# GPU Economics 248 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.

## gpu-economics — 248 statements, 2018–2026

Years: 2018 (4); 2020 (5); 2021 (40); 2022 (33); 2023 (71); 2024 (68); 2025 (19); 2026 (8)

2018-06-15-ten-million-servers-sit-comatose-while-76-struggle-eebq — Ten million servers sit comatose while 76% struggle with cloud costs
2018-11-09-85-of-server-capacity-sits-underutilized-wydh — 85% of server capacity sits underutilized
2018-11-10-85-of-server-capacity-sits-unused-h64d — 85% of server capacity sits unused
2018-11-10-akash-can-cut-cloud-prices-50-now-90-at-scale-m7ub — Akash can cut cloud prices 50% now, 90% at scale
2020-01-12-85-of-capacity-in-4-8-million-data-centers-is-unus-4ojj — 85% of capacity in 4.8 million data centers is unused
2020-01-12-compute-can-t-be-commoditized-it-s-like-diamonds-n-4ojj — Compute can't be commoditized — "it's like diamonds, not like gold"
2020-01-12-for-batch-workloads-cheap-enough-compute-makes-lat-4ojj — For batch workloads, cheap enough compute makes latency irrelevant
2020-06-22-machine-learning-is-akash-s-biggest-use-case-10x-c-6vqn — Machine learning is Akash's biggest use case, \~10x cost advantage
2020-12-16-85-of-data-center-capacity-sits-idle-across-8-2-mi-eace — 85%+ of data-center capacity sits idle across \~8.2 million data centers
2021-01-19-launch-pricing-2-3x-below-market-on-bare-metal-znnb — Launch pricing 2-3x below market on bare metal
2021-01-20-launch-pricing-2-3x-below-market-on-bare-metal-53rw — Launch pricing 2-3x below market on bare metal
2021-01-30-ai-companies-pay-20-of-margins-to-amazon-data-cent-regu — AI companies pay \~20% of margins to Amazon; data centers sit underutilized
2021-03-16-20-of-ai-company-margins-go-to-the-cloud-6yge — 20% of AI company margins go to the cloud
2021-03-16-20-of-ai-company-margins-go-to-the-cloud-cmko — 20% of AI company margins go to the cloud
2021-03-16-ai-companies-idle-gpu-clusters-become-a-peer-to-pe-6yge — AI companies' idle GPU clusters become a peer-to-peer cloud
2021-03-16-ai-companies-idle-gpu-clusters-become-a-peer-to-pe-cmko — AI companies' idle GPU clusters become a peer-to-peer cloud
2021-03-19-20-of-ai-company-margins-go-to-cloud-bills-uie6 — \~20% of AI-company margins go to cloud bills
2021-03-19-proof-of-stake-transitions-will-flood-the-market-w-fprl — Proof-of-stake transitions will flood the market with GPUs for Akash
2021-03-19-the-peer-to-peer-cloud-ml-clusters-earn-akt-in-dow-2exc — The peer-to-peer cloud: ML clusters earn AKT in downtime
2021-03-24-the-peer-to-peer-machine-learning-cloud-zau3 — The peer-to-peer machine learning cloud
2021-04-09-a-futures-market-and-hft-bots-for-compute-pytg — A futures market and HFT bots for compute
2021-04-09-a-power-grid-model-ai-companies-sell-idle-gpus-bac-ksdj — A power-grid model: AI companies sell idle GPUs back to the network
2021-04-09-ethereum-s-move-to-proof-of-stake-will-flood-the-m-ipsc — Ethereum's move to proof of stake will flood the market with GPUs for AI
2021-05-12-85-of-data-center-compute-sits-unused-bsm3 — 85% of data-center compute sits unused
2021-05-12-akash-will-launch-the-world-s-first-gpu-marketplac-bsm3 — Akash will launch "the world's first GPU marketplace" fed by post-Merge Ethereum GPUs
2021-05-12-post-merge-mining-hardware-becomes-useless-without-bsm3 — Post-Merge mining hardware becomes useless without a compute marketplace
2021-05-27-ethereum-s-pos-switch-strands-gpu-infrastructure-a-xvgp — Ethereum's PoS switch strands GPU infrastructure — a huge AI opportunity
2021-05-27-we-ll-buy-stranded-gpus-at-half-price-for-the-ai-m-aw43 — We'll buy stranded GPUs at half price for the AI market
2021-05-28-ethereum-s-merge-will-strand-miner-gpus-akash-will-3njm — Ethereum's merge will strand miner GPUs — Akash will absorb them
2021-05-28-the-invisible-amazon-tax-on-machine-learning-3njm — The invisible Amazon tax on machine learning
2021-05-29-85-of-capacity-in-8-4-million-data-centers-is-idle-x6ih — 85% of capacity in 8.4 million data centers is idle; 8-10x cheaper than Amazon
2021-05-29-world-s-first-open-gpu-marketplace-coming-customer-tcwd — World's first open GPU marketplace coming — customers spend $1-2M/month on GPUs
2021-06-09-20-of-ai-companies-margins-go-to-amazon-xun6 — 20% of AI companies' margins go to Amazon
2021-06-09-ethereum-s-proof-of-stake-shift-will-unlock-enormo-bhqf — Ethereum's proof-of-stake shift will unlock enormous GPU supply
2021-06-22-ethereum-s-merge-will-strand-miner-gpus-akash-will-ftgz — Ethereum's merge will strand miner GPUs — Akash will absorb them
2021-06-22-the-invisible-amazon-tax-on-machine-learning-7cwg — The invisible Amazon tax on machine learning
2021-06-27-akash-costs-one-third-to-one-fourth-of-centralized-nyza — Akash costs one-third to one-fourth of centralized cloud
2021-07-14-85-90-of-data-centers-are-idle-akash-costs-one-thi-2fvx — 85-90% of data centers are idle; Akash costs one-third of Amazon
2021-07-14-gpu-marketplace-in-6-months-will-be-the-holy-grail-digi — GPU marketplace in \~6 months will be "the holy grail"
2021-07-14-gpus-are-ai-s-biggest-cost-20-of-ai-companies-marg-dgv3 — GPUs are AI's biggest cost — \~20% of AI companies' margins go to Amazon
2021-07-15-gpus-on-akash-in-the-next-few-months-n34l — GPUs on Akash "in the next few months"
2021-09-11-85-of-data-center-capacity-sits-unused-zvnr — 85% of data center capacity sits unused
2021-10-08-gpu-support-on-akash-coming-in-about-six-months-z27c — GPU support on Akash coming in about six months
2021-10-12-half-of-saas-margins-go-to-amazon-bkdm — Half of SaaS margins go to Amazon
2021-10-12-half-to-one-fifth-the-cost-of-amazon-pkwh — Half to one-fifth the cost of Amazon
2021-10-12-ml-on-akash-in-six-months-fueled-by-the-ethereum-m-qs2s — ML on Akash in six months, fueled by the Ethereum merge's stranded GPUs
2021-12-15-akash-is-one-third-to-one-fifth-the-price-of-aws-d-mxcm — Akash is one-third to one-fifth the price of AWS, driving miner arbitrage
2021-12-15-akash-plans-a-futures-market-for-compute-hs5m — Akash plans a futures market for compute
2021-12-22-the-ethereum-merge-will-flood-the-market-with-gpus-lcf4 — The Ethereum merge will flood the market with GPUs for Akash
2022-01-10-gpus-on-akash-in-three-to-five-months-bklv — GPUs on Akash in three to five months
2022-01-10-post-merge-ethereum-miners-will-repurpose-gpus-ont-sjcc — Post-merge Ethereum miners will repurpose GPUs onto Akash
2022-01-13-akash-compute-at-one-fifth-to-one-third-of-market-7qxh — Akash compute at one-fifth to one-third of market price
2022-06-16-general-purpose-gpu-cloud-spec-ready-20-customers-q5ts — General-purpose GPU cloud: spec-ready, 20 customers, no one has done it
2022-06-29-akash-gpus-20-50-cheaper-than-market-jigw — Akash GPUs 20-50% cheaper than market
2022-06-29-merge-will-strand-enormous-gpu-capacity-oneu — Merge will strand enormous GPU capacity
2022-07-07-ethereum-merge-gpus-will-be-akash-s-unique-value-p-5q74 — Ethereum merge GPUs will be Akash's unique value proposition
2022-07-14-gpu-machine-learning-workloads-are-the-next-market-wa2u — GPU/machine-learning workloads are the next market after PMF
2022-07-25-post-merge-gpus-as-akash-s-unique-value-propositio-6s3w — Post-merge GPUs as Akash's unique value proposition
2022-07-28-eth-merge-will-flood-akash-with-cheap-gpus-txm7 — ETH merge will flood Akash with cheap GPUs
2022-08-18-gpu-hpc-on-akash-will-undercut-cloud-by-20-30-thpd — GPU/HPC on Akash will undercut cloud by 20-30%
2022-08-18-gpus-managed-services-will-be-a-paradigm-shift-for-jrko — GPUs + managed services will be a paradigm shift for Akash
2022-08-18-over-85-of-the-world-s-compute-sits-unused-ko5v — Over 85% of the world's compute sits unused
2022-09-15-gpu-on-kubernetes-prototype-worked-timeline-hostag-shgf — GPU-on-Kubernetes prototype worked; timeline hostage to upstream
2022-10-04-akash-costs-about-95-cheaper-than-your-traditional-77g5 — "Akash costs about 95% cheaper than your traditional Amazon"
2022-10-04-gpus-coming-to-akash-within-three-months-kmlj — GPUs coming to Akash within three months
2022-10-13-globally-distributed-gpu-clusters-mean-massive-cos-glav — Globally distributed GPU clusters mean massive cost reduction from reused hardware
2022-10-13-gpu-marketplace-prototype-by-end-of-2022-lfoi — GPU marketplace prototype by end of 2022
2022-10-20-gpus-are-the-next-mainnet-milestone-oknr — GPUs are the next mainnet milestone
2022-10-28-gpus-are-the-superiority-play-an-open-gpu-market-t-vitj — GPUs are the superiority play: an open GPU market the cloud doesn't have
2022-10-28-gpus-will-be-akash-s-netflix-moment-ppuz — GPUs will be Akash's "Netflix moment"
2022-10-28-society-doesn-t-have-enough-gpus-for-even-immediat-fkfr — Society doesn't have enough GPUs for even immediate AI adoption
2022-11-03-19b-of-idle-ethereum-mining-gpus-to-unlock-2g5g — $19B of idle Ethereum-mining GPUs to unlock
2022-11-03-gpu-demand-is-doubling-every-four-months-nqvr — GPU demand is doubling every four months
2022-11-03-training-stable-diffusion-600k-cheaper-on-akash-m6bh — Training Stable Diffusion: \~$600k, cheaper on Akash
2022-11-11-19b-of-post-merge-gpus-need-a-market-and-akash-is-v4lo — $19B of post-merge GPUs need a market — and Akash is ready
2022-11-11-generative-ai-compute-demand-doubling-every-four-m-fueo — Generative AI compute demand doubling every four months
2022-11-13-gpus-at-5x-price-akash-uniquely-taps-unused-genera-nmpg — GPUs at 5x price; Akash uniquely taps unused general-purpose GPU capacity
2022-11-17-19b-of-gpus-unlocked-from-ethereum-mining-akash-po-uwj6 — $19B of GPUs unlocked from Ethereum mining, Akash positioned to capture them
2022-11-17-ai-will-be-akash-s-biggest-demand-driver-gpu-need-fb64 — AI will be Akash's biggest demand driver; GPU need doubling every four months
2022-11-21-gpus-move-akash-from-parity-to-superiority-via-the-jqg3 — GPUs move Akash from parity to superiority via the first open GPU market
2022-11-21-gpus-will-be-akash-s-netflix-moment-fa72 — GPUs will be Akash's Netflix moment
2022-12-11-cloud-repatriation-40x-savings-building-in-house-g-p3ub — Cloud repatriation: 40x savings building in-house GPUs
2023-01-13-19b-of-ethereum-mining-gpus-sit-idle-post-merge-ak-mjs7 — \~$19B of Ethereum-mining GPUs sit idle post-merge — Akash will target them
2023-01-13-gpus-on-akash-within-three-months-ai-avatar-apps-a-jols — GPUs on Akash within three months; AI avatar apps as first killer demand
2023-01-13-the-cloud-is-breaking-down-under-ai-gpu-scarcity-indd — The cloud is breaking down under AI GPU scarcity
2023-01-20-gpus-coming-to-akash-within-six-months-maybe-even-5tp7 — GPUs coming to Akash within six months — maybe even to secure Bitcoin
2023-02-01-gpus-are-next-on-akash-yykm — GPUs are next on Akash
2023-02-16-a100s-five-week-waits-and-concentrated-ownership-wftk — A100s: five-week waits and concentrated ownership
2023-02-16-ai-demand-grows-10x-every-six-months-decentralize-oh3q — AI demand grows 10x every six months; decentralize or few companies control AI
2023-02-18-ai-complexity-is-doubling-every-six-months-gvvt — AI complexity is doubling every six months
2023-02-18-akash-gpu-market-to-launch-with-at-least-20-000-gp-azwa — Akash GPU market to launch with at least 20,000 GPUs
2023-02-18-chip-costs-to-rise-five-times-from-protectionism-t-6m5f — Chip costs to rise five times from protectionism/TSMC relocation
2023-02-18-gpu-prices-will-get-worse-not-better-izhp — GPU prices will get worse, not better
2023-02-25-900-000-ethereum-merge-gpus-to-come-online-during-36e6 — 900,000 Ethereum-merge GPUs to come online during 2023
2023-02-25-a100s-at-0-50-hour-vs-4-on-traditional-cloud-eumg — A100s at $0.50/hour vs. $4 on traditional cloud
2023-02-25-gpu-demand-doubling-every-3-5-months-gpt-4-at-a-tr-i2lg — GPU demand doubling every 3.5 months; GPT-4 at a trillion parameters
2023-03-30-ai-workloads-will-dominate-akash-on-cost-1-hr-a100-bccd — AI workloads will dominate Akash on cost: $1/hr A100s vs $4 on AWS
2023-05-09-akash-will-be-the-only-place-ml-devs-can-get-h100s-fxu7 — Akash will be the only place ML devs can get H100s
2023-05-09-gpus-are-the-new-oil-the-new-gold-f5ou — GPUs are the new oil / the new gold
2023-05-12-akash-as-the-only-cloud-with-h100s-no-longer-just-fiak — Akash as the only cloud with H100s — no longer just the cheap alternative
2023-05-12-clouds-selectively-gatekeep-high-end-gpus-openai-g-v2lz — Clouds selectively gatekeep high-end GPUs; OpenAI got a leg up
2023-05-12-ml-companies-should-own-their-own-gpus-for-soverei-2uxj — ML companies should own their own GPUs for sovereignty
2023-05-12-nvidia-is-a-software-company-not-a-hardware-compan-rdbi — Nvidia is a software company, not a hardware company
2023-05-17-gpu-supply-won-t-catch-demand-for-years-n5h4 — GPU supply won't catch demand for years
2023-05-17-gpus-are-the-new-oil-d5ak — GPUs are the new oil
2023-06-03-gpus-are-the-blood-for-ai-whoever-controls-gpus-co-w6pn — GPUs are the blood for AI; whoever controls GPUs controls AI
2023-06-03-the-ethereum-merge-unleashed-19b-of-idle-gpus-as-c-7jsc — The Ethereum merge unleashed $19B of idle GPUs as centralized infrastructure crumbles
2023-06-10-ai-wave-means-exponential-compute-demand-z5xh — AI wave means exponential compute demand
2023-06-10-akash-launches-gpus-from-a-position-of-strength-b6dj — Akash launches GPUs from a position of strength
2023-06-10-chip-market-to-grow-10x-to-400b-in-nine-years-cfop — Chip market to grow \~10x to $400B in nine years
2023-06-10-chips-are-the-new-oil-stwy — Chips are the new oil
2023-06-10-gpu-costs-and-demand-both-rising-for-years-due-to-dh7e — GPU costs and demand both rising for years due to geopolitics
2023-06-10-gpu-scarcity-makes-a-secondary-market-inevitable-ynds — GPU scarcity makes a secondary market inevitable
2023-06-12-ai-compute-market-will-be-worth-trillions-xkaj — AI compute market will be worth trillions
2023-06-12-airbnb-for-data-centers-t52c — Airbnb for data centers
2023-06-12-from-cheap-cloud-to-access-cloud-ngoh — From cheap cloud to access cloud
2023-06-15-1-5t-ai-spend-vs-a-200m-market-cap-3wxi — $1.5T AI spend vs a $200M market cap
2023-06-15-akash-as-the-world-s-first-open-gpu-secondary-mark-wd7a — Akash as the world's first open GPU secondary market
2023-06-15-gpus-are-the-oil-for-ai-supply-chain-broken-for-2-lbkj — GPUs are the oil for AI; supply chain broken for 2-5 years
2023-08-31-akash-is-the-most-capable-gpu-cloud-p3ap — Akash is the most capable GPU cloud
2023-08-31-amd-is-at-least-two-years-behind-nvidia-ygki — AMD is at least two years behind Nvidia
2023-08-31-best-amd-chips-will-land-on-akash-before-any-cloud-zuoi — Best AMD chips will land on Akash before any cloud
2023-08-31-nvidia-is-deliberately-fragmenting-the-gpu-market-64jo — Nvidia is deliberately fragmenting the GPU market
2023-08-31-openai-can-t-buy-gpus-proof-of-demand-wnvv — OpenAI can't buy GPUs — proof of demand
2023-09-06-1-5-trillion-ai-spend-in-the-next-five-years-y7l5 — $1.5 trillion AI spend in the next five years
2023-09-06-gpus-are-the-oil-for-ai-shortage-lasts-2-5-years-y7l5 — GPUs are the oil for AI; shortage lasts 2–5 years
2023-09-25-akash-is-the-airbnb-for-high-end-gpus-adn4 — Akash is the Airbnb for high-end GPUs
2023-09-25-chips-are-the-new-oil-adn4 — Chips are the new oil
2023-09-25-nvidia-is-deliberately-fragmenting-the-gpu-market-adn4 — NVIDIA is deliberately fragmenting the GPU market
2023-09-28-1-of-global-gdp-will-go-to-machine-learning-qs2i — 1% of global GDP will go to machine learning
2023-09-28-globally-distributed-consumer-gpu-clusters-viable-vocw — Globally distributed consumer-GPU clusters viable within two years
2023-09-29-akash-is-the-secondary-market-for-the-gpu-supply-c-ezxl — Akash is the secondary market for the GPU supply crunch
2023-09-29-all-a100s-sold-out-within-three-weeks-of-gpu-launc-hkfm — All A100s sold out within three weeks of GPU launch
2023-09-29-h100s-take-two-years-to-get-harder-than-a-rolex-x2dv — H100s take two years to get — "harder than a Rolex"
2023-09-29-nvidia-deprioritized-the-hyperscalers-as-competito-c4xj — Nvidia deprioritized the hyperscalers as competitors
2023-10-26-80-of-akash-demand-will-be-inference-wgw3 — 80% of Akash demand will be inference
2023-10-26-ai-products-will-be-forced-to-distributed-clusters-ihm3 — AI products will be forced to distributed clusters by unit economics
2023-10-26-chatgpt-needs-one-h100-per-concurrent-user-nitf — ChatGPT needs one H100 per concurrent user
2023-10-26-crypto-s-inverted-problem-high-demand-not-enough-g-l4xp — Crypto's inverted problem: high demand, not enough GPU supply
2023-10-28-akash-is-the-airbnb-of-gpu-contracts-mnyf — Akash is the Airbnb of GPU contracts
2023-10-28-heterogeneous-compute-can-never-be-fully-commoditi-gije — Heterogeneous compute can never be fully commoditized
2023-10-28-nvidia-is-deliberately-depowering-the-big-clouds-bwko — Nvidia is deliberately depowering the big clouds
2023-10-28-what-happens-with-access-to-a-million-gpus-fp7f — What happens with access to a million GPUs
2023-12-11-frontier-launches-on-nvidia-mass-adoption-spreads-t6ni — Frontier launches on NVIDIA; mass adoption spreads to cheaper GPUs
2023-12-11-gpu-supply-must-be-diversified-taiwan-is-a-single-t6ni — GPU supply must be diversified — Taiwan is a single point of failure
2023-12-18-95-utilization-on-a100s-no-incentives-needed-dn3c — 95% utilization on A100s, no incentives needed
2023-12-18-gpus-are-36-of-all-time-revenue-after-90-days-hxg7 — GPUs are 36% of all-time revenue after 90 days
2023-12-18-h100s-12-hr-on-amazon-vs-2-hr-on-akash-5jbw — H100s: \~$12/hr on Amazon vs \~$2/hr on Akash
2023-12-18-incentives-will-make-akash-the-cheapest-place-for-idj2 — Incentives will make Akash the cheapest place for A100s
2023-12-19-50-utilization-is-the-marketplace-sweet-spot-p3na — 50% utilization is the marketplace sweet spot
2023-12-19-akash-is-4-5x-render-s-usage-kfgu — Akash is \~4-5x Render's usage
2023-12-19-chip-supply-is-nowhere-close-to-demand-get-chips-i-2pfy — Chip supply is nowhere close to demand — get chips into people's hands
2023-12-19-milestone-1-million-a-day-in-network-spend-wqt4 — Milestone: $1 million a day in network spend
2024-01-09-a100-providers-can-recover-hardware-cost-within-a-4yvd — A100 providers can recover hardware cost within a year
2024-01-09-akash-growing-150-month-over-month-6w2v — Akash growing 150% month-over-month
2024-01-09-without-100m-or-hyperscaler-backing-high-density-g-xl26 — Without $100M or hyperscaler backing, high-density GPUs are unobtainable
2024-01-16-liquidity-mining-for-gpus-bfpp — Liquidity mining for GPUs
2024-02-14-5-of-global-gdp-will-be-spent-on-ai-hardware-klnh — \~5% of global GDP will be spent on AI hardware
2024-03-02-600-gpus-month-coming-online-h100s-at-2-hour-u2y4 — 600 GPUs/month coming online; H100s at $2/hour
2024-03-02-big-clouds-control-gpu-leasing-startups-are-locked-yaoo — Big clouds control GPU leasing; startups are locked out
2024-03-02-there-are-free-gpus-everywhere-gpus-are-the-new-sp-ce3h — There are free GPUs everywhere; GPUs are the new spice
2024-03-19-gpus-harder-to-get-than-drugs-sub-2-h100s-on-akash-zhzm — GPUs "harder to get than drugs"; sub-$2 H100s on Akash
2024-03-27-gpu-supply-crunch-lasts-until-2026-biggest-war-che-kssc — GPU supply crunch lasts until \~2026; biggest war chest wins
2024-03-27-gpus-are-the-most-important-chipsets-for-humanity-7cen — GPUs are the most important chipsets for humanity
2024-03-27-gpus-are-the-new-spice-wpo3 — GPUs are the new spice
2024-03-30-gpus-are-the-new-spice-oyzi — GPUs are the new spice
2024-05-20-nvidia-is-deprioritizing-hyperscalers-by-design-iho5 — Nvidia is deprioritizing hyperscalers by design
2024-05-29-akash-as-the-first-place-to-get-frontier-chips-ojbq — Akash as the first place to get frontier chips
2024-05-29-claimed-2018-call-that-ml-would-drive-gpu-demand-iv64 — Claimed 2018 call that ML would drive GPU demand
2024-05-29-more-available-gpus-than-amazon-m7k7 — More available GPUs than Amazon
2024-06-09-2018-whitepaper-called-the-ml-compute-crunch-2eia — 2018 whitepaper called the ML compute crunch
2024-06-09-scaling-supply-is-easier-than-scaling-demand-eqqe — Scaling supply is easier than scaling demand
2024-06-09-sub-2-on-demand-h100s-are-unheard-of-mxnc — Sub-$2 on-demand H100s are unheard of
2024-06-09-the-spice-must-flow-gpus-sell-themselves-without-d-jim6 — "The spice must flow" — GPUs sell themselves without demand incentives
2024-06-11-h100s-at-1-46-hour-vs-12-on-amazon-if-you-can-get-wvvg — H100s at $1.46/hour vs. $12 on Amazon — if you can get one at all
2024-06-11-the-2018-whitepaper-called-the-ml-gpu-boom-seven-y-vdrv — The 2018 whitepaper called the ML/GPU boom seven years early
2024-06-13-2-hr-h100s-vs-amazon-s-12-r2mz — $2/hr H100s vs Amazon's $12
2024-06-13-5-of-global-gdp-will-be-spent-on-ai-compute-by-203-kiod — 5% of global GDP will be spent on AI compute by 2030
2024-06-13-ai-can-t-embed-in-everyday-products-not-enough-chi-pcqz — AI can't embed in everyday products — not enough chips
2024-06-13-gpu-supply-chain-eases-in-three-to-five-years-5rhv — GPU supply chain eases in three to five years
2024-06-13-no-challenger-threatens-nvidia-for-five-years-gyt7 — No challenger threatens Nvidia for five years
2024-06-13-revenue-to-triple-or-quadruple-within-two-months-vukk — Revenue to triple or quadruple within two months
2024-06-14-gpu-scarcity-is-a-coordination-problem-blockchains-yctq — GPU scarcity is a coordination problem blockchains solve
2024-06-18-akash-is-the-only-place-to-get-an-h100-on-demand-6evb — Akash is the only place to get an H100 on demand
2024-06-18-the-2017-whitepaper-predicted-ai-demand-would-outr-6evb — The 2017 whitepaper predicted AI demand would outrun GPU supply
2024-06-19-ai-compute-to-5-of-global-gdp-by-2030-supply-chain-fu4u — AI compute to 5% of global GDP by 2030; supply chain eases in 3-5 years
2024-07-22-breaking-nvidia-s-monopoly-via-amd-first-tinybox-w-tdym — Breaking Nvidia's monopoly via AMD — first tinybox will run on Akash
2024-07-22-cutting-the-cloud-middleman-removes-amazon-s-43-ma-byu7 — Cutting the cloud middleman removes Amazon's \~43% margin
2024-07-22-gpus-are-nearly-impossible-to-lease-on-demand-nzib — GPUs are nearly impossible to lease on demand
2024-07-29-supply-is-solved-by-incentives-demand-side-onboard-r3jp — Supply is solved by incentives; demand-side onboarding is the real bottleneck
2024-07-29-trial-wallets-will-unlock-a-significant-demand-inc-y3vb — Trial wallets will unlock a significant demand increase within months
2024-08-02-akash-predicted-the-gpu-shortage-in-its-2018-white-rrno — Akash predicted the GPU shortage in its 2018 whitepaper
2024-08-02-enormous-idle-gpu-supply-sits-in-crypto-miners-and-a32p — Enormous idle GPU supply sits in crypto miners and warehouses
2024-08-02-gpus-are-nearly-impossible-to-lease-on-demand-gc32 — GPUs are nearly impossible to lease on demand
2024-09-10-cloud-providers-dictate-who-gets-to-train-ai-ctcy — Cloud providers dictate who gets to train AI
2024-09-10-data-centers-sit-unused-85-of-the-time-ctcy — Data centers sit unused 85% of the time
2024-09-10-on-demand-h100s-at-1-50-hour-unheard-of-in-traditi-4566 — On-demand H100s at $1.50/hour, unheard of in traditional cloud
2024-09-20-akash-is-80-cheaper-than-amazon-h100s-at-1-49-vs-5-6ceq — Akash is \~80% cheaper than Amazon; H100s at $1.49 vs $5
2024-09-25-akash-is-80-cheaper-than-amazon-h100s-at-1-49-vs-5-u7mq — Akash is \~80% cheaper than Amazon; H100s at $1.49 vs $5
2024-09-26-akash-is-airbnb-economics-vs-cloud-s-hilton-econom-zcbo — Akash is Airbnb economics vs. cloud's Hilton economics
2024-09-26-cheapest-h100s-in-the-industry-7cod — Cheapest H100s in the industry
2024-09-27-aws-s-60-margins-are-incompatible-with-a-public-ut-jdip — AWS's 60%+ margins are incompatible with a public utility
2024-09-27-nvidia-demand-is-already-declining-as-open-models-b3jd — Nvidia demand is already declining as open models run on non-Nvidia chips
2024-09-27-nvidia-won-t-stay-dominant-i-don-t-want-to-live-in-rdvc — Nvidia won't stay dominant — "I don't want to live in a digital feudalism"
2024-10-01-gpu-demand-will-keep-outpacing-supply-akash-is-the-v2wg — GPU demand will keep outpacing supply; Akash is the only on-demand source
2024-10-01-utilization-gated-supply-incentives-50-used-releas-bvk6 — Utilization-gated supply incentives: 50% used releases rewards to double capacity
2024-10-02-nvidia-is-backlogged-18-months-to-2-years-on-h100s-tra3 — NVIDIA is backlogged 18 months to 2 years on H100s
2024-10-02-serving-apple-s-users-would-take-a-billion-h100s-o-tra3 — Serving Apple's users would take a billion H100s — only \~700,000 exist
2024-10-02-the-endgame-is-one-h100-class-gpu-per-user-refresh-tra3 — The endgame is one H100-class GPU per user, refreshed yearly
2024-10-16-akash-daily-spend-grew-10x-in-a-year-wqid — Akash daily spend grew 10x in a year
2024-11-09-akash-is-ebay-for-gpus-iycn — Akash is eBay for GPUs
2024-11-09-nvidia-is-akash-s-biggest-user-jhd3 — Nvidia is Akash's biggest user
2024-11-09-sub-2-hour-h100s-mean-the-market-is-in-oversupply-bkjt — Sub-$2/hour H100s mean the market is in oversupply
2024-11-14-data-centers-are-85-90-underutilized-bjtc — Data centers are 85-90% underutilized
2024-11-26-1700-growth-1-cloud-market-share-in-about-five-yea-icin — 1700% growth; 1% cloud market share in about five years
2024-11-26-75-90-of-data-center-capacity-is-underutilized-201-oojj — 75-90% of data center capacity is underutilized (2017 supercloud paper)
2024-11-26-92-gpu-utilization-highest-of-any-cloud-pfgl — 92% GPU utilization — highest of any cloud
2024-11-26-compute-futures-markets-within-a-year-wnyu — Compute futures markets within a year
2024-11-26-nvidia-s-dominance-won-t-last-long-v33a — Nvidia's dominance won't last long
2024-12-06-akash-delivers-the-same-compute-as-aws-for-80-chea-3xhv — Akash delivers the same compute as AWS for 80% cheaper by reaching idle capacity
2024-12-06-the-cloud-tax-half-of-every-dollar-for-online-serv-r733 — The "cloud tax": half of every dollar for online services goes to four providers
2025-01-02-contribute-compute-own-the-model-earn-inference-re-76yh — Contribute compute, own the model, earn inference revenue
2025-01-19-5-of-gdp-spent-on-compute-by-2030-gartner-zy6g — \~5% of GDP spent on compute by 2030 (Gartner)
2025-02-04-a-60-margin-layer-cannot-be-a-public-utility-every-agyu — A 60%-margin layer cannot be a public utility — everyone pays a cloud tax
2025-02-26-akash-utilization-grew-10-to-70-in-12-months-kb4c — Akash utilization grew 10% to 70% in 12 months
2025-02-26-h200s-are-100-utilized-on-akash-deepseek-demand-ynj2 — H200s are 100% utilized on Akash (DeepSeek demand)
2025-03-02-900-000-ethereum-merge-gpus-will-come-online-on-ak-oti5 — 900,000 Ethereum-merge GPUs will come online on Akash in 2023
2025-03-02-a100s-at-0-50-hour-vs-4-on-traditional-cloud-b5tr — A100s at $0.50/hour vs. $4 on traditional cloud
2025-03-02-gpu-demand-doubling-every-3-5-months-gpt-4-at-a-tr-bnjr — GPU demand doubling every 3.5 months; GPT-4 at a trillion parameters
2025-03-17-enterprise-data-centers-sit-at-15-utilization-zwe7 — Enterprise data centers sit at \~15% utilization
2025-03-17-targets-10k-gpus-in-2025-100k-in-2026-1m-in-three-2igw — Targets: 10K GPUs in 2025, 100K in 2026, 1M in three years
2025-04-18-akash-s-community-has-access-to-1-million-gpus-mn77 — Akash's community has access to \~1 million GPUs
2025-04-18-gpus-are-the-oil-whoever-controls-the-fuel-control-e25n — GPUs are the oil — whoever controls the fuel controls the world
2025-04-18-openai-s-moat-is-gpus-and-data-not-the-model-and-i-3bmy — OpenAI's moat is GPUs and data, not the model — and it's eroding
2025-07-01-a-gigawatt-of-ai-compute-doesn-t-need-one-building-nbdk — A gigawatt of AI compute doesn't need one building
2025-07-31-gpu-access-is-feudalism-p7mv — GPU access is feudalism
2025-10-10-70-utilization-vs-aws-s-40-and-filecoin-s-0-5-mxe2 — 70% utilization vs. AWS's 40% and Filecoin's 0.5%
2025-10-10-80-cheaper-than-cloud-deployed-in-two-minutes-4m3e — 80% cheaper than cloud, deployed in two minutes
2025-11-04-akash-sustains-54-55-utilization-across-1-000-gpus-a5s7 — Akash sustains 54-55% utilization across 1,000+ GPUs
2025-12-03-priced-out-builders-prove-decentralization-works-t-y5rf — Priced-out builders prove decentralization works — the Cornell-to-Nvidia story
2026-01-15-in-the-2023-gpu-crisis-akash-was-the-only-one-with-draf — In the 2023 GPU crisis, Akash was the only one with GPUs
2026-02-12-one-gpu-per-prompt-7-billion-ai-users-implies-stag-clyb — One GPU per prompt — 7 billion AI users implies staggering GPU demand
2026-02-12-the-h100-gap-half-a-million-made-four-million-need-tg7y — The H100 gap: half a million made, four million needed
2026-02-24-physical-goods-going-up-resource-trade-still-early-wkxj — Physical goods going up; resource trade still early
2026-04-23-claimed-track-record-called-the-gpu-and-energy-sho-joas — Claimed track record: called the GPU and energy shortages
2026-04-23-market-shift-from-single-gpus-to-50-100-node-h100-nmuc — Market shift from single GPUs to 50-100 node H100 clusters on long-term contracts
2026-07-17-gpus-appreciate-instead-of-depreciating-6xlr — GPUs appreciate instead of depreciating
2026-07-29-put-5090s-at-home-on-solar-and-earn-income-at-zero-k4hj — Put 5090s at home on solar and earn income at zero marginal cost