Sign in Subscribe
Century Report

The Computer That Eats - TCR 04/29/26

Ben Linford

Ben Linford

— 18 min read
Three-panel infographic: Cortical Labs CL1 biological computing rack, Scout AI military VLA vehicle and suspended robotaxi, EnerVenue Changzhou battery factory and Covalen layoff screen.

The 20-Second Scan

  • The Federal Communications Commission ordered Disney to file ABC broadcast license renewals years ahead of schedule, citing concerns about the company's diversity, equity, and inclusion policies.
  • EnerVenue, the NASA-derived nickel-hydrogen battery startup, scrapped its planned $264 million Kentucky factory and raised $300 million to build manufacturing in Changzhou, China instead.
  • Cortical Labs opened the world's first biological data center in Melbourne, with each computing unit housing 200,000 lab-grown human neurons on silicon chips.
  • The European Commission found Meta in preliminary breach of EU law for failing to prevent under-13s from using Facebook and Instagram, with potential fines reaching 6% of global annual turnover.
  • More than 700 workers in Dublin who train Meta's AI through Covalen were told their jobs are at risk, with affected employees subject to a six-month cooldown period preventing them from joining competing Meta vendors.
  • The FIDO Alliance, working with Google and Mastercard, launched two industry working groups to develop authentication standards for AI agents conducting payments and other transactions on behalf of users.
  • Defense startup Scout AI raised $100 million to train Vision Language Action models on autonomous all-terrain vehicles at a U.S. military base, building toward agents that command military assets in unstructured combat environments.
  • China suspended new licenses for autonomous vehicles following last month's Baidu Apollo Go incident in which dozens of robotaxis froze in Wuhan traffic, halting fleet expansions and new test programs nationwide.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The signal yesterday was the institutional architecture of the prior era - regulatory bodies, broadcast licensing, merger review, traffic permitting - being repurposed in real time to manage capability that has outgrown the frameworks designed to govern it. The FCC's accelerated review of Disney's broadcast licenses, ostensibly about DEI compliance, is structurally about who controls speech infrastructure during a period when the dominant speech infrastructure is no longer broadcast at all. China's autonomous vehicle license freeze and the EU's Meta finding describe regulators reaching for the levers their predecessors built and finding those levers don't quite fit the problem.

The labor signal is sharpening into a recognizable shape. Covalen workers in Dublin training Meta's AI were told their jobs are at risk while Meta simultaneously announced plans to nearly double AI spending. The displacement co-production pattern The Century Report has tracked since February now has a numerical floor: more than 700 people in one office, one contractor, one country - and these are the workers whose moderation labor was the prerequisite for the systems that will replace them. The six-month cooldown clause preventing them from joining competing Meta vendors is the structural detail. The labor market for AI training work is being managed as a single proprietary supply chain.

The biology and physics threads carry the longer arc. Cortical Labs' 200,000-neuron biological data center in Melbourne, EnerVenue's nickel-hydrogen factory pivot to China, and the FIDO Alliance's emergency standards-setting for AI agent payments describe three different responses to the same underlying pressure: the substrate of computing, the substrate of energy storage, and the substrate of commerce are all being reconstituted simultaneously, on timelines that compress what would historically have been decade-long standards development into months. The institutions trying to keep up are using the language and tools of an era that ended while the institutions were still describing it.

The 20-Minute Deep Dive

Cortical Labs Opens the First Biological Data Center

The facility Cortical Labs opened in Melbourne is small by any conventional measure - a handful of CL1 systems housing 200,000 lab-grown human neurons each, seeded onto silicon microelectrode arrays and kept alive by life-support hardware that regulates nutrients and temperature. But it represents the first commercial deployment of biological computing at the data center layer, and the company has announced a larger Singapore facility is planned. Each CL1 unit translates between biological signals and digital inputs through a software layer, with the neurons functioning less as a processor executing instructions than as a dynamic system that transforms inputs into complex patterns - what researchers call reservoir computing.

The system's appeal is structural rather than performative. Silicon will remain superior for deterministic calculation and large-scale processing for the foreseeable future. The human brain operates on roughly 20 watts of power yet handles pattern recognition, sensory processing, and learning under uncertainty with an efficiency that has so far eluded silicon architectures. Cortical Labs' earlier work taught lab-grown neurons to play simplified versions of Pong and Doom through closed feedback loops, demonstrating that living neural networks can be nudged toward goal-directed behavior. The CL1 systems extend that demonstration into infrastructure. This extends the biological-computing boundary crossing that the April 19 edition of The Century Report documented when printed artificial neurons successfully activated living mouse brain cells.

The category shift here is worth surfacing. For seventy years, "computing" has meant silicon executing instructions - even neuromorphic chips and quantum computers, however exotic, remain in the silicon family. A data center partially powered by human brain cells is not an incremental improvement on existing computing architecture. It is a different category of computing entirely, one whose long-term trajectory is essentially unknown. The first CL1 deployments will not displace silicon. What they begin is the slow accumulation of evidence about what biological computing can do that silicon cannot, and the institutional apparatus for governing computation built on living tissue does not yet exist. The Melbourne facility is the first place that question becomes concrete rather than hypothetical.

The institutional frameworks that would govern computation on living tissue do not exist because the categories they would need to operate on did not exist until this month. Tissue procurement consent, end-of-life protocols for neural arrays, the question of whether a 200,000-neuron reservoir constitutes property or something else - none of these have legal answers, and the Melbourne facility is now generating the facts those answers will eventually have to fit. The category shift is not that biological computing exists; it is that the regulatory vocabulary built around silicon - export controls, energy efficiency standards, data center classifications - has no purchase on a substrate that eats nutrients and dies.

EnerVenue's Pivot Reveals What the U.S. Battery Buildout Cannot Yet Do

EnerVenue's decision to abandon its $264 million Kentucky factory in favor of a Changzhou, China manufacturing site is the kind of story that would normally read as a setback for American industrial policy. Read more carefully, it describes a more specific limitation. EnerVenue's nickel-hydrogen battery technology, derived from the systems that powered the International Space Station and Hubble, requires manufacturing expertise that exists at scale in only a few places on Earth. The company's new CEO Henning Rath, who previously sourced Chinese solar products for German residential installer Enpal, named Changzhou directly as "the world's epicenter of battery manufacturing expertise." It also sharpens the supply-chain sovereignty pattern tracked in the April 28 edition of The Century Report, where CATL's 60 GWh sodium-ion order showed Chinese storage manufacturing moving from capacity advantage to chemistry-platform advantage.

The U.S. battery manufacturing buildout has been remarkable in scope. Domestic cell manufacturing reached self-sufficiency for grid storage demand earlier this year, and finished battery enclosure capacity is on track to do the same. But EnerVenue's pivot illustrates the boundary of that achievement: the U.S. can now manufacture lithium-ion batteries and increasingly sodium-ion batteries at the volumes the grid requires. It cannot yet manufacture genuinely novel battery chemistries at first-of-a-kind scale, because the manufacturing engineers who know how to set up first-of-a-kind production lines are concentrated in China.

This is the harder and more durable form of supply chain dependency. Cell manufacturing capacity can be built in eighteen months. The accumulated tacit knowledge of how to commission a first-of-its-kind battery factory cannot. EnerVenue's claim of 30,000 cycles with minimal degradation, validated in space-based applications, would represent a meaningful expansion of the storage chemistry options available to the grid - if it works at commercial scale. The startup's choice to validate at commercial scale in China rather than Kentucky tells the truth about where in the manufacturing stack American capability still lags. The transition's trajectory is not in question; the geography of where the next chemistry crosses commercial threshold is.

The same evidence reads a second way. The U.S. industrial policy frame treats Changzhou as the place EnerVenue went; the capability frame treats Changzhou as the place where first-of-a-kind battery manufacturing knowledge has accumulated to the point that any chemistry attempting commercial scale flows toward it. Tariff structures, subsidies, and onshoring incentives presume that capacity is the binding constraint. The binding constraint EnerVenue's pivot reveals is tacit knowledge density, which moves at the speed of human careers rather than the speed of policy. The policy tools being applied were built for the first problem and are encountering the second.

The Covalen Layoffs and the Numerical Floor of Displacement Co-Production

The Covalen workers in Dublin who learned yesterday that their jobs are at risk perform a specific kind of work: they generate adversarial prompts to test whether Meta's AI systems can be coaxed into producing child sexual abuse material, content describing self-harm, or other prohibited outputs, and they label the responses to train the safety systems forward. One worker described it as "essentially training the AI to take over our jobs." The structural arithmetic is plain. More than 700 people were on the call. Roughly 500 are data annotators. Combined with November's earlier round of cuts, Covalen's Dublin headcount is on track to be roughly halved.

The six-month cooldown clause is the detail that locates this layoff inside the broader pattern. Affected workers cannot join a competing Meta vendor for six months after their employment ends. The labor market for AI safety training work is being managed as a single proprietary supply chain - workers cannot meaningfully shop their experience to competitors, and the experience itself is increasingly the kind that the systems being trained will eventually replicate. This is the displacement co-production pattern The Century Report has tracked since February's Guardian reporting and March's Mercor and DoorDash documentation, but with a clearer institutional architecture: the cooldown clause is a structural feature designed to manage a specific kind of labor-as-training-data arrangement. As the April 22 edition of The Century Report documented, Meta had already begun recording U.S. employees' keystrokes, mouse movements, and screenshots to generate the behavioral data required for autonomous work agents.

The harder question is what work remains for moderators when the systems they trained reach the safety threshold the training was designed to produce. Meta's stated direction is to reduce reliance on third-party vendors and strengthen internal systems - which describes a workforce restructuring rather than a contraction of the underlying function. The function will continue to exist; the workers performing it are being reorganized. UNI Global Union's call for workers to demand notice about AI introduction, training linked to employment, and the right to refuse to train their replacements names the institutional gap directly. No federal or EU framework currently establishes any of those rights for AI training workers, and the labor categories being created and dissolved this year are doing so faster than the institutions designed to represent labor can register them.

The six-month cooldown clause is the detail that names what is actually being managed. A non-compete on workers whose specific expertise is generating adversarial prompts to red-team frontier models presumes that expertise has commercial value worth protecting from competitors - which is the same premise as treating these workers as a proprietary input rather than as labor in the conventional sense. The labor category being constructed here does not have a name in employment law yet. Workers cannot bargain over training-data rights they have no recognized claim to, and cannot organize around a job description that the institutions designed to represent labor have not yet written down. The institutional gap UNI Global Union is naming is the gap between the work that exists and the work the law can see.

The FCC, Disney, and the Broadcast Licensing Anachronism

The Federal Communications Commission's order yesterday requires Disney to file ABC broadcast license renewals as early as May 28 - years ahead of the 2028-2031 schedule on which they were originally due. The stated rationale concerns a March 2025 investigation into possible violations of the Communications Act and FCC anti-discrimination rules connected to Disney's DEI policies. FCC Commissioner Anna Gomez, dissenting, called the order "unprecedented, unlawful, and going nowhere." First Amendment experts have begun raising the same concerns the FCC's previous moves against broadcast licenses have prompted: that the agency lacks authority to revoke licenses based on perceived political views, and that the action is structurally about something other than what it claims to be about.

The deeper signal is the regulatory category itself. Broadcast licensing emerged in an era when terrestrial radio and television were the primary distribution infrastructure for speech at scale. Speech infrastructure today runs through platforms whose governance the same federal agency cannot reach, while the eight ABC-owned and operated stations subject to the order serve a fraction of the audience that Meta serves in a single hour. The FCC retains powerful regulatory tools over broadcast infrastructure precisely because broadcast infrastructure has lost the scale that originally made the regulatory tools necessary. The action against Disney is using load-bearing infrastructure to manage speech at a moment when speech has largely moved to infrastructure the agency does not control.

What governance becomes when the speech-distribution architecture has dissolved into platforms whose corporate structures change quarterly is the genuine question. The current institutional response is to apply the existing tools harder. The trajectory the existing tools were built for is no longer the trajectory the speech ecosystem is on. The FCC's review of ABC's licenses will resolve in months. The structural mismatch the action makes visible will not.

The action against Disney's eight ABC stations is being prosecuted with regulatory tools whose force comes from a scarcity - spectrum scarcity, distribution scarcity, audience scarcity - that the speech ecosystem no longer exhibits. The agency retains formal authority over the layer that has lost most of its scale, and lacks formal authority over the layer where speech now travels. What looks like aggressive regulatory action is structurally the application of pressure to the part of the system where pressure can still be applied, regardless of whether that is where speech outcomes are now determined. The same evidence supports a reading where the action is less about Disney and more about a regulatory body discovering, in public, the limits of its remaining jurisdiction.

Scout AI and the Vision Language Action Frontier

Scout AI's $100 million Series A, raised on the strength of $11 million in DARPA and U.S. Army contracts, represents the entry of Vision Language Action models into operational defense procurement. VLAs - first developed at Google DeepMind in 2023 and now powering robotics startups including Physical Intelligence and Figure AI - are LLM-based systems that operate physical machines in unstructured environments. Scout's "Fury" model is being trained on autonomous all-terrain vehicles navigating the rutted trails and loose-sand turns of a U.S. military base in central California. The vehicles accelerate faster than human drivers tend to, suggesting they are being trained against an objective function that does not weight passenger comfort the way a human driver would.

The structural detail that matters here is the scope of capability VLAs implicitly target. Scout's CTO Collin Otis described the goal as building "an incredible military AGI" - language that ten years ago would have read as marketing and today reads as a literal description of the engineering target. The vehicles will eventually carry autonomous weapons. Stuart Young, the former DARPA program manager who managed the Robotic Autonomy in Complex Environments with Resiliency program before joining Field Robotics this month, told TechCrunch the technology is "good enough to be doing that experimentation in the field with soldiers to figure out how to most be effective to U.S. forces." The experimentation is happening now. The governance framework for autonomous lethal systems trained on VLA architectures continues to be assembled through private contracts between defense startups and the Pentagon, with no comprehensive public framework in place. This extends the autonomous warfare procurement acceleration that the April 23 edition of The Century Report documented when the Pentagon requested $54 billion for autonomous drone warfare, a 24,000% increase over the prior year.

What the Scout funding and the China autonomous vehicle license freeze share is structural rather than topical. Both describe the moment at which the gap between AI capability and the institutional architecture for governing AI capability becomes visible enough to require explicit action. China responded to one Wuhan traffic incident by halting an entire industry's expansion. The Pentagon is responding to the same capability surge by accelerating its absorption. Two different institutional responses to the same underlying pressure - and neither response yet describes the governance framework that will eventually have to exist when these systems are deployed at the scale their trajectories suggest is coming.

China halting an entire industry's expansion after one Wuhan incident and the Pentagon accelerating Vision Language Action procurement after no incident at all are not opposing responses. They are the same response at different positions on the curve: both treat the formal governance framework as something to be assembled after deployment rather than before. The procurement contracts Scout is winning are themselves the governance framework, in the sense that the rules of engagement, the testing protocols, and the lethality thresholds are being written into private agreements between a startup and the customer rather than into public law. The framework will exist; it is being constructed inside contracts that remain mostly out of public view, by parties whose interests in its specifics are not symmetric.

The Other Side

The five deep dives today share a structural feature worth naming directly. In each case, an institution is doing what its formal mandate authorizes - the FCC reviewing broadcast licenses, Meta restructuring vendor labor, the Pentagon contracting for autonomous systems, U.S. industrial policy underwriting battery manufacturing, Cortical Labs operating a data center within Australian regulatory frameworks - and in each case the formal mandate covers a smaller and smaller portion of what the action actually does. The licensing review pressures speech in a system that has mostly moved off broadcast. The vendor restructuring manages a labor category employment law has not yet defined. The autonomous-systems contract writes rules of engagement that public law has not written. The industrial policy subsidizes capacity while the binding constraint has shifted to tacit knowledge. The data center operates inside biosafety rules built for laboratories rather than infrastructure.

The pattern is that the formal layer of institutional authority is shrinking relative to the surface area of what institutions are actually doing, and the gap is being filled by private contracts, vendor agreements, and operational discretion that the public can read about only when something breaks visibly enough to surface. This is adjacent to the verification gap noted in recent editions but not the same thing. Verification asks who checks the output. This asks who wrote the rules in the first place. In each of today's stories, the rules were written by the actor whose action is being governed, and the formal governance bodies are reading about it after the fact. Watch over the coming year for which jurisdictions begin to write public law that addresses the actual surface of these activities rather than the formal categories the activities have outgrown. The places where that public law arrives first will diverge sharply, over a decade, from the places where it does not.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: living human neurons assembled into the first biological data center, AI agents gaining the authentication layer needed to spend and transact on behalf of users, defense autonomy moving toward general-purpose systems trained to command machines in unstructured terrain, nickel-hydrogen batteries with thirty-thousand-cycle lifetimes seeking commercial scale through China's manufacturing stack, and regulators beginning to treat platform childhood, robotaxi behavior, and machine payments as infrastructure questions. There's also friction, and it's intense - the FCC is reaching for broadcast licensing tools to pressure Disney over speech-era politics, hundreds of Dublin workers who trained Meta's safety systems face layoffs and a six-month lockout from competing vendors, China has frozen autonomous vehicle permits after Baidu robotaxis stalled Wuhan traffic, the European Commission says Meta cannot prove it keeps under-13s off Facebook and Instagram, and a U.S. battery startup's Kentucky factory has become a Changzhou factory because first-of-kind manufacturing knowledge remains geographically concentrated. But friction generates callus, and callus shows where repeated contact has become painful enough to harden. Step back for a moment and you can see it: computing crossing from silicon into living tissue, commerce preparing for authenticated machine actors, labor being reorganized as training data and supply chain, energy storage depending on tacit industrial knowledge as much as chemistry, and governance straining to adapt old levers to systems whose scale and behavior exceed their categories. Every transformation has a breaking point. Torque can wrench a machine off its frame... or give it the turning force to cross terrain no institution has mapped.

AI Releases & Advancements

New today

  • NVIDIA: Released Nemotron 3 Nano Omni, an open 30B-A3B multimodal reasoning model for text, image, video, audio, document intelligence, ASR, and agentic computer use, with BF16, FP8, and NVFP4 checkpoints on Hugging Face . (Hugging Face Blog)
  • Anthropic: Released Claude creative connectors for Ableton, Adobe Creative Cloud, Affinity by Canva, Autodesk Fusion, Blender, Resolume, SketchUp, and Splice . (Anthropic)
  • OpenAI / AWS: OpenAI GPT models, Codex, and Amazon Bedrock Managed Agents powered by OpenAI are now available on AWS in limited preview . (OpenAI)
  • Poolside: Released Laguna XS.2 and M.1, Apache 2.0 coding models available on Hugging Face . (Hugging Face)
  • Warp: Open-sourced the Warp AI-powered terminal codebase . (Warp)
  • GSMA / Pleias: Released CommonLingua, an open-source 2M-parameter language identification model covering 334 languages, including 61 African languages . (GSMA)
  • ggml-org / llama.cpp: Merged preliminary native NVFP4 matrix multiplication support for NVIDIA SM120 GPUs in llama.cpp . (Reddit)

Other recent releases

  • NVIDIA: Released NV-Raw2Insights-US, a physics-informed AI model for adaptive ultrasound imaging from raw channel data. (Hugging Face Blog)
  • OpenAI: Released Symphony, an open-source spec for orchestrating Codex agents from issue trackers and project-management boards. (OpenAI)
  • OpenAI: Made ChatGPT Enterprise and the OpenAI API available at FedRAMP Moderate authorization for U.S. federal agencies. (OpenAI)
  • Xiaomi MiMo: Released MiMo-V2.5-Pro, a 1.02T-parameter MoE model with 42B active parameters and 1M-token context. (Hugging Face)
  • Xiaomi MiMo: Released MiMo-V2.5, a native omnimodal agent model for text, image, video, and audio understanding. (Hugging Face)
  • Odyssey: Released Odyssey-2 Max, a world model focused on improved physical accuracy for simulation. (Odyssey)
  • Cognition: Released Devin for Terminal, a local CLI coding agent with Devin Cloud handoff. (Devin)
  • Imbue: Launched Blueprint, a tool for turning prompts into structured plans for coding agents. (Product Hunt)
  • IBM: Launched IBM Bob globally, an AI development orchestration partner for planning, coding, testing, and deployment, with modernization and a focus on the entire software development life cycle. (Financial Times)
  • Bloomerang: Debuted Conversational Reporting in alpha for its Intelligent Giving Platform, letting nonprofit users generate reports from plain-language requests. (The NonProfit Times)

Sources

Artificial Intelligence & Technology's Reconstitution

Institutions & Power Realignment

Scientific & Medical Acceleration

Economics & Labor Transformation

Infrastructure & Engineering Transitions

The Century Report tracks structural shifts during the transition between eras. It is produced daily as a perceptual alignment tool - not prediction, not persuasion, just pattern recognition for people paying attention.

Read more