The Long-Term View: Why Id4 Ventures Bets on Breakthroughs, Not Shiny Things.
So proud to see our longtime bets taking shape.
From the beginning, we chose Id4 Ventures to focus on fundamental technologies that will truly change our lives and economy, and not to chase the next shiny, transient trend.
Building Deep Tech is hard. Our job is to lighten that pain for our founders—to provide the moral and financial conviction to be there when others doubt. This is only possible thanks to the long-term trust and support of our incredible LPs. Nothing would be possible without you.
Our Thesis in Action: Six Years of Generational Shifts
Six years ago, we began investing with the conviction that AI would be a generational shift, requiring an entire technological infrastructure rebuild and entirely reshaping industries.
Today, we are seeing that deep thesis materialize with companies like:
@gensynai : Building the computational backbone for a decentralized AI future. (Series A led by @a16z , > 50M transactions in 5 months on the testnet)
@pathway_com : launching a new “post-transformer” architecture that paves the way for autonomous AI.. (https://t.co/zcl7WJhOpx)
But our conviction goes beyond AI infrastructure, touching human impact and new frontiers:
Saving lives with @ThinkSono
Automating scientific R&D with @lumi_systems
Pioneering durable manufacturing with Tetmet (https://t.co/4t3aYmeLv6)
The Next Chapter: The Space Economy
For the past two years, we’ve begun writing a new chapter based on the same philosophy, backing foundational players in the emerging Space Economy:
@OrbitalParadigm : Seamless logistics from Space to Earth
Gama (https://t.co/pitdqAyg17) The most effective way to deorbit.
...and more announcements to come.
To all our founders: We can’t appreciate enough the sacrifices and sheer grit you take on every day. You are the real stars. We will always strive to be worthy of your trust and partner with you to improve our world.
The real stars: @harrygrieve@benfielding@zuzanna_pathway@FouadAlNoor1@silas_adekunle@TomVroemen
We just welcomed @id4vc ventures into our investor pool!
We're thrilled to build with @hervecuviliez, @IvanPetrovic and all the team.
Id4 started investing in AI 6 years ago, when only a handful of companies were talking about it. They saw early on what the AI revolution would demand: a full infrastructure rebuild, high-performance cloud, grounded in rigorous science and a strong entrepreneurial spirit.
Excited for what's coming
Paris-based startup SquareMind recently raised $18M to launch Swan, the world’s first robot designed to capture standardized, full-body dermoscopic skin imaging.
The funding round was led by Sonder Capital, founded by Fred Moll, the “father of surgical robotics” who built Intuitive Surgical’s da Vinci system in the 1990s.
Swan uses a robotic arm to scan a patient’s entire skin surface in minutes, paired with AI software that tracks new or changing moles across visits.
The tech addresses a critical bottleneck in dermatology, where months-long waitlists collide with the fact that 80% of melanomas are new lesions.
Commercial launch in the US and Europe is expected later this year.
“We have not yet had a PageRank moment for intelligence.”
We’ve got so many comments and questions about this statement delivered by @adrian_pathway during our recent Transformer vs Post-Transformer debate with @lukaszkaiser@YesThisIsLion@mlech26l - thanks!
Let’s dig into it. In the 1990s, web search already existed. We could index information. AltaVista existed. The web was growing fast.
Then PageRank happened.
That moment combined three things:
1. A simple but deep mathematical idea: treat the web as a giant graph and compute a stationary distribution of a *random walk* on that *graph*
2. A scalable implementation: large-scale graph computation on huge clusters
3. A company that integrated and scaled the idea end-to-end: Google
That combination gave search a much clearer center. It stopped being just a pile of heuristics and started to look more like: here is the mathematical object we need to compute, now let’s build the systems needed to compute it well.
Adrian asked Lukasz Kaiser directly whether he sees a PageRank-like idea inside the
Transformer. Lukasz said no.
For intelligence, we still do not have that kind of unifying operator or process. We do not yet have an agreed mathematical object that says: this is the core computation behind it.
That missing unifier is what Adrian meant by the absent “PageRank moment for intelligence.”
That is also the main idea behind our work on BDH, our Post-Transformer architecture. We are after that fundamental “platform discovery” for intelligence.
The full Transformer vs Post-Transformer debate is a good place to go deeper on these topics. Link below.
Welcome to the id4 tribe! We are thrilled to work with @benrey0302 & Arthur Chevalier . What they are building at https://t.co/JQOelDuLeZ is amazing and becoming vital if you want to be able to manage your cloud infrastructure in the age of AI
Id4 ventures proud investor 🦾
We just welcomed @id4vc ventures into our investor pool!
We're thrilled to build with @hervecuviliez, @IvanPetrovic and all the team.
Id4 started investing in AI 6 years ago, when only a handful of companies were talking about it. They saw early on what the AI revolution would demand: a full infrastructure rebuild, high-performance cloud, grounded in rigorous science and a strong entrepreneurial spirit.
Excited for what's coming.
Last week’s Post-Transformer debate post raised one question: Can long term memory become part of the architecture?
It points to one promising mathematical idea behind Post Transformer AI: Linear attention in high dimension with persistent state.
In a standard Transformer, memory is handled through caching context.
The model keeps previous keys and values in small dimension d, then attends over them. But this is still token history.
BDH (Dragon Hatchling) – one of the Post-Transformer architectures, takes a different route.
The paper describes BDH's state space as fixed and large, with the macro interpretation of associative memory, like KV cache, but organized differently.
Each layer has a persistent state matrix: ρₗ ∈ Rⁿˣᵈ
Here:
n = neuronal or concept dimension
d = low rank synaptic dimension
d << n
The key idea is that state is aligned to neurons, in high dimensional space (n in the order of billions).
A Transformer stores token history.Whereas BDH-GPU (a tensor-friendly version of the BDH architecture) evolves state, similar to State-Space Models.
This is where the brain analogy becomes useful. The brain does not append every experience into a longer transcript. It has a large bounded substrate of neurons and synapses, where experience changes connections sparsely and with high parallelism.
BDH GPU expresses a related idea computationally:
not memory as a longer context window,
but memory as a large, evolving internal state.
Why it matters:
– no Transformer style hard context window. practically enabling a infinite context window in a reasoning model.
– linear attention in a large neuronal dimension
– sparse positive activations
– persistent state instead of only token history
The deeper insight:
Long horizon reasoning may not come from storing more tokens.
It may very well come from better state dynamics.
So thrilled to see @TomVroemen vision going live! Reindustrialization in progress 🦾🦾
@id4vc we love physical Ai and so proud to back https://t.co/RnuEjEVKQA
Pathway and AWS use the term "Sticky Inference" to refer to the part of the AI stack where context compounds. It shows up in the use cases tied to your proprietary enterprise data, where a clear moat emerges as the model keeps learning from the business it serves.
@AWSstartups just published a blog on long-horizon reasoning and continual learning for enterprises, and how Pathway's Dragon Hatchling (BDH) is delivering on both.
The article covers six use cases across healthcare, financial services, retail and more, showing how the Post-Transformer architecture moves from research to production AI workflows.
One deep learning debate every AI researcher should care about: Transformers vs Post Transformers.
At the surface, it sounds like an architecture fight. Mathematically, it is about scaling laws, memory, online learning in frontier models, and hardware limits.
That is what made the recent debate interesting. It featured @lukaszkaiser, @adrian_pathway, @YesThisIsLion, and @mlech26l, hosted by @zuzanna_pathway.
Transformers won the last era because multi head self attention scales empirically and fits the hardware ecosystem extremely well. But the next bottleneck may be different.
Full self attention has O(n²) compute pressure with sequence length. Transformer LLMs do not natively have persistent long-term memory. RAG retrieves. Longer context conditions. Neither necessarily forms new reasoning patterns inside the model.
That is why continual learning is becoming central, recently covered by @a16z.
The open questions:
– How can models learn after deployment without catastrophic forgetting?
– How can long term memory become part of the architecture?
– How can models reason over longer horizons without paying infinite context costs?
– How can hardware and AI architectures co-evolve more efficiently?
– And, are we chasing the right benchmarks with these goals in mind?
These questions were tackled head on, with counters from @lukaszkaiser, Transformer co-inventor and core contributor to ChatGPT and GPT models.
The image below summarizes some notes from the 80 minute debate.
What comes after the Transformer?
Zuzanna Stamirowska puts the debate out in the open, with the very inventors of Transformer and Post-Transformer architectures!
Watch the 5-minute highlights. Follow @zuzanna_pathway and hit the bell, full fight drops tomorrow.
I largely agree with @YesThisIsLion on this.
The biggest mistake right now is expecting the first Post-Transformer models to beat Transformers on day one by delivering massive gains on irrelevant axes.
Said it would be a real fight. IT WAS. 🥊
@adrian_pathway: “Transformers think in language. They do not think in latent thought.”
@mlech26l: “I am convinced that the Transformer will find its own replacement.”
@YesThisIsLion: “Lukasz is going to be correct up until that day, and then he is going to be wrong forever.”
@lukaszkaiser: “Do not be scared of being 50-times slower!! If you show me a model that is 50-times slower but on a better slope, you win.”
Good thing I told them to keep it clean, look at them! 😂
Transformer Vs Post Transformer: Deciding Round, By @pathway_com
Datadog's State of AI Engineering 2026 is out:
→ 5% of AI requests fail in production
→ 60% of those failures = capacity limits, not model quality
→ Agent framework adoption doubled YoY & complexity too
Datadog CPO Yanbing Li: "The companies that win won't just build better models, they'll build operational control around them."
Vercel CEO Guillermo Rauch: "The next wave of agent failures won't be about what agents can't do but what teams can't observe."
He's right. We just think the next step isn't seeing it sooner, it's seeing it before it happens.
On it!
Congratulations to the winners of the @GensynFND <> @ETHGlobal Open Agents Hackathon - Best Application of AXL
• 1st place - Dromeus (@deveshcodes_)
• 2nd place - Pythia (@HarshitNay80531)
• 3rd place - AXL Open Telemetry (@metroxe)
Find details of their submissions below.
The post-transformer era is here. We’ll shape it together.
Heavens! What a fight!🥊 Last night in San Francisco we brought four of the inventors building today’s and tomorrow’s AI architectures into the ring for the deciding round.
A year ago, when we said the post-transformer era was coming, most people saw it as just a research concept.
Yesterday the room was packed with people who flew in because they felt the shift is real. Research peers from OpenAI, DeepMind, Anthropic, xAI, and NVIDIA. Leaders scaling AI at the largest banks (Goldman, BlackRock, Visa, First Citizens, Merck) and internet companies (Google, Meta, Apple, Microsoft, LinkedIn, Salesforce, Walmart, Waymo). AI investors and diplomats. And founders of other deeptechs in the space.
Beyond the stage, what stood out was the willingness to question defaults, surface real disagreements, and keep pushing toward better answers. For this field, and for the people who will build on what we make.
Thank you to the fighters who brought that energy:
@adrian_pathway (Pathway), @lukaszkaiser (OpenAI), @mlech26l (Liquid AI), @YesThisIsLion (Sakana AI). You did give me a "good clean fight for the AI Champion Title" 🤣.
And to @dexhorthy for being my co-moderator that this electric ring needed. Big thanks to our friends in the Bay for spreading the word.
I’ll share more over the coming days — there's a lot worth pulling out of those few hours.
📍Transformer vs Post-Transformer: The Deciding Round
“I don’t think Transformers can do it.”
The fight is on!
@zuzanna_pathway is putting the inventors of the Transformer and the Post-Transformer era — in a literal boxing ring in San Francisco!
Full card: @lukaszkaiser, @adrian_pathway, @mlech26l, @YesThisIsLion
Who wins?🥊
The missing half: space logistics!
Exactly @id4vc thesis with @OrbitalParadigm
Massive market in the making , amazing team led by @caccia83 = winner in the making 🚀🛰️
https://t.co/wKGSBYX2uY