# Decentralized AI 165 predictions

URL for any entry below: https://www.gregosuri.com/predictions/{theme}/{id}/
Markdown twin: append `index.md`.
This is a title index. Full quotes and citation details are in individual twins.

## decentralized-ai — 165 statements, 2020–2026

Years: 2020 (1); 2021 (1); 2022 (5); 2023 (22); 2024 (68); 2025 (56); 2026 (12)

2020-12-10-defi-and-machine-learning-first-enterprise-later-t-frdf — DeFi and machine learning first, enterprise later (the S3 playbook)
2021-04-09-the-cloud-is-the-biggest-impediment-to-ai-soht — The cloud is the biggest impediment to AI
2022-10-20-private-deployments-enclaves-and-ml-on-akash-gpus-7icm — Private deployments, enclaves, and ML on Akash GPUs
2022-11-03-web3-and-generative-ai-will-converge-s2kn — Web3 and generative AI will converge
2022-11-11-ai-is-the-future-of-humanity-few-companies-can-t-c-rmz7 — AI is the future of humanity — few companies can't control its supply
2022-11-17-lots-of-value-exchanged-on-akash-from-ml-in-the-ve-3ngp — Lots of value exchanged on Akash from ML in the very short future
2022-12-01-ai-x-generative-art-whole-nft-series-in-hours-art-ldbh — AI x generative art: whole NFT series in hours, art commoditized by GPUs
2023-02-18-12-month-mission-akash-the-best-platform-for-ai-wo-wrgo — 12-month mission: Akash the best platform for AI workloads
2023-02-25-akash-ai-in-45-days-best-ai-developer-product-not-lkee — Akash AI in 45 days — best AI developer product, not a crypto product
2023-03-30-web3-is-the-supply-ai-is-the-demand-ed4b — Web3 is the supply, AI is the demand
2023-05-09-gpus-as-the-power-of-the-future-must-be-decentrali-4f2k — GPUs as the power of the future must be decentralized
2023-06-10-machine-learning-is-very-centralized-decentralizin-2bzk — Machine learning is very centralized — decentralizing it is extremely hard
2023-06-10-whoever-controls-gpus-controls-the-world-ai-must-b-3ljj — Whoever controls GPUs controls the world; AI must be in the hands of people
2023-06-15-akash-runs-15-000-models-render-runs-one-r32u — Akash runs 15,000 models; Render runs one
2023-06-15-an-explosion-in-machine-learning-on-akash-ja7q — An explosion in machine learning on Akash
2023-06-15-decentralized-gpu-networks-will-invisibly-power-ai-bu7g — Decentralized GPU networks will invisibly power AI
2023-08-31-geo-distributed-training-works-on-akash-lkdu — Geo-distributed training works on Akash
2023-09-06-resilience-kill-the-company-the-network-lives-y7l5 — Resilience: kill the company, the network lives
2023-09-25-decentralized-solutions-will-win-ai-not-centralize-adn4 — Decentralized solutions will win AI, not centralized ones
2023-09-25-the-commons-should-own-agi-adn4 — The commons should own AGI
2023-09-29-1-of-global-gdp-will-go-to-machine-learning-that-s-ssqz — \~1% of global GDP will go to machine learning; that's Akash's success metric
2023-09-29-akash-was-built-for-machine-learning-from-the-2017-v57i — Akash was built for machine learning from the 2017 white paper
2023-09-29-centralized-ai-leads-to-feudalism-make-sure-everyo-e7xz — Centralized AI leads to feudalism; make sure everyone has chips
2023-10-26-first-foundation-model-trained-on-a-decentralized-whzc — First foundation model trained on a decentralized cloud; centralized-superiority "going to be debunked"
2023-10-26-gpus-are-the-new-oil-sovereign-ai-3b7r — GPUs are the new oil; sovereign AI
2023-10-28-akash-is-the-only-functional-supercloud-6hue — Akash is the only functional supercloud
2023-10-28-gpu-gatekeepers-serving-shareholders-not-humanity-nebt — GPU gatekeepers serving shareholders, not humanity
2023-12-11-akash-will-be-the-platform-of-choice-for-decentral-t6ni — Akash will be the platform of choice for decentralized ML
2023-12-18-on-chain-ai-is-at-least-5-6-years-away-gmer — On-chain AI is at least 5-6 years away
2024-01-09-ai-access-is-controlled-by-a-few-players-not-a-fut-mqj2 — AI access is controlled by a few players — not a future to accept
2024-01-16-for-the-first-time-crypto-has-something-amazon-doe-bfpp — For the first time, crypto has something Amazon doesn't
2024-03-02-crypto-is-saving-ai-not-the-other-way-around-w7e3 — Crypto is saving AI, not the other way around
2024-03-19-crypto-networks-are-better-positioned-to-prevent-s-ndur — Crypto networks are better positioned to prevent Skynet than corporate boards
2024-03-19-full-stack-decentralization-not-one-layer-virq — Full-stack decentralization, not one layer
2024-03-19-general-purpose-compute-verification-is-unsolved-z-zz3h — General-purpose compute verification is unsolved; ZK is 40 minutes per image
2024-03-27-the-app-layer-captures-the-most-value-users-of-nvi-nxmt — The app layer captures the most value; users of Nvidia beat Nvidia
2024-03-30-bittensor-is-the-orchestration-layer-for-ai-xoer — Bittensor is the orchestration layer for AI
2024-05-20-an-unstoppable-machine-run-gpu-network-bcwz — An unstoppable machine-run GPU network
2024-05-20-first-foundation-model-trained-on-a-decentralized-fk56 — First foundation model trained on a decentralized network
2024-05-29-decentralized-training-is-starting-to-work-mqe7 — Decentralized training is starting to work
2024-05-29-nobody-has-solved-distributed-training-r2ih — Nobody has solved distributed training
2024-06-09-decentralized-training-research-is-promising-ltdc — Decentralized training research is promising
2024-06-09-nobody-has-solved-distributed-ml-training-ae6s — Nobody has solved distributed ML training
2024-06-13-decentralized-training-on-heterogeneous-compute-is-c3jp — Decentralized training on heterogeneous compute is promising
2024-06-14-ai-on-blockchain-is-a-reality-now-vgbt — AI on blockchain is a reality now
2024-06-18-bittensor-s-emissions-already-rival-openai-s-reven-6evb — Bittensor's emissions already rival OpenAI's revenue
2024-06-18-nobody-who-discovers-asi-will-share-it-so-asi-acce-6evb — Nobody who discovers ASI will share it — so ASI access must be a public utility
2024-06-19-ai-concentration-leads-to-a-feudalistic-society-jhzb — AI concentration leads to a feudalistic society
2024-06-19-no-one-has-solved-federated-learning-on-heterogene-mb6i — No one has solved federated learning on heterogeneous GPUs
2024-06-19-separation-of-compute-from-state-5pyp — Separation of compute from state
2024-07-17-crack-distributed-training-and-we-crack-agi-much-s-kxug — Crack distributed training and we crack AGI much sooner
2024-07-22-agi-should-be-decentralized-but-current-trajectory-4wtm — AGI should be decentralized — but current trajectory says otherwise
2024-07-22-akash-is-the-only-network-offering-on-demand-h100s-4hyy — Akash is the only network offering on-demand H100s at reasonable prices
2024-07-22-compute-should-be-a-common-good-shared-clusters-wi-34ka — Compute should be a common good; shared clusters will drive cost down
2024-07-22-federated-learning-on-distributed-clusters-is-live-wobc — Federated learning on distributed clusters is live via Prime Intellect and DiLoCo
2024-08-02-akash-is-the-de-facto-standard-for-decentralized-a-np2z — Akash is the de facto standard for decentralized AI
2024-08-02-the-notion-that-decentralized-systems-cannot-be-fa-sqao — "The notion that decentralized systems cannot be fast is just wrong" — and real products already prove it
2024-08-06-we-cannot-rely-on-centralized-ai-overlords-4j5v — We cannot rely on centralized AI overlords
2024-09-10-firmest-conviction-your-fridge-will-contribute-to-ctcy — Firmest conviction: your fridge will contribute to a global machine learning machine
2024-09-18-crypto-enables-data-sovereignty-for-ai-5a76 — Crypto enables data sovereignty for AI
2024-09-20-without-decentralization-we-become-peasants-under-uhby — Without decentralization, we become peasants under digital lords
2024-09-25-nvidia-integration-and-decentralized-training-cust-7hdz — Nvidia integration and decentralized training customers
2024-09-25-without-decentralization-we-become-peasants-under-w272 — Without decentralization, we become peasants under digital lords
2024-09-26-distributed-training-puzzle-being-solved-on-akash-mjn4 — Distributed training puzzle being solved on Akash
2024-09-26-private-ai-over-chatgpt-for-personal-data-abhi — Private AI over ChatGPT for personal data
2024-09-27-digital-feudalism-corporations-rationing-who-gets-nc77 — Digital feudalism: corporations rationing who gets to do AI
2024-09-27-nvidia-is-akash-s-biggest-user-and-its-only-crypto-ngtw — Nvidia is Akash's biggest user and its only crypto protocol
2024-09-30-crypto-enables-data-sovereignty-for-ai-gjff — Crypto enables data sovereignty for AI
2024-10-01-the-future-of-akash-verification-is-tee-a-reversal-u6xq — The future of Akash verification is TEE (a reversal he concedes)
2024-10-02-the-ai-crossroads-digital-lords-vs-the-people-tra3 — The AI crossroads: digital lords vs. the people
2024-10-15-decentralization-and-transparency-are-needed-to-un-myvv — Decentralization and transparency are needed to understand model bias
2024-10-15-most-data-lives-on-personal-devices-decentralizati-a22s — Most data lives on personal devices; decentralization unlocks it without breaking privacy
2024-10-15-permissionless-peer-to-peer-membership-removes-the-hm2n — Permissionless peer-to-peer membership removes the bound on compute scale
2024-10-16-five-unelected-people-controlled-the-world-s-most-4w5l — Five unelected people controlled the world's most powerful AI
2024-10-16-the-biggest-risk-society-faces-is-digital-feudalis-vlqa — The biggest risk society faces is digital feudalism
2024-10-16-whoever-gets-agi-first-will-keep-it-yjsw — Whoever gets AGI first will keep it
2024-10-18-agi-more-likely-from-many-small-models-than-one-bi-7cce — AGI more likely from many small models than one bigger model
2024-10-18-decentralized-networks-will-challenge-hyperscalers-ie6r — Decentralized networks will challenge hyperscalers
2024-10-24-by-cosmoverse-2025-a-100b-to-1t-parameter-decentra-clra — By Cosmoverse 2025: a 100B-to-1T parameter decentralized model, and reactors under construction
2024-10-24-concentration-of-power-plus-ai-leads-to-digital-fe-n65j — Concentration of power plus AI leads to digital feudalism
2024-10-24-data-centers-in-space-and-they-ll-be-decentralized-6tgv — Data centers in space — and they'll be decentralized
2024-10-24-decentralized-training-has-been-cracked-and-it-hap-4pkn — Decentralized training has been cracked, and it happened on Akash
2024-10-24-decentralized-training-works-and-it-s-happening-on-6lqv — Decentralized training works — and it's happening on Akash
2024-10-24-fractionalized-model-ownership-is-the-missing-ai-b-flw4 — Fractionalized model ownership is the missing AI business model
2024-11-09-agi-centralization-means-digital-feudalism-42c5 — AGI centralization means digital feudalism
2024-11-09-crypto-can-pool-capital-to-train-frontier-models-5-x6ai — Crypto can pool capital to train frontier models (\~$50M for a Llama)
2024-11-14-gpu-scarcity-even-for-the-richest-idle-gpus-are-th-jzab — GPU scarcity even for the richest; idle GPUs are the answer
2024-11-19-ai-will-be-the-most-important-layer-of-life-for-mo-j6zt — AI will be the most important layer of life for most people
2024-11-19-decouple-intelligence-from-the-state-like-bitcoin-u75w — Decouple intelligence from the state, like Bitcoin decoupled money
2024-11-19-host-a-model-pay-with-bitcoin-secure-on-bitcoin-br-arnn — Host a model, pay with Bitcoin, secure on Bitcoin — bringing intelligence to Bitcoin
2024-11-25-deai-summer-arrives-mid-2025-p6vt — "DeAI summer" arrives mid-2025
2024-11-25-distributed-training-is-a-few-generations-behind-a-qwta — Distributed training is a few generations behind and catching up fast
2024-11-26-decentralized-ai-will-shine-by-mid-2025-qukb — Decentralized AI will shine by mid-2025
2024-11-26-internet-scale-training-10b-now-then-100b-then-a-t-zqd2 — Internet-scale training: 10B now, then 100B, then a trillion parameters
2024-12-06-nous-research-s-distro-cracked-decentralized-train-2gag — Nous Research's DisTrO cracked decentralized training on Akash
2024-12-18-5-of-global-gdp-will-be-spent-on-ai-compute-by-203-j4ax — 5% of global GDP will be spent on AI compute by 2030
2024-12-18-own-ai-or-be-owned-by-it-xc4v — Own AI or be owned by it
2025-01-02-decentralized-ai-goes-mainstream-by-mid-2025-76yh — Decentralized AI goes mainstream by mid-2025
2025-01-02-nuclear-can-t-arrive-in-time-distributed-training-76yh — Nuclear can't arrive in time — distributed training is the answer
2025-01-31-agi-is-possible-on-chain-not-in-private-labs-h3np — AGI is possible on-chain, not in private labs
2025-01-31-deepseek-accelerates-deai-summer-expect-a-uniswap-gpe3 — DeepSeek accelerates "DeAI summer" — expect a Uniswap moment
2025-01-31-five-year-window-for-decentralization-to-go-mainst-yqt2 — Five-year window for decentralization to go mainstream
2025-01-31-synthetic-data-broke-the-data-ceiling-and-accelera-4fve — Synthetic data broke the data ceiling and accelerated AGI
2025-02-04-digital-feudalism-corporations-will-dictate-who-ge-5nsr — Digital feudalism: corporations will dictate who gets to do AI
2025-02-04-nvidia-is-akash-s-biggest-user-the-only-crypto-pro-xk67 — Nvidia is Akash's biggest user — the only crypto protocol it uses
2025-02-26-crypto-is-the-only-framework-to-monetize-open-ai-6wn2 — Crypto is the only framework to monetize open AI
2025-03-02-akash-ai-in-45-days-the-best-ai-developer-product-ti43 — Akash AI in 45 days — the best AI developer product, not a crypto product
2025-03-17-decentralization-is-how-you-disrupt-openai-pnus — Decentralization is how you disrupt OpenAI
2025-04-02-distributed-training-over-the-internet-just-became-w7ms — Distributed training over the internet just became real
2025-04-11-distributed-training-got-solved-three-months-ago-o-pcq4 — Distributed training got solved three months ago — on Akash
2025-04-11-in-a-couple-of-years-it-will-be-obvious-ai-can-onl-tst3 — In a couple of years it will be obvious: AI can only scale decentralized
2025-04-18-ai-access-will-create-a-feudal-society-if-left-con-gelb — AI access will create a feudal society if left concentrated
2025-04-18-decentralized-ai-is-close-to-catching-up-with-cent-rxdl — Decentralized AI is close to catching up with centralized AI
2025-04-18-decentralized-training-of-frontier-models-is-not-p-bglh — Decentralized training of frontier models is not possible today
2025-04-18-machine-learning-must-be-decentralized-and-owned-b-4kxc — Machine learning must be decentralized and owned by the public
2025-04-18-your-data-trains-gpt-for-free-data-sovereignty-sho-nir6 — Your data trains GPT for free — data sovereignty should be a primitive
2025-05-13-decentralized-ai-takes-off-mid-2025-every-layer-of-j2w6 — Decentralized AI "takes off" mid-2025; every layer of AI will decentralize
2025-05-13-distributed-training-just-broke-the-co-location-re-j2w6 — Distributed training just broke the co-location requirement
2025-06-03-distributed-training-is-in-its-gpt-3-era-limited-o-soyr — Distributed training is in its GPT-3 era, limited only by compute
2025-06-03-hyper-bullish-on-decentralized-ai-as-the-fix-for-t-4zy2 — "Hyper bullish" on decentralized AI as the fix for the energy crisis
2025-06-03-inference-will-be-much-larger-than-training-oaee — Inference will be much larger than training
2025-06-24-decentralize-ai-s-foundation-before-it-ossifies-womh — Decentralize AI's foundation before it ossifies
2025-06-24-decentralized-training-isn-t-the-future-it-s-the-p-moyj — Decentralized training "isn't the future, it's the present"
2025-06-24-open-compute-as-national-strategy-and-line-of-defe-l3uo — Open compute as national strategy and line of defense
2025-06-27-in-six-months-decentralized-training-went-from-vis-rqld — In six months decentralized training went from vision to proven path
2025-07-01-2025-broke-the-decentralized-ai-is-impossible-assu-y6sg — 2025 broke the "decentralized AI is impossible" assumption
2025-07-01-decentralization-is-a-national-security-advantage-lpot — Decentralization is a national-security advantage
2025-07-01-decentralize-ai-s-foundation-before-it-ossifies-flcp — Decentralize AI's foundation before it ossifies
2025-07-01-decentralized-training-already-works-at-405b-scale-bubc — Decentralized training already works at 405B scale
2025-07-01-inference-today-fine-tuning-tomorrow-training-soon-hu7v — Inference today, fine-tuning tomorrow, training soon after
2025-07-01-restoring-model-control-communities-keep-models-al-vx36 — Restoring model control — communities keep models alive
2025-07-16-distributed-infrastructure-decentralizes-opportuni-ikqp — Distributed infrastructure decentralizes opportunity
2025-07-31-distributed-training-hits-gpt-3-level-by-end-of-ye-p7mv — Distributed training hits GPT-3 level by end of year
2025-07-31-the-fix-is-changing-the-math-875x-less-communicati-p7mv — The fix is changing the math: 875x less communication
2025-07-31-traditional-ai-now-accepts-decentralized-training-p7mv — Traditional AI now accepts decentralized training: five papers at ICML
2025-08-02-decentralized-ai-is-the-only-way-no-new-energy-for-mgok — Decentralized AI is the only way — no new energy for four years
2025-08-02-distributed-training-becomes-reality-this-year-sta-53mo — Distributed training becomes reality this year; state-of-the-art by \~2027
2025-08-02-high-conviction-3-5-years-for-decentralized-ai-or-6qaf — High conviction: 3-5 years for decentralized AI, or never
2025-08-02-token-incentives-could-out-compete-openai-53mo — Token incentives could out-compete OpenAI
2025-08-02-we-re-about-to-blow-up-very-soon-4li4 — "We're about to blow up very soon"
2025-10-10-a-truly-decentralized-trained-model-within-6-12-mo-hcw6 — A truly decentralized-trained model within 6-12 months
2025-10-10-continuous-gpu-verification-is-unsolved-and-akash-qhy4 — Continuous GPU verification is unsolved — and Akash is close
2025-10-20-post-upgrade-roadmap-verifiable-compute-and-vms-wi-wdcy — Post-upgrade roadmap: verifiable compute and VMs, with demand to follow
2025-10-20-we-re-not-bitcoin-we-re-ai-the-ai-market-demands-r-iwjc — "We're not Bitcoin, we're AI" — the AI market demands rapid change
2025-11-04-a-15b-parameter-llm-was-trained-over-the-public-in-5kwu — A 15B-parameter LLM was trained over the public internet, matching centralized accuracy
2025-11-04-energy-gated-compute-means-centralization-by-defau-zmo6 — Energy-gated compute means centralization by default
2025-11-04-the-internet-itself-can-be-the-training-cluster-i2uz — The internet itself can be the training cluster
2025-11-04-we-re-expanding-who-gets-to-build-turning-idleness-jh5l — We're expanding who gets to build — turning idleness into intelligence
2025-12-03-decentralized-inference-is-complete-vaporware-toda-rzac — Decentralized inference is complete vaporware (today)
2025-12-03-most-decentralized-ai-is-junk-ignore-closed-source-axah — Most decentralized AI is junk; ignore closed source and off-chain projects
2025-12-03-tees-solve-verifiability-and-confidentiality-at-20-mtg6 — TEEs solve verifiability and confidentiality at \~20% overhead
2025-12-03-the-holy-grail-contribute-your-computer-to-a-train-4336 — The holy grail: contribute your computer to a training run, get tokens back
2025-12-03-there-is-no-other-future-that-s-not-decentralized-c4fa — There is no other future that's not decentralized
2026-02-12-crypto-ai-isn-t-dead-distributed-training-will-hit-zwzz — Crypto AI isn't dead — distributed training will "hit you hard"
2026-02-12-distributed-training-is-at-its-gpt-2-moment-km27 — Distributed training is at its GPT-2 moment
2026-02-24-distributed-training-will-be-a-thing-big-labs-moat-cqpu — Distributed training will be a thing; big labs' moat in doubt
2026-03-28-distributed-training-becomes-a-thing-in-6-months-t-xvxn — Distributed training becomes a thing in 6 months to a year
2026-04-11-distributed-training-takes-center-stage-53kl — Distributed training takes center stage
2026-04-11-every-lab-is-looking-at-it-mainstage-in-a-couple-o-zz6s — Every lab is looking at it — mainstage in a couple of years, GPUs on farmland
2026-04-23-a-70b-model-no-longer-needs-meta-s-data-centers-ri5t — A 70B model no longer needs Meta's data centers
2026-04-23-distributed-training-reaches-state-of-the-art-in-2-6gku — Distributed training reaches state of the art in 2-3 years
2026-04-23-if-we-rely-on-these-companies-for-intelligence-we-tglf — If we rely on these companies for intelligence, we lost as a society
2026-04-23-moratoriums-and-energy-limits-will-force-distribut-ot5q — Moratoriums and energy limits will force distributed training
2026-07-17-heterogeneous-compute-unlocks-decentralized-networ-acoo — Heterogeneous compute unlocks decentralized networks
2026-07-29-training-now-works-on-distributed-heterogeneous-ne-k4hj — Training now works on distributed, heterogeneous networks