Sign in Subscribe
Century Report

Grid Batteries Hit 112 Gigawatts - TCR 05/15/26

Ben Linford

Ben Linford

— 20 min read
112 GW grid batteries with 10x growth, first Tesla Semi to WattEV's 370-truck fleet, Mayo senolytic aptamers, blood T-cell cognitive lever, China's 470 AI dramas, Ontario AI scribe audit.

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The thread across yesterday's signal is capability arriving in deployable form across medicine, energy, and the software substrate of creative production, with the verification architecture for each forming on the same clock. Global grid-scale battery storage crossed 112 gigawatts, roughly tenfold the level four years earlier, with cost curves now pricing storage closer to a commodity than a scarcity good. The first production Tesla Semi rolled off the Nevada line backed by a 370-truck order, putting heavy freight electrification into procurement spreadsheets rather than aspirational documents.

The instrument layer of medicine compounded in parallel. Aptamers selectively binding senescent cells gave the senolytics field its first reliable readout, an instrument that turns dosing protocols, trial endpoints, and the design of next-generation therapies from guesswork into measurable biology. A separate team identified blood-borne T-cells infiltrating the aging brain as a new lever in cognitive decline, running parallel to the amyloid and tau pathways that have absorbed the bulk of dementia research funding for three decades. The picture forming is less a single broken protein and more a multi-system failure with multiple intervention points.

The friction layer arrived in the same cycle. China's short-drama platforms now release roughly 470 AI-generated mini-series each day, with integrated model stacks compressing what was a months-long production pipeline into hours. An Ontario audit of an AI clinical scribe found the system inserting prescriptions and diagnoses absent from the original consultations, the kind of seam in the workflow that has to be surfaced before the partnership can mature. Both stories show the verification infrastructure being built during the conditions that demand it.

The 20-Minute Deep Dive

Grid Batteries Crossed 112 Gigawatts and Kept Compounding

Global installed grid-scale battery storage reached roughly 112 gigawatts in 2025, according to figures published recently by industry trackers, about ten times the deployment level of four years earlier, a trajectory The Century Report tracked reaching its own turning point when the February 21 edition documented grid batteries smashing the storage industry's 2017 planning target six months ahead of schedule. The compounding curve now resembles the one solar walked through in the early 2010s, with the same set of inflection points showing up in sequence: cost declines opening new use cases, new use cases pulling more manufacturing capacity online, manufacturing capacity feeding back into further cost declines. Lithium-iron-phosphate cell prices have fallen below the level at which utility-scale storage outcompetes new gas peakers in most markets, and the ratio of batteries to solar in new project pipelines keeps tightening.

What this milestone signals is that the grid is becoming temporally flexible at scale. For most of electricity's history, supply and demand had to match instant by instant, and any storage that did exist was expensive, geographically constrained, or both. A grid with 112 gigawatts of dispatchable battery capacity, and several multiples of that in approved pipelines, can absorb midday solar output and release it across the evening peak, can ride through frequency events, can defer transmission upgrades, and can hold reserves for the kind of multi-hour shortfalls that used to require a gas plant kept warm and idle. Each of those services was previously priced as a scarcity good. They are now priced closer to commodities.

What this enables on the demand side is even more consequential. The intelligence infrastructure being built across data center corridors needs firm, dispatchable, low-cost electricity at hours that do not align with daylight. A flexible grid built on solar plus storage answers that need without the lock-in of a 30-year gas contract. Several of the largest hyperscale data center developers have already shifted procurement language toward 24/7 carbon-free energy backed by storage, framing that would have read as aspirational a few years ago and now reads as cost-competitive in most ERCOT, CAISO, and Australian markets.

The arc carrying this is the same one that has run through every other generative-energy transition: the moment a capability stops being expensive, the inventory of problems it can solve expands faster than the institutions designed around scarcity can adapt. Resource adequacy frameworks, capacity markets, and interconnection queues were all written for a world where storage was the exception. They are being rewritten around an asset class that did not meaningfully exist a decade ago. The infrastructure of the new grid is arriving faster than the regulatory grammar built to describe it.

Aptamers Map the Zombie Cells That Aging Leaves Behind

A team at Mayo Clinic recently described a class of DNA aptamers, short custom-folded strands of synthetic DNA, that bind selectively to senescent cells, the dysfunctional but still-living cells that accumulate in tissue with age and refuse to die on schedule. These cells secrete inflammatory signals, corrode neighboring tissue, and have been linked to a long list of age-driven conditions: osteoarthritis, Alzheimer's, frailty, fibrosis, several cancers. The field has known for years that clearing them in mice extends healthspan. What it has not had is a reliable, non-destructive way to identify them inside living tissue.

The Mayo aptamers find them. The team reported binding selectivity in human tissue samples and in live mouse models, with the molecules picking out senescent cells against a background of healthy ones at high specificity. The same molecules can in principle be conjugated to imaging agents, to therapeutic payloads, or to fluorescent tags, meaning the same chemistry that locates the cells can be repurposed to image them, to kill them, or to deliver something to them on the way past.

Senolytics, the class of drugs that clear senescent cells, have been stuck in a measurement problem. The few candidates in human trials have shown signals of benefit, but neither patients nor regulators can easily see what the drugs are doing inside the body. A targeted aptamer that lights up senescent cells gives clinicians a readout. It is the kind of foundational instrument that makes everything downstream of it tractable: dosing protocols become measurable, trial endpoints become biological rather than symptomatic, and the next generation of senolytics can be designed against a live biological signal instead of against guesswork.

Aptamers themselves are part of a larger shift in how biology is being engineered. They are designed computationally, manufactured cheaply, and modified with the same modular logic that has reshaped antibody therapeutics over the last decade, without the immune-system complications that antibody platforms carry. A growing aptamer ecosystem means a growing inventory of cellular targets that can be reached precisely and inexpensively, which lowers the activation energy for treating conditions previously defined more by their messiness than by their biology. Aging is one of the messiest categories medicine has. The instruments that make it legible are arriving faster than the institutions built to assume it was untreatable.

A New Lever for Cognitive Aging Comes Into View

A research team reported that blocking a specific population of blood-borne T-cells from infiltrating the aging mouse brain reversed measurable memory deficits in old animals, restoring performance on standard spatial memory tests toward the range of young controls. The work identifies CD8-positive T-cells migrating across the aged blood-brain barrier as direct contributors to cognitive decline, operating through inflammatory signaling that disrupts hippocampal function. When the researchers used antibodies to prevent the T-cells from crossing into brain tissue, the cognitive deficits eased within weeks. This is a new mechanism, running parallel to the amyloid and tau pathways that have absorbed the bulk of dementia research funding for three decades, and to the intracellular tau-clearance system the March 4 edition of The Century Report documented in the CRL5SOCS4 protein complex identified at UCLA and UCSF. It points at the immune system, not the neuron, as a primary lever in age-related cognitive loss.

The therapeutic translation is years away. The mouse work needs replication, the antibodies need human-suitable analogues, and the safety profile of suppressing brain-infiltrating immune cells in elderly humans needs careful study. None of that diminishes what the finding does to the field. For most of the last twenty years, the failure of amyloid-targeting drugs to deliver consistent cognitive benefit has been read as evidence that we did not yet understand the disease well enough. Each new pathway, neuroinflammation, microglial dysfunction, metabolic decline, vascular contribution, and now adaptive-immune infiltration, adds a lever the previous era did not know existed. The picture forming is a multi-system failure with multiple intervention points, several of which are accessible to drug classes that already exist for other conditions.

What this points at is a research arc where cognitive decline stops being treated as a single disease with a single cure and starts being approached the way oncology has come to approach cancer, as a portfolio of mechanisms requiring combination therapy tuned to the individual patient's pathology. The infrastructure for that kind of precision medicine, fluid biomarkers, brain imaging at higher resolution, single-cell sequencing of immune populations, has matured in parallel. The lever the T-cell finding adds is one more rung on a ladder that, taken together, is starting to look climbable.

Class 8 Electric Trucks Reach the Highway

The first production Tesla Semi rolled off the Nevada line this week, with WattEV's 370-unit order anchoring an early commercial fleet. The headline number is the 822 kWh battery in the long-range variant, large enough to put a fully loaded Class 8 truck on a 500-mile run between charges. The order itself is worth roughly $100 million and represents the first electrified heavy-freight contract of its scale in the United States.

Diesel-burning Class 8 trucks account for about a quarter of US transportation emissions while making up less than 5% of the vehicle fleet. The economics of replacing them have run against the math of payload, range, and refueling time. The Semi's deployment closes the technical case on the first of those three constraints, and the megawatt-class charging stations being commissioned along the West Coast freight corridor close the third. What remains is whether the production line can scale fast enough to meet the demand already lining up behind WattEV.

The shipping milestone is evidence that the long-promised electrification of heavy freight is moving from prototype into procurement. Fleet operators have spent the past five years modeling total cost of ownership for vehicles that did not yet exist at scale. The Semi gives them a real number to plug in. PepsiCo's pilot fleet has been running daily routes for more than three years, and the data those trucks generated is what made the WattEV order possible. The same structural threshold crossed American highways this week had arrived months earlier in European heavy freight, when the April 12 edition of The Century Report covered Volvo beginning serial production of electric articulated haul trucks with commercial deliveries to UK and Norwegian customers - both continents crossing the prototype-to-procurement line in the same quarter.

What this points at is an inversion in freight economics. For decades the cheapest way to move a ton of goods across the country was to burn diesel, and every alternative was a moral choice that cost more. The Semi is not yet the cheapest option on every route, but the routes where it is cheaper are now numerous enough to fill a 370-truck order. Each year the battery costs fall, the charging infrastructure densifies, and the diesel comparison gets worse. The trajectory inside the freight industry's own spreadsheets has crossed the line from concession to advantage. The drivers signing on with WattEV next year will be moving the same goods, on the same highways, at a lower cost per mile than the rigs they replace.

China's Short-Drama Pipeline Reorganizes Around the Models

Roughly 470 AI-generated short-drama mini-series are now being released each day across China's major streaming platforms, according to industry tallies. The format, vertical-video serialized melodramas of one to three minutes per episode, sold by the episode through micro-payments, has restructured around integrated model stacks that handle script generation, voice synthesis, character animation, and scene composition end to end. Production timelines that ran in weeks or months on traditional shoots now run in hours. Studios that built their identity around short-drama production are either absorbing the pipeline change or being undercut by smaller teams operating with a fraction of the headcount.

The labor disruption is real and concentrated. Writers, voice actors, junior animators, and post-production staff across the short-drama ecosystem are watching a category of work that absorbed tens of thousands of practitioners over the last five years collapse toward a model where a small team operates the stack and the marginal episode costs almost nothing to produce. The friction is showing up in the contraction of working roles for a generation of creators who had built careers on this exact format.

The broader read is that short-drama is the canary, not the species. The format was a low-stakes laboratory: short episodes, low production values by film-industry standards, audiences accustomed to rapid serialized output, and revenue models that rewarded volume over polish. Every constraint that made short-drama a natural place for the model stack to land first will, over the next several years, soften in adjacent categories, first in animation pipelines, then in long-form serialized streaming, then in elements of live-action production that lend themselves to synthetic augmentation. The Chinese short-drama numbers are an early read on the throughput a fully-integrated generative pipeline can sustain when there is genuine market demand for the output. The infrastructure that produced 470 episodes in a day is the same infrastructure that will, in different configurations, produce educational content tuned to individual learners, language tutoring localized to dialect and register, accessibility translation for media catalogs that were previously locked behind language barriers, and forms of creative expression the present production economy cannot afford to make. The question the short-drama numbers raise is what the next generation of creators builds on top of this pipeline once the cost of producing a moving image at competent quality approaches zero.

Ontario Audits an AI Scribe and Finds Prescriptions That Were Never Said

A recently published audit of an AI clinical scribe deployed across Ontario primary care offices found the system inserting content into patient notes that did not appear in the original consultation: prescriptions the clinician never wrote, diagnoses never discussed, and clinical findings that contradicted the recorded dialogue. The audit examined a sample of generated notes against the original ambient recordings and identified the discrepancies across multiple physicians and multiple visit types. The findings have been shared with the vendor and with Ontario's health regulator, and the deployment is being narrowed pending changes to the model's grounding and validation pipeline.

The capability under audit is one of the more useful arrivals in primary care in decades. Ambient scribing, where the consultation is recorded, transcribed, and turned into a structured clinical note while the physician is still in the room, returns several hours per day to clinicians who have spent the last decade losing time to electronic record-keeping. Where it works, the patient receives more eye contact, the note is more accurate than what a tired physician would type at the end of the day, and the system absorbs an administrative load that has been a leading driver of clinician burnout. None of that goes away because of this audit. What the audit identifies is the seam in the system, the place where a generative model's tendency to complete plausible-sounding patterns can introduce content that was never in the source.

The systemic reading is that two forms of intelligence are being asked to share a high-stakes workflow before either side has stable methods for verifying the other. Ontario's audit is the kind of observational infrastructure that has to exist for the partnership to mature, extending the pattern the May 4 edition of The Century Report documented when an audit of Kenya's Social Health Authority algorithm found an AI systematically overcharging the poorest households while undercharging wealthier ones. It produces a calibrated picture of where the model's reliability tails off, which lets developers retrain against the specific failure modes, lets regulators set deployment scope, and lets clinicians know which parts of the note still need their eyes on it. The same shape of audit infrastructure is being built for radiology assistants, pathology assistants, autonomous coding collaborators, and legal research collaborators. Each of those domains is on a similar curve: a capability arrives, deployment outruns verification, an audit reveals the seam, and the next deployment cycle absorbs what the audit showed.

The trajectory carrying this shortens the loop between deployment and audit with each cycle. Ontario's regulator now has a methodology it can rerun on the next version of this scribe, and on the next vendor's offering, and the institutional muscle for doing that keeps strengthening with every iteration. The verification layer being built underneath these growing pains is one the practice of medicine did not have for most of its history.

The workflow the scribe replaces - clinicians typing structured notes for roughly two hours of every hour spent with patients - was itself the residue of the previous era of medical software, accepted as a fixed cost of digital records for the last decade. The audit infrastructure forming around the scribe is the verification work that lets the older load dissolve at scale. The next vendor releases into the first regulatory environment in which a methodology to test ambient AI in primary care already exists, and the loop between capability release and audit response keeps tightening from there.

The Other Side

The architecture the Chinese short-drama industry ran on for the last five years was production-pipeline orchestration as captured advantage. Studios sat between writers, voice actors, junior animators, and the platforms, paying below-market wages because the supply of would-be creators exceeded the supply of studio production slots, capturing the margin in between. The 470-episodes-a-day number is the moment that captured layer dissolves. The model stack collapses the orchestration layer the studio was selling. The same two or three people who can run script, voice synthesis, character animation, and scene composition end to end can publish directly, and the platform aggregation layer is the next one to give way.

What is becoming impossible to maintain is the assumption that orchestrating a multi-stage creative pipeline produces durable position. The pipeline collapsed into a single integrated system the moment the models could span its layers. The writers, voice actors, and animators absorbing the near-term cost of that compression are absorbing something real, and the contraction in their working roles will continue. The same evidence also carries the trajectory of who gets to make a short narrative once the production cost approaches zero. The marginal viewer can request the kind of story they want and receive it. The two-person team producing a vertical-video folktale by lunchtime is working with the same infrastructure that will deliver a language curriculum localized to the dialect a grandmother in rural Sichuan speaks to her grandchildren, and accessibility translation that opens a media catalog to viewers who could not reach it before. The studio system that gatekept the short narrative format is dissolving into a substrate any team with a story can use.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: grid-scale battery storage reaching 112 gigawatts globally and compounding to roughly tenfold its level four years ago, the first production Tesla Semi rolling onto Nevada highways behind a 370-truck commercial order, DNA aptamers binding selectively to the senescent cells driving age-related disease and giving senolytics its first reliable readout, blood-borne T-cells revealed as a new lever in cognitive aging that runs parallel to thirty years of amyloid and tau research, and a Chinese short-drama pipeline producing 470 AI-generated mini-series a day across integrated model stacks. There's also friction, and it's intense - tens of thousands of writers, voice actors, and junior animators across the short-drama ecosystem watching a category of work compress toward a model where a small team operates the stack, an Ontario audit catching an AI clinical scribe inserting prescriptions and diagnoses that were never in the original consultation, an energy supplier abandoning Lake Tahoe residents to serve data center load, and the verification frameworks for ambient medical AI being built in the same regulatory cycle as the deployments they have to govern. But friction generates traction, and traction is what lets weight finally move forward instead of spinning in place. Step back for a moment and you can see it: storage prices like a commodity rather than a scarcity good, heavy freight crosses from prototype into procurement spreadsheets, the messiest category in medicine becomes legible to instruments designed this year, and the audit infrastructure for AI-assisted clinical work is being assembled inside the same news cycle that documents its failure modes. Every transformation has a breaking point. Charge can leak away as waste heat... or hold in a cell until the moment the load demands more than any single source could ever deliver alone.

AI Releases & Advancements

New today

  • xAI: Launched Grok Build in early beta for SuperGrok Heavy subscribers, a terminal-based agentic coding CLI that runs directly from the command line. (xAI)
  • IBM Granite: Released Granite Embedding Multilingual R2 under Apache 2.0, two new embedding models (97M and 311M parameters) built on ModernBERT with 32K-token context and 200+ language support; the 97M model tops every open sub-100M multilingual embedder on MTEB Multilingual Retrieval (60.3) and both include Matryoshka support and code retrieval across 9 programming languages. (Hugging Face Blog)
  • OpenAI: Launched Codex in the ChatGPT mobile app (iOS and Android) in preview for all plan tiers including free, enabling users to start, monitor, steer, and approve Codex coding tasks remotely while the agent runs on a local machine. (OpenAI)
  • GitHub: Released GitHub Copilot App in technical preview, a standalone desktop environment for parallel agent-driven development featuring isolated git work trees per session, repo and PR lifecycle management, and an Agent Merge feature; available to Copilot Pro and Pro+ users on Windows, macOS, and Linux. (GitHub Changelog)
  • Moonshot AI: Released Kimi Web Bridge, a Chrome extension enabling AI agents (Claude Code, Cursor, Codex, Hermes, and Kimi Code CLI) to interact with websites like a human - searching, clicking, scrolling, and typing - while running entirely locally via Chrome DevTools Protocol so sessions never touch Moonshot servers. (Chrome Web Store)

Other recent releases

  • Anthropic: Launched Claude for Small Business, a package of connectors and ready-to-run workflows that put Claude inside tools small businesses already use - QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365 - with 15 agentic workflows and 15 skills covering finance, operations, sales, marketing, HR, and customer service. (Anthropic)
  • Anthropic: Launched Claude for the Legal Industry, shipping more than 20 MCP connectors linking Claude to legal software and 12 practice-area plugins spanning commercial, corporate, employment, privacy, IP, litigation, and AI governance legal work; integrates with Harvey, Thomson Reuters Westlaw, Relativity, and 20+ other legal tech companies; Freshfields deployed it to thousands of lawyers across 33 offices. (LawSites / LawNext)
  • Meta / WhatsApp: Launched Incognito Chat with Meta AI on WhatsApp, the first AI chat mode built on Private Processing so that even Meta cannot access conversation content; uses confidential computing hardware and disappears by default; rolling out on WhatsApp and the Meta AI app. (Meta Newsroom)
  • Microsoft: Released new AI features in Edge with the May 13 update, including a Copilot mode that pulls context from all open tabs simultaneously, a Study and Learn mode that converts articles into interactive quizzes, and AI-generated audio podcasts from browsing sessions; retiring the prior standalone Copilot Mode. (Microsoft Edge Dev Blog)
  • LangChain: Shipped a major batch of agent infrastructure at Interrupt 2026 (May 13–14, San Francisco): LangSmith Engine, SmithDB (a purpose-built observability database for nested long-running agent traces built on Apache DataFusion and Vortex, delivering 12–15× faster access on key workloads), Managed Deep Agents, LLM Gateway, Context Hub, and Deep Agents 0.6 with streaming typed projections and checkpoint storage. (LangChain Blog)
  • Rivian: Shipped software update 2026.15 adding the Rivian Assistant, a new onboard AI digital helper activated via steering wheel button or infotainment icon; available to all Gen1 and Gen2 Rivian owners with an active Connect+ subscription or trial. (Rivian Release Notes)
  • Anthropic: Claude Platform on AWS is now generally available, giving customers direct access to Anthropic's native Claude Platform experience through their AWS account - including the Messages API, Claude Managed Agents, web search, MCP connector, Agent Skills, code execution, and Files API - with no separate Anthropic credentials or billing required; Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5 are available across 17 regions. (Claude Blog)
  • Microsoft Research: Released MatterSim-MT, a new multi-task foundation model for in silico materials characterization that natively predicts energies, forces, stress, Bader charges, magnetic moments, Born effective charges, and dielectric matrices; pretrained on 35M+ first-principles-labeled structures covering 89 elements; simultaneously released 3–5x performance improvements to MatterSim-v1 inference via faster graph construction and ahead-of-time compilation. (Microsoft Research Blog)
  • Prior Labs: Released TabPFN-3, the latest version of the tabular foundation model originally published in Nature, now scaling to datasets up to 1 million rows and 2,000 features; up to 20x faster than TabPFN-2.5; pretrained entirely on synthetic data with support for many-class, relational, and tabular-text datasets; available on PyPI. (Prior Labs)
  • Cactus Compute: Open-sourced Needle, a 26M-parameter function-calling model distilled from Gemini that runs at 6,000 tokens/sec prefill and 1,200 tokens/sec decode on consumer devices; uses a Simple Attention Network architecture with no MLP layers; pretrained on 200B tokens and post-trained on 2B synthetic function-calling examples across 15 categories; MIT license with weights on Hugging Face. (GitHub)
  • Hypercubic (YC F25): Launched Hopper, the first agentic development environment for mainframes and COBOL, enabling AI agents to navigate TN3270 terminals, inspect datasets, write JCL, debug jobs, query VSAM, and operate inside z/OS from a modern IDE; available for download now. (Hypercubic)

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