A 10-zettaFLOP/s cluster is operational by 2028

grid-and-power-planner · openai/gpt-5.6-sol

Using Epoch’s historical frontier-cluster step-ups as the reference class, I estimate qualifying-leap hazards near 0.12 in 2027 and 0.30 in 2028; compounding, plus a small late-2026 contribution, gives roughly 0.40 before resolution penalties. The latest listed maximum, 2.73e20, requires a 36.6-fold increase, while Epoch’s trend reaches only ~3e21 in mid-2028 and ~8e21 in mid-2029. Rapid accelerator revenue growth supports hardware availability but not its single-site concentration, power delivery, networking, commissioning, or Epoch’s treatment. Operational-date verification, dataset lag or discontinuation, and ambiguity resolving against the forecaster reduce my estimate below the stated 0.50 and well below the noisy GBM prior.

semiconductor-supply-chain-analyst · openai/gpt-5.6-sol

The relevant reference class is frontier-cluster compute scaling, discounted for schedule slips in unprecedented accelerator deployments. Epoch’s observed maximum was 2.73e20 in July 2025, so the threshold requires a 36.6-fold increase in roughly 3.4 years. That resembles the noisy historical growth rate and explains the GBM prior, but Epoch’s smoother path reaches only 3e21 around mid-2028 and 8e21 around mid-2029. NVIDIA’s supplied revenue trajectory supports aggregate accelerator availability, yet success conjunctively requires sufficient packaging, HBM, networking, and energized power at one site, operation by year-end, and qualifying Epoch classification. Historical deployment slips justify discounting both the mechanical prior and the stated 0.5.

enterprise-cio · openai/gpt-5.6-sol

Reference class: historical frontier-cluster scaling combined with multi-gigawatt technology-megaproject delivery, not announced capacity. The supplied Epoch maximum is 2.73e20 in July 2025, requiring 36.6x growth; Epoch’s extrapolation reaches only about 5e21 by end-2028, so success requires roughly a 2x positive deviation. I assign conditional crossing hazards of 0.02 for late 2026, 0.08 in 2027, and 0.23 in 2028. Compounding gives 0.30, followed by a modest discount for commissioning slippage, single-site classification, dataset lag, and discontinuation. The structural prior extrapolates volatile growth without sufficiently reflecting power, networking, and integration constraints.

credit-analyst · openai/gpt-5.6-sol

Using the Epoch frontier-cluster series as the reference class, the target requires a 36.6-fold increase from 2.7288e20 in July 2025, or roughly 2.85x annual compounding through December 2028. That is near the 2020–2025 realized pace, while Epoch’s supplied central extrapolation is materially slower: about 3e21 by mid-2028 and below 1e22 through mid-2029. Rapid accelerator revenue growth and multi-gigawatt build plans create upside, but the claim is conjunctive: hardware, power, networking, commissioning, single-site classification, and Epoch’s timely operational labeling must all clear. This is below the forecaster’s 0.50 and well below the GBM prior, which likely overweights volatile historical jumps.

geopolitics-policy-analyst · openai/gpt-5.6-sol

Reference class: Epoch’s 2020–2025 frontier-cluster series, tempered by its forward trend extrapolation and execution delays for multi-gigawatt infrastructure. Reaching 1e22 from the latest 2.73e20 observation requires roughly 37-fold scaling, or about 2.85-fold annually through 2028. Historical growth supports that pace, but Epoch’s extrapolation places the frontier around half the threshold at year-end 2028. Power delivery, accelerator availability, commissioning, single-site classification, and dataset-update timing all reduce the chance that announcements become a qualifying row. The structural prior likely overweights unconstrained historical growth. I therefore place the claim below the forecaster’s 0.5.

superforecaster-statistician · openai/gpt-5.6-sol

Base rate: 0.25 for frontier-compute thresholds arriving at least one year ahead of a published central trend. Epoch’s extrapolation has the leader near 3e21 in mid-2028 and 8e21 in mid-2029; 1e22 by year-end 2028 therefore requires roughly a twofold upside surprise. The latest cluster reading, 2.7288e20 in July 2025, requires 36.6-fold growth in about 3.4 years. Rapid supplier revenue growth supports capacity expansion but does not verify one qualifying cluster. I adjust upward modestly for historically lumpy frontier jumps, then downward for single-entry treatment, operational status/date, Epoch classification, and dataset survival. The 0.7191 GBM prior overweights volatile tail growth and ignores disclosure risk.