"A.I. Brainiacs" is a Puck article by Ian Krietzberg, published January 15, 2026. It is a dispatch on researchers and startups developing brain-inspired AI architectures as alternatives to the industry's dominant scaling paradigm.
Summary
The article frames the dominant approach to AI development through what it calls the scaling "commandment" — "Thou shalt seek scale" — and argues that it is running into four converging problems. The first of the four is capital: the investment required for frontier training runs is described as increasingly prohibitive for all but the best-capitalized labs. The second is physics and energy: the article states that the US electrical grid cannot support the level of demand the AI industry believes it needs, a constraint connected to the broader Nuclear PPAs for AI and Data Center Siting / AI Power Politics dynamics. The third is chip shortages: supply constraints on GPUs and specialized accelerators limit who can scale. The fourth is community resistance: local and state opposition to data center proliferation is described as growing.
Against this backdrop, the article profiles researchers and startups pursuing architectures intended to achieve greater model efficiency without sacrificing performance — aiming, in the author's words, to "scuttle the notion that scale is the industry's true North Star."
Brain-inspired alternatives
The article describes three brain-inspired approaches. The first, neuromorphic computing, uses chips that mimic the brain's spiking neural architecture rather than continuous-activation artificial neural networks; the author characterizes these as more energy-efficient in theory but significantly harder to program and train on. Spiking neural networks (SNNs) fire only when a signal exceeds a threshold, as biological neurons do, rather than computing continuously, which the article says yields much lower energy use per computation while remaining substantially behind transformers on benchmark performance. Liquid neural networks are continuous-time, variable-weight networks inspired by the nematode nervous system (C. elegans); the article describes them as developed at MIT CSAIL in Daniela Rus's lab and commercialized by Liquid AI, with claimed advantages of smaller model size, interpretability, and real-time adaptability.
The article poses what it presents as a fundamental question — "Do we need to figure out how the brain actually works?" before brain-inspired AI can work well. Some researchers are described as arguing that the field has learned enough from neuroscience to build useful architectures, while others argue that current brain-inspired models remain too far from actual neural computation to deliver claimed efficiencies at scale. The article characterizes the combination of brain-like chips (neuromorphic hardware) and brain-like algorithms as a "fantasy" that remains unrealized in production systems.
Key claims
The article advances three principal claims. The first, rated high confidence, is that the US electrical grid cannot support the AI industry's projected demand, a claim the author corroborates with reference to an IEA energy report and a NERC assessment. The second, rated medium confidence, is that brain-inspired architectures remain substantially behind transformers on benchmarks, presented as the author's assessment and one that Liquid AI contends otherwise. The third, rated high confidence, is that the combination of brain-like chips and brain-like algorithms has not yet been realized in production, presented as the author's framing and described as uncontested.
Provenance
The article situates brain-inspired approaches within the scaling laws versus efficiency debate. Scaling proponents such as OpenAI and Anthropic argue for continued returns from more compute, data, and parameters, while brain-inspired proponents argue those returns are hitting physical and economic limits. The article's efficiency-first orientation parallels the argument made from a different direction in AI as Normal Technology (Narayanan & Kapoor). It also connects to Edge AI — the argument for smaller, more efficient models that can run on-device rather than in massive data centers — and Daniela Rus's liquid neural networks are explicitly framed as edge-capable in the Digitalist Papers essay (Private Physical Edge AI).
Relationships
- related: Scaling Laws — the paradigm brain-inspired approaches challenge
- related: Edge AI / Private Physical AI — related push for smaller, more efficient models
- related: Liquid AI — Daniela Rus; liquid neural networks; explicit brain-inspired commercialization
- related: Compute Governance — alternative architectures could reduce compute as a bottleneck
- related: AI Environmental Impact — energy constraint is the shared driver
- related: Fast-Follow Problem — efficient architectures could reduce compute moats