Blab Business
Local AI, Open-Weight Models, and Crypto Wallets Could Replace SaaS Tools, Says Balaji Srinivasan
When Balaji Srinivasan stepped onto the stage at SuperAI in June 2026, he outlined a future where your laptop, not a cloud server, becomes the brain of your AI tools. The former Coinbase CTO and current Andreessen Horowitz venture partner argued that the gap between open‑weight and closed‑weight AI models has narrowed to just a few months.
Srinivasan said that the billions of dollars poured into training proprietary models now subsidise the open‑weight ecosystem that follows a few months later, making it harder for closed‑model vendors to justify API‑based revenue. Local inference tools such as LM Studio, Docker Model Runner, and Ollama already allow users to download and run models offline, and quantised versions can even operate on smartphones.
If models can run locally, the front‑end of a SaaS application becomes easy for AI to replicate. Visual interfaces can be cloned by feeding screenshots or screen‑recordings into a model, while the real value lies in the back‑end data model, integrations, and state management. The Obsidian‑Claude example illustrates how local Markdown files can be queried by a local AI, surfacing connections across years of notes without sending data to a cloud.
The shift from “app‑centric” to “file‑centric” is enabled by open file formats such as Markdown, mbox, docx, and xlsx. Local AI can compute on these files, making the file itself the primary asset rather than the application that displays it. Export‑to‑data features become a compliance checkbox and an escape route, and developers are increasingly choosing tools that are interoperable with AI agents.
Crypto wallets solve the public‑key infrastructure problem that has traditionally limited decentralised applications. Hundreds of millions of users hold private keys in wallets such as MetaMask, Trust Wallet, and hardware wallets. These keys enable secure multi‑party computation, encrypted messaging, and authenticated packet exchange, providing the identity and encryption layer for local‑first apps.
Srinivasan also warned that cloud security is becoming increasingly indefensible. AI‑powered exploit tools can probe attack surfaces faster than human teams, and a single breach can expose large amounts of data. He noted that blockchains, because they hold real value, receive the highest financial incentives for hardening, whereas private enterprise back‑ends may be more vulnerable.
The fragmentation of social media, described by Srinivasan as a “Tower of Babel” moment, is also accelerating decentralisation. Platforms such as Nostr, Lens, and Farcaster are emerging as smaller, higher‑trust networks, each with its own identity system and communication protocol. This aligns with the “personal, private, programmable” thesis that local‑first infrastructure can operate without a single global platform.
AI also increases noise in communication. The ability to generate realistic spam, scams, and fake messages makes it harder to distinguish genuine outreach, leading users to rely more on trusted networks. Srinivasan predicts that this will create new jobs in verification, attestation, and notarisation.
However, the practical adoption of a local‑AI, open‑file, crypto‑wallet stack faces significant hurdles. The usage gap is large; most users still prefer cloud services for their convenience and real‑time collaboration features. The complexity of setting up local models, choosing from hundreds of open‑weight models, and managing crypto keys presents a high transaction cost that many users are unwilling to pay.
At present, local‑first tools are largely niche. They offer privacy and control but lack the ease of use and network effects of established SaaS platforms. Whether the stack can overtake cloud‑based services remains uncertain, and the next few years will test whether the convergence of local AI, open file formats, and crypto wallets can deliver a compelling alternative.
Srinivasan said that the billions of dollars poured into training proprietary models now subsidise the open‑weight ecosystem that follows a few months later, making it harder for closed‑model vendors to justify API‑based revenue. Local inference tools such as LM Studio, Docker Model Runner, and Ollama already allow users to download and run models offline, and quantised versions can even operate on smartphones.
If models can run locally, the front‑end of a SaaS application becomes easy for AI to replicate. Visual interfaces can be cloned by feeding screenshots or screen‑recordings into a model, while the real value lies in the back‑end data model, integrations, and state management. The Obsidian‑Claude example illustrates how local Markdown files can be queried by a local AI, surfacing connections across years of notes without sending data to a cloud.
The shift from “app‑centric” to “file‑centric” is enabled by open file formats such as Markdown, mbox, docx, and xlsx. Local AI can compute on these files, making the file itself the primary asset rather than the application that displays it. Export‑to‑data features become a compliance checkbox and an escape route, and developers are increasingly choosing tools that are interoperable with AI agents.
Crypto wallets solve the public‑key infrastructure problem that has traditionally limited decentralised applications. Hundreds of millions of users hold private keys in wallets such as MetaMask, Trust Wallet, and hardware wallets. These keys enable secure multi‑party computation, encrypted messaging, and authenticated packet exchange, providing the identity and encryption layer for local‑first apps.
Srinivasan also warned that cloud security is becoming increasingly indefensible. AI‑powered exploit tools can probe attack surfaces faster than human teams, and a single breach can expose large amounts of data. He noted that blockchains, because they hold real value, receive the highest financial incentives for hardening, whereas private enterprise back‑ends may be more vulnerable.
The fragmentation of social media, described by Srinivasan as a “Tower of Babel” moment, is also accelerating decentralisation. Platforms such as Nostr, Lens, and Farcaster are emerging as smaller, higher‑trust networks, each with its own identity system and communication protocol. This aligns with the “personal, private, programmable” thesis that local‑first infrastructure can operate without a single global platform.
AI also increases noise in communication. The ability to generate realistic spam, scams, and fake messages makes it harder to distinguish genuine outreach, leading users to rely more on trusted networks. Srinivasan predicts that this will create new jobs in verification, attestation, and notarisation.
However, the practical adoption of a local‑AI, open‑file, crypto‑wallet stack faces significant hurdles. The usage gap is large; most users still prefer cloud services for their convenience and real‑time collaboration features. The complexity of setting up local models, choosing from hundreds of open‑weight models, and managing crypto keys presents a high transaction cost that many users are unwilling to pay.
At present, local‑first tools are largely niche. They offer privacy and control but lack the ease of use and network effects of established SaaS platforms. Whether the stack can overtake cloud‑based services remains uncertain, and the next few years will test whether the convergence of local AI, open file formats, and crypto wallets can deliver a compelling alternative.
