BITKRAFT Ventures

The broader takeaway from the session: The next generation of AI products may not be won by the best model alone. They may be won by whoever builds the strongest system around that model. Full talk here: youtube.com/watch?v=0HwpkT…

This is where crypto x AI becomes most compelling: when there is a real coordination problem underneath the product. For infrastructure, that can mean coordinating data, incentives, verification, and ownership. For applications, it can mean enabling more open, persistent, and user-shaped digital worlds.

In both cases, what matters is not just model capability, but the system built around it. For Oro, that means building the data layer required to train the next generation of multimodal systems. For SEED, it means building the memory and simulation systems that make AI behavior feel real over time.

At BITKRAFT, one of the ways we think about crypto x AI is through the square and the tower: networks vs. hierarchies, open systems vs. closed ones (h/t to @PonderingDurian) As new stacks form, modular ecosystems often take share over time. In AI, that may mean the pendulum is starting to swing back toward the application layer: 🧵

Once characters start to feel more lifelike, player expectations rise immediately. They should remember prior conversations. They should not react to the same information twice. They should interpret events differently depending on who they are and what they know. That is where simulation starts to get interesting.

SEED points to a different frontier: The hard part of AI-native characters is not just generating dialogue. It is enabling them to carry memory, continuity, and context through a persistent world.

That is especially true where centralized datasets are thin or nonexistent. Oro’s view is that decentralized coordination can be uniquely effective at sourcing and verifying this kind of data, especially in low-resource languages and multimodal settings where the public internet falls short.

Oro’s core view is that frontier AI is becoming increasingly data-constrained, not just compute-constrained. If the next generation of models is voice, video, robotics, and embodied AI, the bottleneck shifts toward access to high-quality, private, long-tail training data.

Two BITKRAFT portfolio companies illustrate different parts of that opportunity set particularly well: @getoro_xyz is building the private intelligence layer for multimodal AI and RL environments. @KlangGames SEED is exploring what AI-native simulation can look like at the application layer.

That framing matters because the AI stack has spent the last few years disproportionately rewarding infrastructure. But technology tends to move in cycles: infrastructure gets built, applications emerge on top, those applications hit bottlenecks, and that demand shapes the next wave of infrastructure.

Thrilled to congratulate the @lostlakegames team on revealing Vaunted — and what a reveal it is! 🎮🚀 We've long believed in this team and watching them step into the spotlight with something this ambitious and polished is exactly the moment you invest for. The galaxy of Vaunted is just getting started and we can't wait for players everywhere to dive in. 🌌⚔️ Wishlist on Steam. You'll want to be early on this one. 👇

@JensHilgers talks AI, world models, games and esports on @etnshow 💪 Watch the full interview: youtube.com/live/YoeTUDIfi…

BREAKING: Jens Hilgers (@JensHilgers), Janitor and Founding GP @BITKRAFTVC is joining us tomorrow at 1PM GMT

The BITKRAFT Report. Don’t just read it. Experience it. reports.bitkraft.vc/r/consumer-app…

We believe: Game-design intuition + AI-native infrastructure wins. Inside the report: • 6-category sector analysis • Framework for building durable moats • The enabling infrastructure stack • Predictions for the next 10 years

The consumer tech landscape just hit a massive inflection point: In 2025, non-gaming apps officially out-earned games for the first time. 🧵