loading the coin.
THE REAL-WORLD GOAL / art
We trained an AI on every memecoin we could find.
Explore the projectCookr is an open-source meme image model trained on memecoin culture. Two light versions are public on Hugging Face under CookrAI (cookr-v1-light on Z-Image, cookr-v1-light-qwen on Qwen-Image), and the three data collectors are on GitHub. The dataset is 11,000+ cleaned, captioned memecoin logos and meme templates and grows every day from pump.fun. Every run so far was rented by the hour on RunPod and stopped the moment it ended, because the budget was a few dollars. That caps Cookr at small LoRA runs. Cookr v2 needs full-weight training on the 20B Qwen-Image base and weekly retraining on fresh launches. This goal buys one dedicated training node: a new NVIDIA H100 80GB PCIe (eBay item 167577727476, seller yudr3m, Ottawa, Canada, C$38,435.78, free tracked shipping) in a Supermicro SYS-741GE-TNRT tower GPU server (eBay item 168323768877, used barebone with dual 2000W Titanium PSUs, officially supports H100 PCIe), with one Intel Xeon Silver 4510, 128 GB DDR5-4800 ECC and a 2 TB NVMe. Built and run by [Name], the developer behind @CookrAI on GitHub and @CookrPro on X.
80 GB of HBM3 is the smallest card that trains Qwen-Image full-weight without sharding or offloading; the 48 GB cards cannot merge the 20B checkpoint at all, which is why v1 shipped as LoRA-only. Renting an H100 24/7 costs about $2,000 a month on Lambda, so the card pays for itself in 14 months of the use we need, and after that Cookr trains for the price of electricity. Buying one card rather than cloud credit also means the machine keeps serving the model on cookr.pro between runs.
The Cookr developer only, for training, evaluating and serving Cookr models. Every model the machine produces is published under the CookrAI Hugging Face account (light versions open, full model served free on cookr.pro), and each run's config, log and sample grid goes to the public GitHub repo. Holders get what everyone gets: the models and the receipts. Buying the coin does not give ownership of the hardware, access to it, or any share of what it produces.
Item: H100 80GB PCIe, C$38,435.78 = 26,961 USD at 0.7014 USD/CAD on 7 Oct 2026, free international shipping. Other purchase costs, 13,984 USD: Supermicro SYS-741GE-TNRT barebone 8,946 (eBay 168323768877, THB 301,391, free shipping and returns), Xeon Silver 4510 new 765 (eBay, UAE seller, incl. delivery), 4x 32GB DDR5-4800 ECC RDIMM 4,073 (eBay mem-store, US, incl. delivery), 2 TB NVMe and cables 200 (Bangkok retail, estimate). Taxes: Thai import VAT 7% on the CIF value of all imported parts (40,745 USD) = 2,852 USD. Computer parts under HS 8471/8473 are assumed duty-free; if customs classifies the card differently the reserve absorbs it. Reserve: 10% (4,380 USD) for currency moves, a replacement seller at a higher price, or a customs surprise. Total 48,177 USD at SOL = 120.89 USD (Jupiter, 7 Oct 2026) = 398.5 SOL. Target: 400 SOL. If SOL rises the surplus stays in the goal wallet and is published; if it falls, see Risks.
Day 0: goal reached. Hours 0 to 48: platform release delay, nothing moves. Day 2 to 3: buy the H100 from the documented listing and the barebone, CPU and RAM from theirs; post every order confirmation with amount and listing screenshot. Day 3 to 21: international delivery to Bangkok (eBay estimates 3 to 4 weeks for the card), customs clearance, VAT receipt published. Day 21 to 25: assembly, serial-number photo, nvidia-smi output, 24-hour burn-in log. Day 25 to 30: first Cookr v2 training run starts; config and sample grid published on GitHub and X. Any slip beyond day 30 gets a public note with the reason.
(@CookrAI on GitHub, @CookrPro on X), sole developer of Cookr, is the custodian. No relationship to any seller; these are retail purchases on eBay. The machine lives at the developer's studio in Bangkok. Evidence published on cookr.pro and pinned on X: order confirmations, invoices with serial numbers, customs VAT receipt, assembled-machine photo, nvidia-smi output, first training log. Every later model release states it was trained on this machine.
Paid by the developer, not by the goal. Full load is about 700 W for the whole node (350 W card, CPU, fans), roughly 17 kWh a day, about 85 THB (2.5 USD) a day at Bangkok rates, under 80 USD a month. No insurance; the H100 listing has no seller returns, so dead-on-arrival is covered by the eBay Money Back Guarantee, the barebone by the seller's free returns, and anything after that by the developer. Cooling, space and internet are already in place.
H100 listing disappears or the seller cannot ship: buy the same part from Supermicro Store (SKU GPU-NVH100-80, 27,295.40 USD, listed as backorderable) or the next public listing at or below that price, and publish the new link; the reserve covers the difference. Card arrives dead or not as described: eBay Money Back Guarantee claim, then re-buy. Customs applies duty beyond the 7% VAT: paid from the reserve, receipt published. SOL falls so the wallet no longer covers the purchase: the goal stays open until it does, or the developer tops up the gap and says so. Purchase fails entirely: funds remain in the goal wallet and the next step is announced publicly; there is no automatic refund.
Creator-supplied documents. A listing or quote does not prove an acquisition.
What is bought: one new NVIDIA H100 80GB PCIe (eBay item 167577727476, seller yudr3m, Ottawa, Canada, C$38,435.78 = 26,961 USD, free tracked shipping to Thailand, 10 in stock on 7 Oct 2026) installed in a used Supermicro SYS-741GE-TNRT tower GPU server (eBay item 168323768877, 8,946 USD, dual 2000W Titanium PSUs, H100 PCIe officially supported) with one Intel Xeon Silver 4510 (765 USD), 4x 32GB DDR5-4800 ECC RDIMM (4,073 USD) and a 2 TB NVMe (200 USD, Bangkok retail). Thai import VAT 7% on the imported parts: 2,852 USD. Documented budget 43,797 USD plus a 10% reserve = 48,177 USD = 398.5 SOL at 120.89 USD/SOL (Jupiter, 7 Oct 2026), rounded to 400 SOL. What it is for: training Cookr v2, the full-weight successor of the public cookr-v1-light models (huggingface.co/CookrAI). 80 GB of HBM3 is the smallest card that trains and merges the 20B Qwen-Image base without sharding. Every run so far was rented by the hour on RunPod; this node runs 24/7 for the cost of electricity. Who holds it: [Name] (@CookrAI on GitHub, @CookrPro on X), sole developer of Cookr, at the developer's studio in Bangkok. Not affiliated with any seller. Upkeep: electricity (about 2.5 USD a day at full load), cooling and internet are paid by the developer, never from the goal wallet. Receipts published on cookr.pro and pinned on X: order confirmations, invoices with serial numbers, customs VAT receipt, assembled-machine photo, nvidia-smi output, first training log. Every model trained on it says so on its model card. Fallback: if the eBay listing is gone, the same part from Supermicro Store (SKU GPU-NVH100-80, 27,295.40 USD). No returns from the card seller; eBay Money Back Guarantee covers dead on arrival. No automatic refund if the purchase fails; funds stay in the goal wallet and the next step is announced publicly. Holders own the coin, not the hardware.
proof link| side | trader | SOL | $Cookr | price | when |
|---|---|---|---|---|---|
| buy | B3wR…eiyw | 0.065 | 1.8M | $0.00000424 | 2h ago |
| sell | 7Epc…sANW | 0.11 | 3.0M | $0.00000424 | 2h ago |
| sell | 6XEz…3YVv | 0.078 | 2.2M | $0.00000426 | 2h ago |
| buy | 6XEz…3YVv | 0.041 | 1.2M | $0.00000427 | 2h ago |
| buy | 4DK7…PJEm | 0.41 | 11.7M | $0.00000421 | 2h ago |
| buy | 4DK7…PJEm | 0.41 | 12.0M | $0.00000411 | 2h ago |
| buy | 4DK7…PJEm | 0.082 | 2.4M | $0.00000405 | 2h ago |
| sell | bark…uXYJ | 0.09 | 2.7M | $0.00000405 | 2h ago |
| sell | Gcjz…eCtb | 0.048 | 1.4M | $0.00000407 | 2h ago |
| sell | 4u4e…CwRj | 1.08 | 27.6M | $0.00000471 | 2h ago |
| sell | 7cjb…VxrP | 0.88 | 25.4M | $0.00000418 | 2h ago |
| sell | D3Da…K1HB | 1.08 | 29.3M | $0.00000443 | 2h ago |