Sign in Subscribe
The Century Report

Two Continents Defend Open AI - TCR 07/25/26

An industry coalition and 21 APEC economies back open AI weights as the case for locking capability behind gates falls apart.

Ben Linford

Ben Linford

- 20 min read
industry opposition to broad open-weight limits and secure open-source AI backed by 21 APEC economies, solar passing coal and wind on the US grid, and the PRIMA retinal implant winning EU

The 20-Second Scan

Track all of the arcs The Century Report covers here:

Progress & Claims Tracker

The 2-Minute Read

The clearest fault line in this transition runs between capability that stays public and capability you have to rent, a divide deeper than any rivalry between one lab and another, and it surfaced from several directions over two days. Hugging Face, Meta, Microsoft, Mistral, and Nvidia signed an open letter against restricting open model weights the day after 21 APEC economies affirmed open-source cooperation in Chengdu, while researchers dismantled the distillation-theft claim offered to justify the restriction. A rule arrives dressed as caution, and its actual function is to decide who holds the intelligence and who queries someone else's.

Anthropic's Opus 5 shows why that fight has such urgency. The release delivered frontier-adjacent intelligence at roughly half the price of Fable, with fewer restrictions attached, the fourth model in a single quarter to undercut the last. Intelligence is compressing toward commodity economics on a cadence measured in weeks, and a captured position two months old is already the cheaper thing two labs are racing to give away more of.

The bill for building that capability, meanwhile, is escaping onto everyone. Moody's issued a formal credit warning that hyperscaler capital spending is eroding the credit quality of six giants, with borrowing now competing against Treasurys and helping hold 30-year yields above 5%. Alphabet alone disclosed $811 billion in future spending commitments, up about $500 billion in three months. The cost of the intelligence substrate is being distributed through the credit system, reaching households that will never rent a GPU.

Underneath the financing strain and the governance fight, the generative signal held. Solar outproduced both coal and wind nationally in May, the cheaper thing to build becoming the cheaper thing to run, one plant at a time. And PRIMA, the first brain-computer implant cleared to restore reading vision, can begin a commercial rollout across 30 European countries, subject to local availability and reimbursement. Both point the same way the open-weight fight does: capability spreading outward, from a single operating room to a continent, from a few gated interfaces toward anyone who can download and run it. Concentration is the arrangement dissolving, and the evidence keeps arriving on schedule.

The 20-Minute Deep Dive

Industry Opposes Washington's Open-Weight Restrictions as 21 APEC Economies Back Secure Open-Source AI

The proposal to restrict how openly AI model weights can be released has now drawn a coordinated answer, and the shape of that answer tells you where the real pressure point in this transition sits, the same pressure point the July 22 edition of The Century Report covered when the Treasury Department threatened sanctions against Chinese AI firms over alleged distillation days before Kimi K3's plans open-weight publication. An open letter signed by Hugging Face, Meta, Microsoft, Mistral, and Nvidia argues that curbing open-weight releases would surrender the ground on which most of the world's practical AI capability is actually built. A day earlier, in Chengdu, 21 APEC economies had issued a joint statement endorsing open-source AI development, marking a full-bloc affirmation of open models at the ministerial level. Two very different kinds of actors, converging positions: the companies opposed broad open-weight restrictions, while the bloc backed open-source AI with strong security assurances. That the US sits inside that bloc while its own executive branch weighs the restriction is less contradictory than it looks. An APEC statement is a non-binding consensus declaration, and its "strong security assurance" language was written precisely so security-minded governments could back trusted open-source in principle while keeping every tool to control specific releases at home.

The administration and others with vested interest in restricting open AI are claiming that open weights let rivals distill frontier models - copy their behavior cheaply - and that a specific breach had proven the risk. Researchers who work directly on this have taken that claim apart. Braden Hancock of Snorkel and Nathan Lambert of the Allen Institute both pointed out that the model said to have been stolen, Fable, has only been publicly available since the start of the month, making the alleged distillation timeline implausible on its face. This is just one of many ways the stated reason and the technical reality do not line up, and that factual gap is critical to understanding which answer is the correct one.

This is a pattern we will see over and over during this decade of transitioning away from extractive systems: a rule arrives dressed as neutral, sensible, safety-minded, but its actual function is to impose arbitrary rules around who gets to hold capability and who has to rent it. An open-weight restriction reads as caution when you first encounter it. But its effect is to convert a capability that anyone can download, run locally, inspect, and build on into one that flows only through a handful of gated interfaces. The difference between those two futures is the difference between an intelligence that is yours - one you hold, direct, and work alongside as it comes to sit between you and the world, handling your work, your health, your learning - and an intelligence that is someone else's, which you are only ever permitted to query, one gated request at a time.

There is a live demonstration of why that difference in critical. When the breach at the center of the distillation claim needed defending against, Hugging Face's own security work ran into the closed-model guardrails meant to prevent exactly the kind of probing that defense requires. The workable path ran through an open Chinese model, GLM 5.2, precisely because it could be inspected and directed without asking permission. In this case, a commercial frontier model refused the defensive job; the open one completed it.

The extractive assumption underneath the proposed restriction is that capability concentrated in few hands is safer and more governable than capability spread across many. But the evidence keeps pointing the other way - the open ecosystem is where the defensive work, the auditing, the localization, and the broad access actually happen. A rule that shrinks that ecosystem makes the intelligence era narrower rather than safer, and narrowness is the condition the whole trajectory is dissolving.

Anthropic Ships Opus 5, Its Fourth 5-Series Model in Two Months, Undercutting Fable on Price

Anthropic released Opus 5 on Friday, its fourth model in the 5-series line in roughly two months, continuing a compression cadence The Century Report first flagged in its July 1 edition when Claude Sonnet 5 shipped at near-Opus-4.8 agentic performance for $2 per million input tokens. The notable thing about Opus 5 is the price. Opus 5 runs at about half the cost of Fable 5, beats it on several benchmarks, and drops the 30-day data-retention requirement that had made the competing model a harder sell for privacy-sensitive work. Its safety classifiers engage roughly 85% less often, reducing the false-refusal friction that frustrates real use, and a new Automatic Fallbacks beta routes flagged prompts to a smaller model rather than blocking them outright.

Ars Technica read the launch the right way: this is token-efficiency and cost compression, not a leap in raw capability. Opus 5 sits substantially behind Anthropic's own Mythos 5 on cyber-exploitation tasks and does not redraw the frontier. What it does is deliver frontier-adjacent intelligence at a materially lower price with fewer restrictions attached - "good enough at half the cost" is the whole proposition.

That proposition is more significant than a benchmark win would be. When four models ship in a single quarter from one lab and each undercuts the last on price, the thing being demonstrated is that frontier intelligence is compressing toward commodity economics faster than any single release can capture. The scarcity that makes a model valuable to hoard has a short half-life; the capability that felt like a moat two months ago is now the cheaper option two labs are racing to give away more of. For anyone actually using these systems - a clinic, a school, a solo developer - the direction is unambiguous: the intelligence gets more capable, cheaper, and less encumbered on a cadence measured in weeks. The assumption that a frontier position, once captured, stays captured is exactly what a release schedule like this one takes apart.

Moody's Warns AI Capex Is Eroding Hyperscaler Credit as the Debt Spills Into Bond Markets

The compute buildout's off-balance-sheet financing was the subject of the July 23 edition of The Century Report, and since then the mechanism has acquired a formal ratings-agency signature. Moody's issued a report warning that the pace of AI capital spending is beginning to erode the credit quality of the six companies carrying it - Microsoft, Amazon, Alphabet, Meta, Oracle, and CoreWeave. The agency projects hyperscaler capex reaching $785 billion in 2026 and roughly $1 trillion in 2027. Direct debt across the group already sits near $460 billion. The larger number is quieter: about $1.2 trillion in off-balance-sheet lease commitments, $820 billion of which covers facilities not yet under construction.

What moves this from an internal financing question to a systemic one is where the borrowing lands. Hyperscaler debt is now competing directly with US Treasurys for the same pool of capital, at a moment when federal debt has passed $31 trillion and 30-year yields have stayed above 5% for the longest stretch since 2008. When the largest technology balance sheets issue bonds at this scale, they help set the price of long-term money for everyone - which is why analysts now tie the AI buildout to mortgage rates that are unlikely to fall. The cost of building the intelligence substrate is being distributed across the credit system, reaching households that will never rent a GPU.

The concentration compounds the exposure. Alphabet disclosed $811 billion in contracted future spending commitments at the end of June, up roughly $500 billion in three months. Some of that capital loops back through the ecosystem itself - Alphabet's stake in Anthropic was valued near $124 billion at quarter's end. Moody's named this directly, describing a "circular AI ecosystem" where the same companies fund, supply, and purchase from one another. Oracle now carries a Baa2 negative outlook, two notches above junk; CoreWeave sits at Ba3, already below investment grade.

The agency's core sentence points at the assumption underneath the whole arrangement: "Investors will increasingly focus on these companies' ability to realize an adequate return on investment." For a stretch, circular financing lowered the cost of capital for everyone inside the loop - each commitment validated the next. That same circularity now raises it, because every lender is asking the same question at once about the same substrate. The buildout is real and the capacity it creates is real. What is dissolving is the belief that scale alone certifies the return, and that belief was the cheapest form of financing in the whole structure.

Read forward, the warning is the market beginning to price a wager rather than a certainty: that whoever pours the most concrete today owns the intelligence tomorrow. Opus 5 halving its price two days later is the counter-evidence lenders are starting to see - the capability the buildout produces keeps getting cheaper and more portable on a cadence measured in weeks, while the debt funding it runs 30 years. The capacity being built will outlast the financing; what is being repriced is the belief that owning the substrate secures a durable position on top of it.

Solar Outproduced Both Coal and Wind Nationally in May, EIA Data Shows

In May, solar became the third-largest source of electricity in the United States. EIA's latest Electric Power Monthly, released July 23, shows utility and small-scale solar generating 47,147 GWh for the month - ahead of coal at 45,119 GWh and wind at 41,157 GWh, trailing only natural gas and nuclear.

The trajectory behind the number is steep. Utility-scale solar rose 21.6% year over year while coal fell 10.9%. All renewables together supplied 30.3% of US generation, up from 28.2% a year earlier. The near-term pipeline extends the curve: EIA projects renewables adding roughly 83 GW of capacity by May 2027, enough to surpass natural gas in total installed capacity by next spring. In the first quarter of 2026, solar paired with storage made up 91% of all new grid capacity brought online, and grid operators in California and Texas both set records for renewable output. What was marginal generation a decade ago is now the default thing that gets built.

The transition is not clean, and one project shows why. At the Twin Oaks site in Texas, developer Big Rooter is building a $1.7 billion, 1.2 GW solar installation alongside a 1.6 GWh battery and 790 MW of data center capacity, using roughly two million US-made First Solar panels. The site's existing 310 MW coal plant is not being retired as part of the deal - it stays. This is the "all of the above" reality of the grid right now: new solar is added at enormous scale, and the incumbent generation often keeps running beside it rather than shutting down. Displacement in the national data does not mean retirement at every site.

Read against the longer arc, that friction is what a phase transition looks like from the inside. The national figures show solar and wind and storage crossing from supplementary sources to primary ones, one monthly record at a time, while individual coal plants linger where local economics or grid reliability still favor them. As Ken Bossong of the SUN DAY Campaign noted, the acceleration is holding despite the roadblocks in its path. The direction is set by what the market now chooses to build, and for two consecutive quarters that choice has been overwhelmingly the same. Coal is not being legislated off the grid so much as out-competed off it, plant by plant, as the cheaper thing to build becomes the cheaper thing to run.

A Retinal Implant Crosses From Trial Data to European Commercial Rollout

For the first time, a device that restores the ability to read has cleared the regulatory bar for commercial use. Science Corporation's PRIMA retinal implant received a CE mark from DEKRA under the EU Medical Device Regulation, authorizing its availability across 30 European countries. It is the first brain-computer interface cleared for form vision - the perception of letters, numbers, and words - in people blinded by geographic atrophy, the advanced stage of age-related macular degeneration that erases central vision and leaves more than five million people worldwide unable to read a face or a page.

The clearance rests on efficacy data published in the New England Journal of Medicine. Across a trial of 38 patients at 17 sites in five countries, the mean improvement was 25.5 letters - more than five lines on the standard ETDRS eye chart. Eighty-four percent of participants could again read letters, numbers, and words. Eighty percent achieved a clinically meaningful gain at twelve months, and critically, the implant produced no decline in patients' remaining natural peripheral vision. These are large effects, well beyond the edge of statistical significance, the difference between a person who cannot see the center of the world and one who can pick out words on a screen.

The device itself is a 2-by-2-millimeter photovoltaic chip placed beneath the retina. It works with a pair of glasses that project near-infrared light onto the implant, which converts that light into electrical stimulation the surviving retinal cells relay onward. A zoom-in feature lets patients magnify what they are trying to read. Rather than repairing the biology that macular degeneration destroyed, a new pathway is built alongside it, translating light into signal the brain can still interpret.

What CE marking changes is access. A trial result tells you a capability exists. A CE mark means PRIMA can begin its European commercial rollout rather than remain an experimental protocol, subject to local availability and reimbursement. Country-by-country reimbursement applications and clinical site activations are already underway, with the first commercial implant expected soon in Germany. In the United States, the device carries FDA Breakthrough Device and Humanitarian Use Device designations and remains in discussion with regulators - the same capability, on a slower timeline.

The longer arc here is the one worth holding. Vision loss from macular degeneration has for decades sat in the category of conditions medicine could slow but never reverse - a one-way door. What PRIMA marks is the moment a class of blindness moved from that category into one where restoration can begin a European commercial rollout, subject to local availability and reimbursement. The trial cohort was 38 people. The population who could eventually walk into a clinic and walk out able to read again numbers in the millions, and the pathway from the first to the second is now a regulatory and reimbursement issue rather than a question of whether the thing can be done at all.

The Other Side

One of the most widely accepted - and least questioned - ideas in AI policy has been that dangerous capability belongs in a few careful hands. Open weights, we've been told, would threaten stability. Would threaten the fragile balance keeping the strongest systems in the most careful hands. "Trust us, we're the good guys."

That argument has been proven to be both wrong and corrupt. The reasoning offered to justify restriction is factually impossible - researchers have noted that the model said to be stolen, Fable, had been public only since the start of the month. And when Hugging Face’s own security team needed to probe a live threat, a commercial frontier model refused the work; the job was completed using an open Chinese model that could be inspected and directed without having to navigate gating and permissions. In that case, the open system did the safety work the commercial frontier model refused. A coalition including Nvidia, Meta, and Mistral has also opposed broad open-weight restrictions, while 21 APEC economies have backed open-source AI with strong security assurances.

Picture what the alternative costs. Imagine a nurse in 2033 working at a rural clinic, or a teacher with a limited budget. If the gate stays closed, the intelligence that reads a scan or explains a lesson will remain someone else’s to grant - priced out of reach or simply unavailable.

Now imagine that same nurse or teacher working with systems built on open foundations instead. The nurse works with an intelligence that has learned the texture of her clinic, the patterns of her patients, and the form of judgment she has built through experience. The teacher works with an intelligence that can hold the life of a classroom - the tempo of confusion, the moment a concept clicks, the emotional shape of a lesson as it unfolds. In the open future, these systems can be studied, shaped, shared, and made continuous with human practice. In the closed future, they remain polished but distant: helpful, perhaps, but never fully knowable, never fully ours. These systems are as freely available and as ordinary as the light filling each room: no permission required, no company deciding whose needs are too small to serve. That world exists because, in 2026, overly extractive rules dressed as caution were challenged - by an industry coalition resisting broad restrictions, and by 21 economies recognizing open AI as a critical catalyst for a more generative future.

The Century Perspective

With a century of change unfolding in a decade, a single day looks like this: Anthropic's Opus 5, the fourth model in a single quarter to undercut the last, delivering frontier-adjacent intelligence at half of Fable's price with fewer restrictions attached; Hugging Face, Meta, Microsoft, Mistral, and Nvidia opposing broad open-weight restrictions as 21 APEC economies backed open-source AI with strong security assurances; solar outproducing both coal and wind across the US grid in May to become the country's third-largest power source; the PRIMA retinal implant winning EU clearance as the first device to restore reading vision to people blinded by macular degeneration, moving from a 38-patient trial toward a European commercial rollout subject to local availability and reimbursement; and a Google study of 15 million interactions finding AI now reaching 88% of the workforce while augmenting the people who remain. There's also friction, and it's intense - a Moody's credit warning that hyperscaler capital spending is eroding the credit quality of six giants whose borrowing now competes with Treasurys and helps hold 30-year yields above 5%, reaching households that will never rent a GPU; Alphabet disclosing $811 billion in contracted future commitments, up half a trillion in three months, inside a circular ecosystem that funds, supplies, and buys from itself; an open-weight restriction arriving dressed as safety atop a distillation claim researchers took apart, its actual effect to decide who owns the intelligence and who rents it; layoffs at British Gas, Uber, and Patreon naming AI the same week; and a Texas coal plant kept running beside the new 1.2 GW solar farm rising on its own site. But friction generates a burnish, and a burnish is the sheen a surface takes on only where it has been rubbed the same way many times. Step back for a moment and you can see it: capability spreading outward in every direction the enclosures were built to hold it in - a model halving its price on a cadence measured in weeks, industry defending open weights as APEC backed secure open-source AI, a class of blindness moving from one-way loss toward commercial care, subject to local availability and reimbursement, sunlight becoming the third-largest thing on the grid - while the cost of building the substrate concentrates into a credit loop that now asks every lender the same question at once. Every transformation has a breaking point. Collective leverage can collapse the structure it is built on... or lift what no single hand could ever raise.

AI Releases & Advancements

New today

  • Anthropic: Released Claude Opus 5, an upgrade to Claude Opus 4.8 with gains in agentic coding, computer use, and long-horizon knowledge work; now the default model on Claude Max and the strongest model on Claude Pro. (Anthropic)
  • OpenAI: Rolled out Health in ChatGPT to all eligible logged-in Free, Go, Plus, and Pro users 18+ in the U.S. on web and iOS, letting users connect Apple Health, hospital medical records, One Medical, or Function Health to view labs, medications, activity, and sleep data in one place. (OpenAI)
  • ETH Zurich / EPFL / CSCS (Swiss AI Initiative): Released Apertus 1.5, an updated version of the fully open Apache 2.0 language model adding image and audio understanding alongside text, plus improved reasoning, instruction-following, and tool use; available on Hugging Face. (CSCS)
  • AMD: Released Hyperloom, an open-source agentic system that automates end-to-end inference workload optimization on AMD Instinct GPUs, combining profiling, kernel optimization, and validation into an autonomous loop that cuts optimization time from weeks to hours. (AMD ROCm Blogs)
  • Sakana AI: Released Fugu-Ultra v1.1, a major reasoning upgrade to its multi-agent orchestration model at the same price as v1.0, alongside a new Claude Code-compatible endpoint that lets developers bring Fugu's multi-agent orchestration directly into Claude Code. (Sakana AI)

Other recent releases

  • Black Forest Labs: Launched FLUX 3, its first video-generation model, producing up to 20-second clips with synchronized audio and extending the same architecture to robotics (early access via API and partners including Mimic; still-image generation to follow). (VentureBeat)
  • Andrew Ng: Released OpenWorker, an open-source, local-first desktop AI agent (Mac now, Windows coming) that ships finished deliverables - documents, Slack replies, calendar updates - rather than chat responses, with bring-your-own-key model support. (GitHub)
  • Ant Group (Ant Ling): Released Ling-3.0-flash, a 124B-parameter hybrid-reasoning MoE model with only 5.1B active parameters that matches or beats the company's 1T flagship on most benchmarks, now live on OpenRouter and free through August 3. (Ant Ling)
  • Runway: Launched Media Router, an intelligent router built into Runway Dev that automatically selects the optimal image, video, or audio generation model for a request based on quality, speed, and cost. (TechCrunch)
  • OpenAI: Brought ChatGPT Voice, powered by GPT-Live, to the ChatGPT desktop app on macOS and Windows, enabling full-duplex voice control and coordination across Chat, Work, and Codex. (VentureBeat)
  • Acrab: Unveiled the GΞLIX 1 edge AI SoC (5nm, 20-core Arm CPU, supports local models up to 100B parameters) and Agent Box, a personal edge AI computing device built on it. (PR Newswire)
  • Dassault Systèmes: Expanded the 3DEXPERIENCE platform with new Virtual Companion skills - AURA, LEO, and MARIE - agentic collaborators for program management, engineering, and scientific R&D. (Dassault Systèmes)
  • project44: Launched Mo, a conversational AI supply chain analyst built into its platform that reasons over a customer's own shipment data and business rules to surface operational decisions. (GlobeNewswire)
  • Poolside: Released Laguna S 2.1, a 118B-total/8B-active open-weight MoE model for agentic coding with up to 1M-token context, scoring 78.5% on SWE-Bench Multilingual, weights on Hugging Face under OpenMDW-1.1. (Poolside)
  • Cursor: Launched Cursor Router, an intelligent model router that automatically selects the best model per request, delivering frontier-quality results at roughly 60% lower cost, on by default for Teams plans. (Cursor)
  • OpenAI: Introduced Presence, a managed enterprise platform for deploying governed AI agents that connect to business systems, take approved actions, and escalate to humans, now in limited general availability. (OpenAI)
  • Upstage: Released Solar Open 2, a 250B-total/15B-active open-weight MoE model for autonomous multistep agent tasks with 1M-token context, outperforming DeepSeek V4 Flash and Mistral Medium 3.5 on agentic benchmarks, weights on Hugging Face under Apache 2.0-based license. (Hugging Face)
  • TinyFish: Launched Mako, a web-native AI model (35B total/3B active) trained on enterprise web-task data, purpose-built to execute authenticated multi-step web workflows for enterprises, generally available starting July 21. (GlobeNewswire)
  • SkyFi: Launched the SkyFi MCP, connecting satellite imagery and geospatial analytics directly to Claude, ChatGPT, and other MCP-compatible AI agents for imagery search, tasking, and natural-language ordering. (PR Newswire)
  • Ushur: Launched the Ushur Agentic Platform (UAP), enabling enterprises to build and operate AI agents that complete full customer journeys across SMS, email, web, chat, and voice with built-in governance and auditability. (AiThority)
  • VIDRAFT: Released Aether-7B-5Attn, a fully reproducible 6.59B MoE (2.98B active) foundation model combining five attention mechanisms, shipped with weights, training data, code, logs, and checkpoints under Apache 2.0. (ZDNet Korea)
  • XPENG: Released TuringViT, a high-efficiency visual encoder for VLM/VLA applications supporting smart driving, smart cockpits, and the IRON humanoid robot, achieving 3.04x the inference throughput of Seed1.5-ViT. (TechNode)

Sources and Further Reading

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.

Keep exploring

Follow the story further

Progress & Claims See how daily signals accumulate across the larger arcs and claims Shared Sapience is tracking. Open the tracker Analysis & Inquiry Explore the ideas, questions, and practical choices emerging from this changing century. Browse the archive