spotify, and Zillow. In the distance, I could almost hear the screams of countless AI travel apps experiencing overnight death syndrome.

Other than avoiding the critical path of LLM giants, Cyril and Briac had this advice on where not to build a company in 2026:

• 👨‍✈️ Avoid copilots. There is very little differentiation and most of these tools are too shallow to have a lasting moat.

• 🍬 By all means avoid the category of AI wrappers. VCs are already ignoring this space after seeing the LLMs wipe out entire categories of ChatGPT wrapper apps.

• 🙅‍♂️ Avoid the easy path. If it’s easy for you to build it’s also easy for your competitors. The rising wave of AI note takers, AI sales tools, and AI email tools will turn into a red ocean bloodbath as these undifferentiated competitors fight to the death. If it’s easy to build, it’s easy to kill.

How and where to build in 2026

Now you know what to avoid, so where should you build? Briac and Cyril’s answer: do the hard things.

• ⛰️ Take the hard road. As Sam Altman himself wrote in 2023, “It is easier for a team to do a hard thing that really matters than to do an easy thing that doesn’t really matter; audacious ideas motivate people.” Use AI’s enabler role to tackle hard problems never before solved.

• 🔬 Get very specific on a topic and stick to it. Spaces like accounting, payroll, and regulated verticals all have very specific challenges associated with them. Building in these areas lets you carve out a domain too narrow or too complex for the LLMs to copy what you create.

• ➡️ Pursue adjacencies. The billions in AI funding are largely going towards infrastructure and the energy to power it. Hardware, modular nuclear reactors, and electricity companies are seeing a resurgence in demand and a need for further innovation.

• 🧠 Think beyond AI. The hype around the space has captivated the Sauron’s eye of the tech world, but many important sectors are going unloved. Take climate tech for example. In Cyril’s view, there is “almost no chance we escape climate problems in the long term.” This is just one of many ignored spaces that is worthy of attention now. 

If you’re looking for an idea to pursue, here are a few specific ones Cyril and Briac are excited to see built:

1️⃣ A true pan-European accounting solution. This space is very difficult to crack since each country has its own set of complex rules, but the results can be massive. Pennylane spent 6 years building a business accounting solution just for France and that alone was enough to create a unicorn. With AI, conquering all of Europe may now be possible.

2️⃣ Speech interpretation with 99.99% accuracy. People make mistakes when they speak (as anyone whose received one of the minutes-long voice notes Europeans are famous for can attest). Accurately picking through the errors to capture the true intent of speech will build the trust needed to cement voice AI’s place in sensitive areas like payments and healthcare.

3️⃣ A first foundational model for the human brain. Some research is already making progress on this topic. Accurately capturing and modeling human behavior would be a tremendous unlock for whoever can tackle this difficult problem.

And if you do end up creating one of these next-gen AI giants I know at least two VCs who would be interested in funding your seed round.

Podcast Recommendation 🫶

Venture Lab (With Luis) 👨‍🔬

Why you’ll like it: In a world where capital has become a commodity, the question emerges: How can investors provide value beyond funding and help founders build exceptional businesses?

The Venture Lab podcast explores this challenge through conversations with the most disruptive minds in entrepreneurship, venture capital, and media, capturing the fundamental shift occurring in today’s venture landscape. Click below to check out a recent episode with Seedcamp founder and prolific investor Saul Klein.