Why this needs untangling
Recent coverage of brain-inspired and beyond-silicon computing, including this site's own coverage of Cortical Labs' biological data centres, routinely links together three genuinely different technologies as though they are variations on one idea. They are not, and the difference matters enormously for governance, because only one of the three touches living tissue and raises any bioethics question at all.
Three categories, three different stakes
Neuromorphic computing: silicon chips, brain-inspired architecture, no living tissue, no bioethics question at all. These are conventional semiconductor chips designed to mimic how neurons fire and connect, not actual biological material. Intel's Loihi 2 research chip and IBM's NorthPole are the best-documented examples; both remain research and prototype stage per the companies' own technical materials, with genuinely impressive efficiency results on narrow, sparse workloads that have not translated into commercial data-centre deployment. The one clearly commercial, shipping example is BrainChip's Akida AKD1500 processor, in production as of June 2026 and targeted at low-power edge applications like defence sensors and wearables, not large-scale AI training. The governance question here is ordinary hardware and IP due diligence, nothing more; there is no tissue, no consent question, no welfare question.
Biological computing: actual living neurons, real bioethics stakes. This is Cortical Labs' CL1, Switzerland's FinalSpark, which already rents cloud access to living brain organoids to universities across three continents, and academic Organoid Intelligence research led out of Johns Hopkins, whose own team explicitly frames its work around toxicology and disease modelling, not building anything resembling a mind. This is the only one of the three categories where a real, published bioethics framework exists, and where that framework's own gap, that it covers research use but not commercial infrastructure, is the live governance story.
DNA and glass data storage: inert media, not living tissue, a different governance question entirely. DNA data storage, an industry alliance including Illumina, Twist Bioscience, and Western Digital, and Boston-based Catalog Technologies, recently acquired by French startup Biomemory with commercial deployment targeted for late 2026, uses synthetic, non-living DNA purely as an extremely dense, durable storage medium. It raises no sentience or consent question, since there is no donor and no living cell; its governance question is closer to biosecurity, oversight of synthetic-DNA synthesis capability generally, and technical standards than bioethics. A related but entirely separate technology, Microsoft's Project Silica, stores data by etching patterns into glass with lasers and claims roughly 10,000-year durability. It is frequently and incorrectly described in casual coverage as a DNA-storage breakthrough. It is not DNA at all, and conflating the two is a factual error worth actively avoiding.
The honest state of the "exotic computing will fix AI's energy crisis" narrative
All of this activity is being funded by the same pressure this site tracks in its cloud-emissions and data-centre-regulation coverage: the IEA's 16 April 2026 figures show data-centre electricity demand up 17 percent in 2025 and AI-specific demand on track to triple by 2030. That pressure has genuinely opened investor appetite for speculative bets. Unconventional AI, a startup founded by former Databricks AI chief Naveen Rao, raised 475 million US dollars in December 2025 at a 4.5 billion dollar valuation for an oscillator-based computing approach that, by the company's own account, exists today only as a software simulation. It has not built a physical chip, and its widely repeated claim of 1,000-times energy efficiency is explicitly a theoretical projection, not a measured result. Probabilistic computing, built on stochastic hardware, has a more credible academic pedigree centred at Purdue University, with active peer-reviewed work through 2025 and 2026, but remains firmly in the research phase.
Worth flagging plainly: several widely circulated 2026 claims in this space, including reports of an Intel Loihi 3 chip and IBM NorthPole reaching full-scale production, could not be traced to any primary statement from Intel or IBM in the research behind this article. They appear to originate from low-quality syndicated aggregator content rather than the companies themselves. Some of the enthusiasm behind the moonshot-computing narrative is being manufactured by uncorroborated trade-press repetition, not just startup marketing. Anyone making a technology or investment decision on the strength of these efficiency claims should ask for the primary source before treating any of the headline multipliers, in any of these three categories, as an established fact rather than a projection.
Sources: IBM NorthPole research blog; BrainChip AKD1500 press release; FinalSpark Neuroplatform; Microsoft Project Silica, Nature; IEA data-centre electricity release.
