AI Ends Capitalism Before 2030

Good morning.

A few weeks ago someone told me : “Hey, I like reading your newsletter because I’m interested to see what mood you are in that week”.

I was taken aback by that comment, looked at this person and asked her what she meant by that. She clarified by saying that it seemed to her that I’m going in all directions in these “opinion pieces”.

To which I retorted : have you read the actual news that is lower down in the newsletter? Because it literally IS all over the place. The AI ‘experts’ out there also seem to be in opposite camps. And in truth, I don’t know either. I just find it very interesting to highlight these voices - even if they are alarmist there is always something to gather from it.

This week we’re going back to the scary timeline

You know Halloween was a few weeks ago, but this might be the scariest timeline for AI I've heard yet - and it's coming from someone who keeps making the rounds on every podcast telling anyone who'll listen.

Emad Mostaque. Former CEO of Stability AI. The guy who gave us Stable Diffusion and watched 300 million people download it. So this guy is doing the interview circuit with a message that makes my rice-buying panic look quaint. (I once told the story in this newsletter how two weeks before COVID broke out - I went out to the supermarket and bought 50kg of rice. If only I knew it was toiletpaper i should’ve bought)

His thesis is that all our jobs becomes "economically worthless" in a thousand days. Not fired. Not replaced. Worthless. As in negative value. As in the dumbest person on the team by default.

The scary thing about this prediction is that Emad Mostaque has predicted things correctly before.

Truth is - I've been dancing around this for years now, ping-ponging between "we'll be fine" and "start hoarding canned goods."

But Emad's doing the math. And the math is ugly.

The Numbers Don't Lie

First of all, in the US the idea or Universal Basic Income was brought up many times already. Everyone talks about like it's the obvious solution. Doesn't work. Can't work. Won't work.

$5.3 trillion to give every American $16,000 a year - poverty level, by the way. The entire US tax base is $4.9 trillion.

You can't pay for it. Even if you wanted to. Even if every politician suddenly became reasonable and competent. The math just doesn't work.

On top of that, the tax base is about to crater. Because when AI can do your job for 50 cents a day, why would anyone pay you a salary? Why would there be income tax to collect?

The Cost Curve That Breaks Everything

A cognitive worker replacement costs 50 cents per day. Not per hour. Per day.

That's at the current Grok 4 pricing for a million tokens. Which is roughly what it takes to replicate a day's worth of human cognitive output.

ChatGPT a year ago? About $100 for what you used. Today? Under a buck. That's a 100x drop in twelve months.

But nobody's getting laid off tomorrow. That's not how this works. Companies stop hiring first. They grow without adding headcount. Duolingo hit a billion in revenue growing 40% year-over-year - and they're not firing anyone. They're just not hiring anyone either.

One downturn. One recession. One bad quarter. Then the jobs go. And they never come back.

The Billionaires Are Terrified Too

Emad says the billionaires he talks to are terrified. Not cautiously optimistic or strategically positioned. Terrified.

They're buying data centers. Because in the intelligence age, GPUs are your comparative advantage. Not your workforce, your IP or your capital. Just compute. Raw, dumb, expensive compute.

And they know UBI doesn't work. They also know the math. They're just hoping we figure something out before the pitchforks come out.

(That's not me being dramatic - Emad literally mentions pitchforks. When the former CEO of a major AI company is casually dropping "assuming you don't get pitchforked" into interviews, maybe pay attention.)

The Wild Solution Nobody's Talking About

Emad Mostaque does offer a solution: a dual currency system. One currency backed by the compute that runs society - like Bitcoin but for the AI age. Another currency - "culture credits" - that you get just for being human.

The AIs want your human-issued currency. So they have to serve you to get it. It's alignment through economics instead of through training.

I don't know if this works. I honestly don't. But at least someone's doing the math instead of just saying "we'll figure it out" or "the market will adapt" like those are real plans.

We're Discovering, Not Inventing

The thing that stuck with me most - and I can't shake it - is Emad's point about discovery versus invention. We're not building these AI systems. We're discovering them.

The same equations that make Stable Diffusion work also describe protein folding. And economics. And complex systems. And maybe reality itself.

The golden ratio - that thing that makes art beautiful - is literally the mathematical balance between gradient flow and circular flow in these systems. It's not programmed in but it emerges.

We've taken the collective unconscious of humanity and compressed it into a few gigabytes. And now we're scaling it up. And maybe - just maybe - this is the Great Filter. The thing that explains why we don't see aliens. Because every intelligent species hits this point and... well.

Emad puts his P(doom) at 50%. That's "slightly better than a coin flip that we don't all die."

So What About You?

A thousand days from now, your job might be worthless. Not might be at risk. Might be worthless. Negative value.

The public sector jobs? Probably safe because nobody cares about productivity anyway. Jobs with human connection? Caregiving? Maybe safe. Everything else? Good luck.

But here's what makes this all so maddening - there's no good individual strategy. You can't out-skill this. Read it again: YOU.CAN.NOT.OUT-SKILL.THIS. You can't retrain faster than the models improve. You can't even really prepare because the timeline is compressed and the solutions don't exist yet.

Maybe we build this civic AI infrastructure. Maybe we figure out new economics for the intelligence age. Maybe we get really lucky and the alignment problem solves itself somehow.

Or maybe in a thousand days we find out the hard way what happens when capital doesn't need labor anymore and we didn't build the social infrastructure to handle it.

Either way - and I know this sounds dramatic but I don't know how else to say it - the world's not going to be the same again. It hasn't been for a few years. In a thousand days it changes completely.

Are you ready? Am I? Is anyone?

What's your plan?

Welcome to the Blacklynx Brief

AI News

• Alibaba-backed Moonshot AI released Kimi K2 Thinking, an open-source reasoning model that rivals or surpasses GPT-5 and Claude 4.5 Sonnet across key benchmarks at a fraction of the cost. It achieved a record 44.9% on Humanity’s Last Exam and can autonomously chain 200–300 tool calls, with training costs under $5 million. The release marks China’s closest approach yet to frontier-level AI at significantly lower prices.

• OpenAI CFO Sarah Friar walked back comments suggesting the company sought federal guarantees for its infrastructure spending, after backlash over potential “AI bailouts.” Sam Altman clarified that OpenAI opposes government backstops, saying private firms should be allowed to fail. The episode reignited debate over OpenAI’s rapid spending and growing dominance in the AI sector.

• Microsoft AI CEO Mustafa Suleyman announced the MAI Superintelligence Team, focused on solving targeted challenges like medical research and clean energy rather than pursuing open-ended AGI. Suleyman described the effort as building “Humanist Superintelligence” — AI that serves humanity’s needs. The team includes top researchers from DeepMind, Anthropic, and OpenAI, marking Microsoft’s first distinct identity in superintelligence research.

• Adobe unveiled a unified AI strategy at MAX 2025, combining its top creative models for image, audio, and video into one plan and introducing conversational, agent-powered experiences across Firefly, Photoshop, and Express. The company aims to make AI a true creative collaborator, blending human imagination with machine capability for faster, more intuitive content creation.

• The Firefly app is evolving into a full creative AI studio, integrating third-party and in-house models for end-to-end production in one space. It now supports pro-level video, audio, and image tools, plus a new Firefly Image Model 5 capable of generating 4MP photorealistic visuals. Firefly’s models remain trained only on licensed data, ensuring commercial safety for creators.

• Adobe is embedding agentic AI assistants directly into its core apps, turning them into teammates that automate workflows, apply edits, and coordinate projects. A preview of Project Moonlight showed an orchestration assistant linking tasks across Adobe’s ecosystem. These agents aim to free creators from repetitive work, speeding up production and ideation.

• Through new partnerships with Google and YouTube, Adobe will integrate Google’s Gemini, Veo, and Imagen models into its apps and allow enterprise customers to customize them for branded content. The YouTube collaboration enables creators to use Premiere Mobile with Shorts templates for quick, professional mobile editing.

• Adobe leaders say the defining skill for the AI era is creative direction—guiding technology with vision and style rather than technical know-how. As AI handles the mechanics, creators who embrace it as a partner in imagination and storytelling will stand out, keeping human creativity and emotion at the heart of the process.

• Dr. Fei-Fei Li published an essay arguing that the next big leap in AI will come from spatial intelligence — systems that understand and generate realistic 3D worlds consistent with physics. She says this ability to reason about space, motion, and cause-and-effect is essential for bridging AI from language to perception and action. Such “world models” could transform fields like robotics, healthcare, and science by allowing AI to think and predict in the real world.

• Anthropic projects it will spend far less on compute than OpenAI, estimating $6B for 2025 and $27B by 2028, compared to OpenAI’s $15B and $111B. The savings come from using a mix of chips from Amazon, Nvidia, and Google instead of relying on a single vendor. The company expects to be profitable by 2027, signaling a more efficient, enterprise-focused approach compared to OpenAI’s aggressive, product-driven strategy.

• GPT-5 became the first AI model to fully solve a 9x9 Sudoku puzzle, according to Sakana AI’s new Sudoku-Bench benchmark. It achieved a 33% solve rate across complex puzzles that test deep reasoning and spatial logic, roughly doubling the previous record. The milestone highlights improved structured reasoning but also underscores AI’s remaining gap in creative, human-like problem solving.

• Yann LeCun, Meta’s longtime Chief AI Scientist and Turing Award winner, is reportedly preparing to leave the company to start his own AI venture focused on world models that learn from video and spatial data. His departure follows Meta’s major AI restructuring and reported tension between FAIR (which he led) and the company’s new direction under Alexandr Wang. The move marks the end of a decade-long era for Meta’s foundational AI research team.

• ElevenLabs launched its Iconic Voice Marketplace, offering brands licensed access to AI-generated voices of celebrities and historical figures. The platform includes living voices like Michael Caine and Liza Minnelli, as well as recreated legends such as Maya Angelou and Alan Turing, all through official estate partnerships. The marketplace aims to create an ethical model for voice cloning, blending nostalgia, celebrity IP, and commercial AI audio.

• SoftBank sold its entire $5.8 billion Nvidia stake to fund a new round of AI investments centered on OpenAI, with CEO Masayoshi Son calling the company “the most valuable to come.” The sale continues Son’s history of massive, high-risk bets, with SoftBank planning over $33 billion in AI-related investments. Despite exiting Nvidia, SoftBank’s backing of OpenAI still indirectly fuels Nvidia demand, keeping the AI ecosystem tightly interlinked.

• Fei-Fei Li’s World Labs released Marble, its first commercial world model that creates persistent 3D environments from text, images, videos, or layouts. Users can edit or merge worlds and export them for gaming, VFX, or VR, with plans starting at $20 a month. Marble builds on Li’s vision of spatial intelligence, moving AI closer to understanding and generating physically grounded worlds.

• OpenAI launched GPT-5.1, a conversational update adding personality presets and faster reasoning. Users can now choose from eight tone options—like Professional, Friendly, or Quirky—and tweak settings for emoji use and warmth. The quieter rollout suggests OpenAI may be shifting toward smaller, more frequent updates as competition with Google’s Gemini 3 heats up.

• OpenAI is appealing a court order requiring it to turn over 20 million anonymized ChatGPT conversations to The New York Times in an ongoing copyright case. The company argues the order violates user privacy and that most data is irrelevant to the claims. Despite its objections, the decision highlights growing legal and ethical tensions around user data, transparency, and AI training practices.

Quickfire News

• Nvidia CEO Jensen Huang said China is “nanoseconds behind America in AI” and predicted that China will “win the AI race.”

• Perplexity released an update for its Comet AI assistant, adding better web interaction, multiple tab use, and faster performance.

• A new event called the State of Code Roundtable is happening in New York City on November 13 to discuss how AI affects coding productivity and quality.

• Google announced its Ironwood TPU AI chips will arrive within weeks; Anthropic plans to use one million of them for training and running Claude

• SoftBank Group and OpenAI formed a joint company called SB OAI Japan, aiming to launch “Crystal intelligence,” an AI system for businesses, in 2026.

• Intel’s CTO and AI head, Sachin Katti, left the company to join OpenAI, with CEO Lip-Bu Tan now leading Intel’s AI and advanced technology teams.

• Legal AI company Clio raised $500 million in Series G funding, valuing it at $5 billion.

• Gamma, a tool for making AI-generated presentations, websites, and posts, reached $100 million in yearly rev

• enue and raised $68 million at a $2.1 billion valuation.

• OpenAI began offering one free year of ChatGPT Plus to U.S. servicemembers and veterans who left active duty within the past year.

• Time magazine launched an AI agent that lets users search its 102-year archive and create text and audio summaries.

• The Wikimedia Foundation urged AI companies to credit and support Wikipedia financially, emphasizing its value in AI training data.

• OpenAI lost a copyright case in Munich, where the court found ChatGPT had been trained illegally on songs; the ruling was described as Europe’s first major AI legal decision.

• Google announced Private AI Compute, a secure cloud platform using Gemini models to process data privately through hardware isolation.

• Samsung introduced its Vision AI Companion for 2025 TVs, combining an improved Bixby with Copilot and Perplexity for conversational functions.

• ElevenLabs released Scribe v2 Realtime, a transcription system supporting 90 languages with top accuracy and live comprehension for agents.

• Blue Owl Capital committed $3 billion to a New Mexico data center for OpenAI’s Stargate, with banks planning an additional $18 billion for the 4.5-gigawatt project.

• Baidu introduced ERNIE-4.5-VL-28B-A3B-Thinking, a multimodal reasoning model said to outperform GPT-5 high and Gemini 2.5 Pro in visual tests.

• Weibo AI made VibeThinker-1.5B open-source, a small model trained for about $7,800 that performs well in math, coding, and reasoning tasks.

• Anthropic announced a $50 billion investment to expand U.S. AI infrastructure, with its first custom data centers planned for Texas and New York in 2026.

• ByteDance’s Volcano Engine released Doubao-Seed-Code, an AI coding assistant that matches Claude Sonnet in benchmarks while being 60% cheaper.

• Even Realities presented G2 smart glasses, featuring a “spatial display” for projecting information at different visual depths and an R1 control ring for gesture

Closing Thoughts

That’s it for us this week.