The Social History of Compute — Credit, Crypto, AI, and Space

CPUs, GPUs, memory, storage, networks, electricity, cooling, and operating time are finite resources claimed by several uses at once. An “idle cycle” still has marginal power and heat costs, while an owner chooses among science, gaming, cryptocurrency, AI, business work, or doing nothing. Capacity does not flow naturally to one purpose; software support, reward, story, cost, and institutional design allocate it.

A straight story in which benevolent science was taken by Bitcoin, passed through Ethereum, and finally became AI erases evidence. Bitcoin’s GPU era was a transition toward ASICs, AI use of GPUs predates Ethereum, and Folding@home expanded dramatically under an urgent social goal after cryptocurrency markets already existed. Overlapping uses and different hardware classes must remain separate.

This article separates documented institutions and dated observations from limited inferences and unknown causal claims. “Human desire never stops” is not a historical fact a source can verify. Evidence can show which systems measured, recognised, or paid for which behaviour.

BOINC credit is an account of validated computational contribution. It is not money, ownership, voting power, or the scientific proof itself. Total credit records accumulation, recent average credit (RAC) exposes recent activity, and leaderboards for hosts, volunteers, and teams—plus third-party cross-project statistics—made worldwide competition visible.

BOINC’s designers recorded that leaderboards encouraged volunteers to add several PCs, upgrade hardware, or buy dedicated machines, while team competition recruited friends, relatives, and coworkers. SETI@home’s retrospective likewise describes a culture combining a screensaver, scientific purpose, rank, and team affiliation. Points were not decoration; they became a social institution for persistence and belonging.

Not every project was entirely unrewarded. GIMPS includes discovery attribution and awards, while distributed.net shared RC5 challenge prize money. The accurate generalization is that routine work was usually returned with scientific participation, credit, rank, badges, and certificates—not with a tradable payment priced per job.

Credit design cannot be separated from integrity. Once a number matters, it can encourage inflated claims, modified clients, unauthorised machines, or selective return. Validators, replication, passkeys, return-rate conditions, and fair-play rules developed alongside the game. Replication and quorum remain project-specific rather than one universal validation method.

Bitcoin did not attach a price to arbitrary scientific computation. It embedded a SHA-256d target search into block proposal and chain selection, allocating the block subsidy and transaction fees to a successful block. When a pool is used, a lower-difficulty share measures internal contribution; it is not itself a Bitcoin consensus block.

BOINC credit mostly remembers who contributed how much validated research computation. Bitcoin reward determines who satisfied a protocol condition and obtained an economically transferable claim. A pool may resemble a project server from the outside, but purpose, verifier, reuse of output, and transferability of reward differ.

This difference can change an owner’s opportunity cost: the same electricity and machine might be assigned to credit and scientific participation, to a market-priced reward, or to neither. Yet the institutional possibility of changing choices is not a measurement of how many machines actually moved.

No cross-platform dataset establishing that causal claim was found. SETI@home active participation had declined from its early publicity peak before Bitcoin appeared. A self-selected 2006 BOINC survey recorded several reasons for leaving—computer problems, complexity, forgetting to restart, lost interest, and electricity cost—but it is not a representative sample from which causal shares can be calculated.

A 2012 account estimated about 900,000 active volunteer computers, roughly 10 PFLOPS, and GPUs supplying about 70 percent of capacity. BOINC’s 2019 platform paper reports an around-2018 snapshot of about 700,000 devices, 560,000 GPUs, and an average 93 PFLOPS. Different scope and definitions prevent those dots from becoming one trend line, and faster hardware separates participant count from FLOPS.

During the 2020 COVID-19 response, Folding@home reported that more than 700,000 citizen scientists joined over three weeks and that participation increased twentyfold; its project timeline records reaching roughly 1.5 exaFLOPS on 25 March 2020. This is neither sustained output nor directly comparable with BOINC, but it is a counterexample to the claim that cryptocurrency markets had made large-scale voluntary mobilisation impossible.

CUDA arrived in 2006 and OpenCL 1.0 in 2008, expanding software support for GPU work beyond graphics. BOINC widened GPU support in 2008, while SETI@home and GPUGrid distributed scientific applications. AlexNet trained on two GeForce GTX 580 cards in 2012, so GPU-based AI research predates Ethereum.

Bitcoin saw a public OpenCL miner in 2010 and an FPGA implementation in 2011; the first Avalon ASIC shipment followed in January 2013. Repetitive fixed SHA-256d tied to revenue created a strong incentive to leave general GPUs for dedicated circuitry. Modern competitive Bitcoin mining is SHA-256 ASIC-led; Bitcoin did not remain a long-term consumer-GPU workload.

Science, gaming, Bitcoin, Ethereum, and AI overlapped rather than replacing each other one at a time. Public evidence does not track how many GPUs moved out of BOINC or Folding@home into cryptocurrency, so lane width or a timeline cannot be interpreted as market share.

Ethereum Frontier launched on 30 July 2015, and Mainnet produced blocks under Ethash proof of work. Ethash used a large dataset and memory-hard design that kept GPU mining viable for years, but it did not exclude ASICs forever; dedicated hardware later appeared. ASIC-resistant and ASIC-proof are different claims.

Company filings establish that cryptocurrency demand affected GPU markets. AMD’s 2017 Form 10-K recorded demand for GPUs suited to mining currencies including Ethereum. In Q3 2018 AMD said blockchain-related GPU sales had fallen from a high-single-digit percentage of total revenue in the prior-year quarter to negligible. NVIDIA’s FY2018 Form 10-K also recorded a large increase in mainstream OEM GeForce products targeted at cryptocurrency mining.

Vendors also disclosed limited visibility into end use and an inability to measure the impact precisely. Demand, channel inventory, and resale effects are observable; attributing all of them to Ethereum alone—or to devices leaving BOINC—is not.

Ethereum Mainnet switched to proof of stake in the Merge on 15 September 2022, ending mining as a means of producing valid Mainnet blocks. ethereum.org estimates an energy-consumption reduction of about 99.95 percent. This was a consensus change for Ethereum Mainnet, not the end of other proof-of-work chains or Bitcoin mining.

NVIDIA later said the Merge may have reduced the utility of GPUs for mining and increased resale into secondary markets, potentially affecting demand for lower-end products. The same disclosure listed macroeconomic conditions, China, and channel inventory among other factors; it does not support assigning gaming decline to the Merge alone.

Former Ethereum GPUs could be sold, returned to gaming, used by another proof-of-work chain, run compatible science or small AI jobs, or be shut down. No primary dataset was found that measures those destinations. “Mining GPUs became AI” turns a compatibility possibility into an unsupported migration claim.

GPU use in AI did not suddenly begin after cryptocurrency. AlexNet reported training its 2012 ImageNet model on two GTX 580 GPUs for five to six days. Low-precision matrix operations, HBM, ECC, fast interconnects, collective communication, compilers, and frameworks later grew this approach into data-centre systems for training and inference.

NVIDIA’s FY2026 Form 10-K reports Data Center revenue of $193.7 billion, total revenue of $215.9 billion, and Gaming revenue of $16.0 billion. Data Center includes networking and other products; it is not all generative-AI GPU revenue. AMD’s FY2025 Form 10-K/A reports $16.6 billion of Data Center revenue and demand for EPYC and Instinct products, while Intel’s FY2025 Form 10-K says it missed the major demand shift toward AI-optimised GPUs and had not become a meaningful participant through Gaudi.

“Now AI” is therefore supportable as a market emphasis visible in several companies’ product roadmaps and revenue mix. Outcomes differ sharply by company, and those filings cannot be summed into an industry-wide AI total. Nor does this mean household mining boards physically became H100-class clusters. Consumer boards and data-centre systems differ in memory capacity, precision, interconnect, reliability, software support, cooling, and operational guarantees.

FLOPS, hashes per second, tokens per second, and BOINC credit are not convertible units. Even within one silicon family, application, precision, memory traffic, communication, and validation differ. Resource allocation must ask which workload runs under which software and system boundary at which energy cost—not merely count devices.

Space computing is not solely future tense. NASA says the first Spaceborne Computer sent to the ISS in 2017 operated a roughly one-TFLOP COTS system for 207 days without a reset. For Spaceborne Computer-2, sent to the ISS in 2021, NASA’s 30 December 2022 operations report records the units being prepared for return and describes an experiment that tested onboard data processing, AI, and error mitigation.

ESA’s PhiSat-1 launched on 3 September 2020 and used an Intel Movidius Myriad 2 to classify clouds in Earth-observation images, discarding low-value images before downlink. This demonstrated edge AI near the sensor; it was not a generally available orbital cloud data centre.

NASA’s High-Performance Spaceflight Computing project is developing a new processor that combines higher performance with radiation tolerance. Its 2026 status reports manufactured chips and testing in progress. It should not be called space-qualified or deployed before those tests establish that status.

ESA’s Cognitive Cloud Computing in Space campaign distinguishes application demonstrations on existing orbital hardware from short studies toward a future networked-data-centre vision. ESA’s 2023 review identifies prototypes and higher technology readiness as subsequent work. These are important development steps, not evidence that a permanent public utility already replaces terrestrial cloud infrastructure.

Space lacks atmospheric convection, so heat moves through structures and leaves through radiators. Thermal design, single-event radiation effects, spacecraft power and mass budgets, and communication delay or disruption belong inside the system boundary. Solar exposure does not make electricity or cooling free.

No primary evidence located through 23 August 2026 showed a commercially operating Bitcoin-mining farm or BOINC data centre in space. That does not prove such a system can never exist, but it cannot be shown as a current deployment. A demonstrated near-term use case is processing sensor data where it is produced to reduce downlink and latency.

People run computers for scientific participation, recognition, affiliation, competition, income, curiosity, and public purpose. Evidence can document behaviour and institutions but cannot reduce human desire to one metric. A world-class library should preserve a traceable relationship: who computed what, on which machine, at whose cost, under whose validation, and for whose benefit.