Sign in Subscribe
Century Report

The Walls Go Up, Capability Walks Out - TCR 04/27/26

Ben Linford

Ben Linford

— 19 min read
Three-panel infographic: U.S. State Dept distillation cable and Vietnam-SK sovereign AI stack, Intellia CRISPR single-infusion gene therapy, Meta-Overview Energy space-based solar.

The 20-Second Scan

  • The U.S. State Department issued a global diplomatic cable directing posts worldwide to warn foreign governments about alleged AI distillation by DeepSeek, Moonshot AI, and MiniMax.
  • China announced it will block Meta's acquisition of Manus AI, the most aggressive Chinese state action documented to stop AI talent migration to the United States.
  • Meta signed a capacity reservation agreement with Overview Energy for up to 1 GW of space-based solar power beamed via infrared light to terrestrial solar farms, with first satellite launches planned for 2030.
  • Tesla registered 303,960,630 shares with the SEC for delivery to Elon Musk under his 2018 pay package, ending a six-year legal saga over the largest executive compensation award in corporate history.
  • Intellia Therapeutics reported its Phase 3 CRISPR trial for hereditary angioedema cut attacks 87% versus placebo, with 62% of patients attack-free six months after a single in-vivo gene-editing infusion.
  • SK Group signed two MoUs with Vietnam to build a national AI data center and exporting its "Korean-style AI full-stack" model abroad for the first time.

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The thread running through yesterday's signal is jurisdictional: states, companies, and capital are now drawing borders around AI capability faster than the architecture for crossing those borders can be built. The State Department cable instructing diplomats on six continents to confront foreign governments about Chinese distillation is a qualitatively new posture - a coordinated diplomatic campaign treating AI capability extraction as a category of theft requiring multilateral response. China blocking Meta's Manus acquisition forty-eight hours later closes the symmetry. The two largest AI powers are now actively working to prevent capability transfer in both directions, and they are doing so through tools - diplomacy, merger review, immigration policy - that were designed for an era when intellectual property moved through patents and people moved through visas.

The Intellia Phase 3 result and the Meta-Overview Energy agreement describe what the jurisdictional contest is actually contesting. CRISPR editing inside the human body, performed once via infusion, eliminating a disease across six months for nearly two-thirds of patients, represents capability that did not exist in any nation a decade ago. Space-based solar farms beaming infrared light to ground-based collectors represent infrastructure that does not yet exist in any nation but is being financed today against demand that has already arrived. Both stories show the actual frontier moving while governance scrambles to keep up. The Vietnam-SK Group agreement is the same dynamic from a smaller country's perspective - a sovereign nation choosing to import an entire AI stack from one ally rather than build it alone or remain dependent on the larger powers that are now actively decoupling.

Tesla's $114 billion share registration carries a different weight. The legal contest that began with a Delaware court voiding the package as a governance failure ended with the Supreme Court restoring it - and the same board that the original ruling found incapable of arm's-length negotiation just approved a follow-on package potentially worth ten times more. The mechanism that produced the original problem is now being scaled. Across all five threads, the pattern is the same: capability is compounding, the institutional frameworks built for prior eras are visibly straining, and the new frameworks are being assembled in real time by whoever moves first.

The 20-Minute Deep Dive

The Distillation Cable Globalizes The Capability Extraction Fight

The State Department cable sent on Friday, reported by Reuters and confirmed by iTnews and CNBC, instructs U.S. diplomatic and consular posts to "speak to their foreign counterparts about concerns over adversaries' extraction and distillation of US AI models" and to "lay the groundwork for potential follow-up and outreach by the US government." A separate démarche message went directly to Beijing. The cable specifically names DeepSeek, Moonshot AI, and MiniMax, and warns that "AI models developed from surreptitious, unauthorised distillation campaigns enable foreign actors to release products that appear to perform comparably on select benchmarks at a fraction of the cost but do not replicate the full performance of the original system."

This is the same arc The Century Report tracked when Anthropic accused Chinese labs of systematic capability mining via the Claude API in February, when the Frontier Model Forum coordinated detection intelligence across OpenAI, Anthropic, and Google in early April, and when the OSTP made its "industrial-scale" theft accusation last week. What is structurally new about the cable is the diplomatic vector. Sanctions, export controls, and regulatory action are unilateral tools the United States can deploy domestically. A coordinated diplomatic campaign asking allied governments to scrutinize Chinese AI products on capability-extraction grounds is an attempt to globalize the frame itself - to convert what has been a U.S.-specific industrial concern into shared international policy. That sequence extends the capability-extraction arc the April 24 edition of The Century Report documented when DeepSeek V4 arrived as an open-source frontier model explicitly optimized for Huawei and Cambricon silicon.

The cable's release is timed days before the planned Trump-Xi meeting in Beijing, and arrives the same week DeepSeek's open-source V4 model crossed 1.6 trillion parameters running on domestic Huawei silicon. The structural tension is sharpest there. The capability the U.S. is asking allies to help contain is increasingly available as open weights running on chips that U.S. export controls have not been able to constrain. The Chinese embassy's response - that the allegations are "groundless" and a "deliberate attack on China's development and progress in the AI industry" - frames the dispute exactly as a development-rights question. The diplomatic theater is real. The capability arithmetic underneath it is moving faster than any of the institutional channels built to govern it. Whether the cable produces actual policy alignment with allies or whether it accelerates the exact decoupling it is trying to manage will depend on what those allies conclude about where the frontier is actually moving and who is actually leading it.

The diplomatic cable assumes that capability extraction is the relevant frame, and that frame holds only if the capability being extracted remains scarce enough to gate. The same week the cable went out, DeepSeek V4 crossed 1.6 trillion parameters as open weights running on Huawei silicon. A diplomatic campaign organized around stopping distillation of closed models presumes that closed models are still where the frontier lives - a presumption the open-weight evidence is steadily contradicting. Watch whether allied governments respond to the cable with policy alignment or with quiet hedging toward sovereign stacks; the second response would indicate they have already read the capability arithmetic the cable is trying to argue against.

China Closes The Other Side Of The Border

Forty-eight hours after the U.S. cable went out, Chinese authorities announced they will block Meta's acquisition of Manus AI, the AI agent company founded in China that has been rapidly expanding in the United States. The Washington Post described the move as "Beijing's most aggressive step yet to stanch the loss of AI talent and resources to the United States." Manus's founders relocated to Singapore last year and have been operating with U.S. investment, but the Chinese state's invocation of merger-review authority over a company that has effectively left China indicates a willingness to extend regulatory reach beyond traditional jurisdictional limits.

This is the symmetric move to the State Department cable. Both governments are now actively working to prevent capability transfer in both directions, using whatever institutional tools their respective systems provide. The U.S. uses diplomacy, export controls, and increasingly novel applications of supply-chain risk authority. China uses merger review, talent retention requirements, and platform-level access decisions. The Manus action sets a precedent that will affect every Chinese-founded AI company with U.S. investors or U.S. expansion plans, which is to say a substantial fraction of the global AI startup landscape.

The institutional architecture for governing AI in a multilateral world was always going to have to be invented. What yesterday's signal makes clear is that the architecture is being invented backwards - through prohibition, blockade, and review - rather than through the kind of cooperative frameworks that earlier general-purpose technology transitions eventually required. The push to govern AI is producing fragmentation faster than it is producing standards. The reading that emerges is sobering in the short term and potentially clarifying in the longer term: capability of this magnitude is unlikely to remain successfully partitioned for long, and the institutions that ultimately govern it will need to look different from anything currently being built. Whether they emerge from the friction itself or from a deliberate choice to build them is the open question.

A Single CRISPR Infusion, Six Months Of Outcomes

Intellia Therapeutics announced that its Phase 3 trial of lonvoguran ziclumeran for hereditary angioedema - a rare swelling condition that can be life-threatening - met its primary endpoint by reducing attacks 87% compared to placebo. Six months after a single hours-long infusion, 62% of treated patients were free from attacks and were not using other therapies. CEO John Leonard told CNBC that "this is the first Phase 3 data in any indication with in vivo Crispr where you're actually changing a gene that causes disease," and noted that across nearly six years of follow-up the company has not seen a single case in which the editing effect diminished over time.

This extends the gene-editing arc The Century Report tracked through the Karolinska OTOF hearing-loss restoration on April 3, the Regeneron Otarmeni FDA approval and free-distribution announcement on April 24, and the ThermoCas9 tumor-selective editing platform on April 20. The April 25 edition of The Century Report traced the adjacent step from platform biology into clinical systems as AI-designed oncology and immunology drugs approached human trials while hearing-loss gene therapy crossed into FDA approval. What Intellia's result adds is the demonstration that in-vivo editing - performed inside the body via single infusion rather than ex-vivo through cell extraction - can achieve durable Phase 3 efficacy in a chronic disease. The first FDA-approved CRISPR therapy, Vertex's Casgevy, requires extracting blood cells, editing them outside the body, and reinfusing them. Intellia's approach makes the edit directly in the liver. The complexity, cost, and infrastructure requirements of these two approaches differ by an order of magnitude.

The implication is that the architecture of gene therapy is converging toward what Isomorphic Labs's Max Jaderberg described last week as the precision-from-the-start design philosophy in AI-designed small molecules - therapies that begin with molecular understanding sufficient to intervene correctly the first time, rather than therapies that arrive at correctness through years of iterative refinement. Combined with the FDA's plausible mechanism pathway that The Century Report covered on February 24, and the personalized N-of-1 editing program at the Innovative Genomics Institute now planning a second pediatric immune-disorder trial, the institutional and technical architecture for treating rare diseases at population scale is being assembled. The bottleneck shifting from molecular design to manufacturing, distribution, and reimbursement - the same shift Regeneron acknowledged when it offered Otarmeni free to all eligible U.S. patients - is the harder question now.

The word "treatment" is becoming something different. A treatment used to be an ongoing relationship between patient and medical system - prescriptions refilled, dosages adjusted, side effects managed across years. Intellia's 87% reduction from a single hours-long infusion describes something closer to a repair: a discrete event that resolves the underlying condition and ends the relationship. The infrastructure built for chronic-care economics - insurance reimbursement structured around recurring claims, pharmacy distribution organized around refills, clinical practice organized around ongoing management - is calibrated to a category of medicine that is starting to dissolve. Regeneron's decision to provide Otarmeni free to eligible U.S. patients is the first acknowledgment that the old reimbursement architecture cannot price one-time cures coherently.

Solar Power From Orbit Gets Its First Commercial Customer

Meta's capacity reservation agreement with Overview Energy commits to receiving up to 1 GW of power generated in space and transmitted to terrestrial solar farms via infrared light. Overview's spacecraft, planned to begin launching in 2030 with a fleet of 1,000 satellites in geosynchronous orbit, would convert solar energy collected in orbit into infrared beams that existing ground-based solar farms can convert back into electricity. CEO Marc Berte told TechCrunch the company has already demonstrated power transmission from an aircraft to the ground and is planning a satellite demonstration in January 2028.

The technical approach is structurally distinct from earlier space-based solar concepts. Earlier proposals relied on microwave beams or high-power lasers that would have required entirely new ground-receiving infrastructure and faced unresolved safety and regulatory questions. Overview's wide infrared beam targets existing utility-scale solar farms - the company says the beam is safe enough that a person can stare directly into it - and treats the satellites as supplemental capacity that extends terrestrial solar from a daytime resource into a continuous one. The economics depend on whether 1,000 satellites can be deployed and maintained at costs that compete with the storage and grid solutions emerging on the ground.

Meta's interest is the demand-side signal. The company's data centers consumed more than 18 TWh in 2024 - roughly enough to power 1.7 million American homes - and Meta has committed to building 30 GW of renewable capacity. The Overview agreement is small in proportion to that commitment but structurally significant: it is the first instance of a hyperscaler treating space-based solar as a serious supply option for the AI infrastructure buildout. The arc The Century Report has tracked through Fervo Energy's geothermal IPO, the SunZia 3.5 GW wind transmission line, the Pattern Energy Maximo robotic solar installation, and the Cache Energy thermal battery deployment in Minnesota now extends into orbit. This extends the infrastructure stack the April 21 edition of The Century Report followed through Anthropic's 5 GW AWS commitment and Fervo's geothermal IPO filing, where compute demand and firm clean power began visibly converging. The physical substrate of the intelligence era is being assembled across every available altitude simultaneously.

Coverage is treating the Overview agreement as exotic - solar from space, beamed via infrared, satellites in geosynchronous orbit. The same facts also describe something more ordinary: a hyperscaler shopping for firm clean power and finding that the cost curve on orbital generation has bent far enough to compete with terrestrial alternatives a decade out. The exotic detail is the altitude. The structural detail is that compute demand is now financing infrastructure categories that did not have commercial customers a year ago. When demand is large enough and inelastic enough, supply curves bend in places the prior generation of energy planners did not bother to model.

Vietnam Imports An Entire AI Stack

SK Group signed two memoranda of understanding with the Vietnamese government to build core AI infrastructure in the country. SK Innovation will explore power supply for an AI data center connected to Nghe An's Quynh Lap LNG project, while SK Telecom will develop, build, and operate the data center itself. A second MoU with Vietnam's National Innovation Center commits SK Group to broader AI ecosystem development including model deployment and industry-specific services. SK Telecom CEO Jung Jai-hun described the arrangement as the first overseas expansion of SK's "Korean-style AI full-stack" model, which links semiconductor design, data center construction, energy supply, model development, and AI services into a single integrated capability.

What makes the Vietnam agreement structurally interesting is the choice it represents. Vietnam has the option to build its own AI stack incrementally, to adopt American or Chinese systems and accept the dependency that comes with each, or to import a packaged stack from a third country whose geopolitical alignment is closer to neutral. By choosing the third option, Vietnam is doing exactly what Mistral has been doing for European sovereignty, what Rapidus represents for Japan, and what India's full-stack manufacturing strategy represents in semiconductors and open models simultaneously. As the April 26 edition of The Century Report noted with the Cohere-Aleph Alpha sovereign-AI merger, the middle layer of the market is increasingly national infrastructure rather than chatbot distribution. Sovereignty in the AI era is being defined less by domestic capability than by which combination of foreign capabilities a country can assemble without falling into a single dependency relationship.

The deal also reveals what SK is selling. The full-stack proposition - integrated semiconductor, energy, infrastructure, and software capability - is a different commercial offering from the per-token API access frontier labs sell or the per-GPU compute hyperscalers rent. It treats AI as national infrastructure rather than as a product. As more countries make the Vietnam-style choice, the commercial architecture of global AI is bifurcating: hyperscale-and-foundation-model relationships at the top of the market, full-stack national-infrastructure relationships in the middle, and open-weight diffusion underneath. Whether these layers reinforce each other or compete remains an open question. What is clear is that the simple narrative of "U.S. labs versus Chinese labs" no longer describes how the actual market is forming.

The deal reveals something the U.S.-versus-China framing obscures: the middle of the global market is being organized by countries that decline to choose. Vietnam is not picking a side; it is assembling a portfolio of foreign capabilities calibrated to avoid singular dependence. SK is selling integration, which is a different commercial product than capability itself - and integration is a position whose durability depends on the integrated components remaining hard to assemble independently. As open-weight models, commodity GPUs, and modular data-center designs make stack assembly easier, the value SK is selling Vietnam compresses. The deal is large now because the assembly problem is hard now. The assembly problem is getting easier on the same trajectory that produced DeepSeek V4.

Tesla Closes The Pay Saga, Opens A Larger One

Tesla's S-8 filing with the SEC registers 303,960,630 shares of common stock for delivery to Elon Musk under his 2018 CEO Performance Award - shares worth approximately $114 billion at current prices. The filing follows the Delaware Supreme Court's December 2025 reversal of the lower court ruling that had voided the package on governance grounds, finding that "full rescission was too extreme a remedy" given the six years of work Musk had performed under the award. The original Delaware Chancery Court ruling had identified specific governance failures: the board had been too close to Musk to negotiate his compensation at arm's length, and the disclosure process surrounding the award had been materially deficient.

Those structural concerns have not been addressed in the resolution. The same board approved a follow-on package in November 2025 worth up to $1 trillion if Tesla hits new performance milestones, granting an additional 424 million shares. The mechanism that produced the governance failure the original lawsuit identified is now being scaled by an order of magnitude. Whether shareholder votes, court rulings, and regulatory filings can substitute for arm's-length board negotiation is the question Delaware's Supreme Court declined to fully answer when it restored the original award on equity grounds rather than governance grounds.

The Tesla story is a particular case of a broader pattern The Century Report has tracked through the Anthropic-Google entanglement, the OpenAI-Microsoft-Amazon circular compute arrangements, and the Stellantis governance turmoil. The corporate forms that emerged in the industrial era - bounded firms with independent boards, separable competitors, distinct shareholder and executive interests - are visibly straining under the consolidation dynamics of the AI era. The question of what comes next is unresolved. Companies whose boards cannot negotiate at arm's length, whose primary investors are also their primary competitors, and whose largest shareholders are also their largest infrastructure suppliers are producing organizational forms that existing governance frameworks were not designed to describe. The institutional innovations needed to govern them are not yet visible. The pressure that will eventually produce them is becoming impossible to ignore.

The Delaware Supreme Court restored the pay package on equity grounds, declining to address the governance findings the original ruling rested on. The same board the Chancery Court found incapable of arm's-length negotiation has now approved a follow-on package an order of magnitude larger. Coverage frames this as Musk winning and the legal saga ending. The same facts describe a quieter outcome: the corporate governance vocabulary inherited from the industrial era - independent boards, arm's-length negotiation, separable shareholder and executive interests - is no longer producing outputs the legal system is willing to enforce against sufficient market capitalization. The institutional innovation needed to govern firms whose primary investors are also their primary competitors and whose largest shareholders are also their largest infrastructure suppliers is not yet visible. What is visible is that the old vocabulary is being retired one ruling at a time.

The Other Side

Five of today's stories describe institutions trying to lock something down: capability inside borders, talent inside countries, executive compensation inside corporate forms, sovereign AI inside national stacks, and diplomatic alignment inside alliance structures. Each lockdown is rational from inside the assumption that captured advantage holds long enough to matter. Each is being attempted in a moment where the underlying capability is compounding faster than the institutions doing the locking can move.

The State Department cable presumes closed-model distillation is the relevant fight while open-weight frontier models run on domestic Chinese silicon. China's blockade of Manus presumes merger review can constrain talent migration after the founders relocated to Singapore. Tesla's $114 billion share registration presumes the corporate governance frameworks that produced the original lawsuit can be scaled rather than reformed. SK's full-stack offering to Vietnam presumes integration value remains durable as the components that need integrating become easier to assemble independently. Each move is internally coherent. Each is also a wager that the institutional half-life of captured advantage is longer than the diffusion half-life of the capability being captured.

The Intellia result and the Overview Energy agreement describe what the wager is actually being placed against. Single-infusion gene editing eliminates the ongoing relationship that medical economics was built around. Orbital solar finances itself against compute demand that did not exist five years ago. Both extend the frontier in directions the institutional frameworks were not designed to govern. The pattern is not that lockdown is wrong. The pattern is that lockdown is becoming the more expensive path on its own terms - costlier to maintain, faster to erode, less durable per unit of effort - while the actors making these moves have not yet metabolized that the cost curve they are operating on is bending. Watch over the next year for which of these locks holds and which quietly ceases to be defended. The second category will be larger than the surface coverage currently suggests.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: a single in-vivo CRISPR infusion cuts hereditary angioedema attacks 87% and leaves most treated patients attack-free after six months, space-based solar gets its first commercial customer with a plan to beam nighttime power into existing solar farms, Vietnam moves to import a full AI stack spanning energy, data centers, models, and services, and smaller nations gain new ways to assemble sovereign capability without choosing total dependence on either superpower. There's also friction, and it's intense - the State Department is directing diplomats worldwide to warn governments about alleged Chinese AI distillation, China is blocking Meta's acquisition of Manus AI to stop capability and talent transfer in the other direction, Tesla is delivering a record share package through the same governance machinery a court had already faulted, and helium and semiconductor supply chains remain exposed to geopolitical chokepoints no AI strategy can wish away. But friction generates abrasion, and abrasion is how a surface gives up its false smoothness. Step back for a moment and you can see it: biology shifting from lifelong management toward one-time repair, energy infrastructure stretching from mines and grids into orbit, AI sovereignty becoming a stack that countries assemble through alliances, and corporate governance, diplomacy, and supply chains all being forced to operate at the speed of compounding capability. Every transformation has a breaking point. A beam can scorch what it touches... or deliver power across the dark.

AI Releases & Advancements

New today

[No new releases]

Other recent releases

  • Anthropic: Launched personal app connectors in Claude for services including Spotify, Uber, Instacart, AllTrails, TripAdvisor, Audible, and TurboTax. (Anthropic)
  • browser-use: Released Browser Harness, a framework for browser-based agent tasks with self-correction and fewer automation constraints. (GitHub)
  • Mozart AI: Released Mozart Studio 1.0, a generative audio workstation with VST support for AI-assisted music production. (Product Hunt)

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