Elon Musk's Metastate of Mega-Tunnels — complete with spreadsheets, politics, etc: escapist.city @boringcompany is still my favorite adjective 🔥 @elonmusk your next 13 Boring Company projects just got pitched O_o

The Compute Standard: Why the AI Token is Eclipsing the Silicon Empire? In the 1930s, the Technocracy movement proposed a radical economic paradigm shift: abolishing traditional fiat money to replace it with a currency backed entirely by physical energy. Supporters of this vision, most notably Elon Musk's grandfather Joshua Haldeman, advocated for an economy where the fundamental unit of exchange was measured in ergs, joules, or kilowatt-hours. In this system, prices would reflect the exact energy required to produce goods, effectively eliminating arbitrary inflation because the money supply would be anchored to the immutable laws of physics. Nearly a century later, this esoteric vision is materializing not through political revolution, but through the structural evolution of artificial intelligence. In 2026, the global economy is quietly transitioning to a new standard of value: the AI token. NVIDIA sits at the apex of this transition with a staggering $5 trillion valuation, acting as the de facto central bank of the early AI era. Yet, the company is walking into a classic innovator's trap engineered by its own perfection. The architecture of NVIDIA's dominance is built upon a 75% gross margin, which essentially functions as a massive tax on the hyperscalers and frontier labs purchasing their hardware. The nature of AI computation, however, has fundamentally changed. The industry's center of gravity has shifted from training massive frontier models—a task requiring maximum precision and bandwidth—to inference, which is the continuous, high-volume process of serving these models in production. Inference is relentlessly cost-sensitive. The primary metric of success is no longer raw supercomputing power, but the cost, latency, and power consumption required to generate a single token. Because of this shift, the "AI token" is evolving into a form of programmable money. We are witnessing the rise of Decentralized Physical Infrastructure Networks (DePIN), which operate as an "Airbnb for AI". These decentralized networks aggregate millions of idle prosumer GPUs and convert them into a unified, fluid inference network. By leveraging amortized hardware, DePINs can offer compute capacity at 30% to 50% of the cost of traditional centralized hyperscaler clouds. The bedrock of this new tokenized economy is energy, specifically brownfield energy assets. The binding constraint on AI scaling is no longer the supply of silicon chips, but the sheer availability of electrical power. To bypass decade-long grid interconnection queues, operators are acquiring retired industrial sites, such as decommissioned coal plants and aluminum smelters, to co-locate compute directly at the power source. This fusion of stranded energy assets and decentralized compute capacity creates the ultimate substrate for minting the new currency: raw electricity converted seamlessly into inference tokens. NVIDIA is structurally paralyzed and unable to participate in this decentralized financialization of compute. The company's internal resource allocation and public-market valuation are entirely tethered to defending its premium 75% gross margins. Pursuing DePIN orchestration or operating at the 20% to 30% margins of the decentralized compute market would require an autonomous spin-out unit, an organizational leap that comfortable incumbents almost never survive. Meanwhile, universal software abstraction layers like OpenAI's Triton and vLLM are making the AI software stack hardware-agnostic, bypassing the lock-in of NVIDIA's once-impenetrable CUDA ecosystem. The inference workload is becoming entirely portable. When inference is portable, the token becomes a liquid, interchangeable asset. The future belongs not to the monopolists of the silicon wafer, but to the orchestrators of the compute standard. As envisioned by early proponents of energy-backed economies, the integration of physical energy generation and digital labor is forging a reality where the kilowatt-hour and the AI token are fundamentally indistinguishable. NVIDIA built the forge, but they are structurally priced out of adopting the very currency it mints.

What we can do more with @worldnetwork open.substack.com/pub/slavas… and @worldcoinfnd

Wes Anderson should make #Hašek's 'The Good Soldier Švejk': solodkiy.lol/ Why the hell not? Pastel trenches. Symmetrical court-martials. #Švejk’s cheerful “yes sir” weaponized into bureaucratic Armageddon while the empire collapses in perfect dollhouse symmetry, and modern geopolitical assymetry. Andersonian in form. Hašekian in content. Malicious compliance never looked this stylish. @AccidentalWes @QuotingWes @WesAndersonfc @rWesAnderson @SvejkSoldat @Svejk @GoodSvejk @chmnkhbt what do you think?:) there is notable X-post from 2023 by a Catalan publisher account @LesMalesHerbes that explicitly compares a book (Miràbilis) to the styles of both #WesAnderson and #JaroslavHašek

Buy TWO #AI gadgets on @Kickstarter: Wake up with a secret THIRD one for free:) @TiinyAILab Tiiny Pocket Lab does the heavy uncensored thinking. @DecoKeeTech Decokee Quake becomes its face, hands and soul. No laptop. No cloud. Just a book-sized offline AI station that actually belongs to you. I already explained the cheat code 👇 open.substack.com/pub/slavas…

My #OlaresOne just landed in London, and I actually need to take a second to thank the @Olares_OS founders and the team behind it. You can always tell when tech is built by people who genuinely care about the DETAILS. From the physical packaging down to the UX with LaresPass, it’s just incredibly well thought out. The device itself looks like a piece of art sitting on my desk. With my new friend, this 'remote smart active personal AI-first/focused server,' the reality is that while it’s clearly built for hardcore AI enthusiasts, it’s completely painless (!) for a curious beginner like me who just wants to learn without breaking things. The weirdest (and coolest) part? I already feel like I’m part of a community. I don’t even fully know what that community is yet, but the vibe is definitely there:) That’s what happens when real people build things for other people. Huge thanks to the real humans at Olares for the brainpower and hard work you put into this! (P.S. If you are ever curious about how the setup and daily use actually feels from a total newbie's perspective, hit me up — I’d be happy to write down some 'dummy' feedback)

Two cool #AI #miniPC from @SapphireTech and @iMoreFine on @eBay nextdoorai.medium.com/morefi… @eBay_UK

30 Books For Non-Russians To Understand Modern #Russia: solodkiy.cv/navalny-books.ht… '#Navalny-like Books' vs '#Putin's People' - by @Navalny @MikhailFishman @Kira_Yarmysh, Roman Badanin (aka @Proekt_media), Mikhail Rubin, Ilya Zhegulev, Anton Dolin, Viktor Jerofejew, @amenka @BorisAkuninTw @CatherineBelton @shustry @zygaro @skazal_on @EvgenyFeldman @ArkadyOstrovsky @FD @TomBurgis @BillBrowder @McFaul @SGuriev @dstreisman @Sasha_Etkind @SlavaSolodkiy -> wish you to see&read more books from Shura Burtin, Roman @Dobrokhotov @ChristoGrozev @the_ins_ru @AnnaVeduta @VAshurkov too! Buy/read books here: @BAbook_Official Echo books @echofm_online @meduza_en book store @meduzaproject -> and support anti-Putin #opposition via your purchases (and donations)

Same “high-end” bracket. Two completely different souls 🔥gpu-new.lovable.app: @AorusOfficial @NVIDIAGeForce RTX 5090 INFINITY → raw, overkill, “I might melt your PSU” energy (32GB GDDR7 600W) by @GIGABYTEUSA @ASRockInfo Radeon RX 9070 XT Taichi White 16GB OC → clean, elegant, “I have an LCD screen and still sips power” vibes / & @AMD Radeon RX 7900 XTX One is the flex. One is the flex that doesn’t bankrupt you. Which one you riding with? notion.so/ASRock-Radeon-RX-9… #RTX5090 #RX9070XT #RX7900XTX @nvidia

#PocketNuke vs desktop apocalypse 🔥@TiinyAILab Pulse running 120B models in your jeans, fully offline, zero cloud cope → #TiinyAI tiiny-ai.lovable.app → You decide who’s the real local AI king. #LocalAI #PocketSupercomputer #CloudIsDead Which one you riding with? Tiiny AI vs Olares One vs Apple Mac Studio M4 vs Nvidia Jetson: notion.so/Tiiny-AI-vs-Olares…

:: #FOL-languages for the next #AI // solodkiy.my

Upvote pls 🚀 on @ProductHunt 🔥 l.solodkiy.cv/producthunt thnx, mates 🙏

A Secretive $11 Billion @DRAMgold Market Exists — And You're Not Invited: DRAM.gold <- The Biggest Opportunity Isn't in Hardware, But in Software That Makes It More Efficient Beneath the consumer market, a massive and sophisticated capital market for compute is now firmly established. This isn't an isolated trend; major deals confirm that billions in financing are being collateralized by GPUs. CoreWeave has secured $7.5 billion in debt financing, Lambda Labs has obtained $500 million, and UK-based Fluidstack has arranged a massive $10 billion GPU-backed facility. However, the secret to this market is counter-intuitive: lenders aren't primarily betting on the hardware's resale value. Instead, they are securing loans against the predictable cash flows generated by that hardware. CoreWeave’s GPU fleet, for example, generates an estimated $1.9 billion in annual revenue from cloud rental contracts with clients like Microsoft and OpenAI. The hardware depreciates, but the service contracts provide a steady, bankable revenue stream. This is precisely why consumer-grade hardware is excluded. Your gaming PC or stockpiled memory sticks don't generate cash flow, can't be easily tracked at scale, and lack a standardized liquidation channel, making them unattractive as collateral. While headlines focus on the eye-watering cost of hardware, the largest and most defensible market opportunity lies in software that optimizes its use. The AI inference market alone is projected to reach $106 billion in 2025, and a huge portion of that value will go to companies that help hardware run more efficiently. The high valuations and acquisitions in this space—from Multiverse Computing's $215M raise to NVIDIA's strategic purchase of OctoAI—signal where the "smart money" is flowing: not into the hardware itself, but into the intelligence that multiplies its efficiency. The impact of optimization software is staggering. For example: • Open-source frameworks like vLLM and DeepSpeed-FastGen have delivered up to 2.7x improvements in processing throughput by optimizing memory management. • Multiverse Computing, which raised $215 million, developed "CompactifAI," a technology that can reduce the size of Large Language Models (LLMs) by up to 95% with minimal precision loss. • The M&A landscape is heating up, with Red Hat acquiring Neural Magic and NVIDIA acquiring OctoAI, confirming the immense strategic value of optimization technology. This is a critical insight. Rather than speculating on the fluctuating price of a depreciating asset, optimization software directly attacks the problem of hardware scarcity by making existing resources more powerful. It's a scalable and defensible business model that creates lasting value across the entire ecosystem. You Can Turn Your Gaming PC into a Cash-Flowing Asset! While individuals can't access the enterprise lending market, a new opportunity has emerged for turning consumer hardware into a revenue-generating asset: decentralized GPU-as-a-Service (GPUaaS) networks. These peer-to-peer marketplaces allow anyone to earn money by renting out their idle compute power. Platforms like Cocoon, and Gonka AI connect hardware owners with users who need processing power for AI tasks. The business model is simple: you connect your GPU to the network, and the platform handles the marketplace, billing, and job distribution, paying you for the time your hardware is used. The economics can be compelling, particularly for inference workloads which run profitably on consumer GPUs. To illustrate the potential, a margin analysis from Lambda Labs shows that an enterprise-grade A10 GPU, costing around $3,500, can generate over $5,201 in annual revenue, achieving a payback in under eight months. While the revenue potential for consumer cards will differ, this model provides a direct path for individuals with high-end gaming PCs to monetize an existing asset that would otherwise sit idle. Yesterday's Tech is Literally a Gold Mine The rapid pace of AI development is creating a mountain of electronic waste—and a massive entrepreneurial opportunity in the circular economy. With fewer than one in five discarded GPUs being properly recycled, a fortune in valuable materials is waiting to be recovered. Consider these powerful statistics: • E-waste contains 40 to 800 times more gold per ton than ore extracted from traditional mines. • The global value of recoverable raw materials from e-waste is estimated at $57 billion. This has opened several entry points for new businesses with varying capital requirements. Key strategies include GPU/Memory Component Brokerage, Data Center Decommissioning, establishing a Regional Collection and Sorting Hub, and creating a Memory Testing and Certification Service. For vertically integrated processing operations that can handle precious metal extraction, margins can reach over 40%. In the rush to acquire the newest hardware, the value locked in yesterday's technology is one of the most overlooked opportunities. The AI infrastructure crisis is real, but treating hardware like a speculative commodity is a losing strategy. The rapid pace of technological obsolescence guarantees that today's prized chip is tomorrow's e-waste. Chasing short-term price spikes is a high-risk gamble. The real, sustainable opportunities lie elsewhere. They are found in providing services that make hardware useful (like GPUaaS), in developing software that makes it more efficient, and in building the circular systems that recover its value at the end of its life. These are the "shovels" of the great AI build-out. As this technological revolution continues, who will truly get rich: those speculating on the price of silicon, or those building the services and software that make it useful? @dram_gold