Beyond OCPI futures on ICE — the tradable basket for the compute-as-commodity thesis
The question
Beyond OCPI futures on ICE, which existing tradable instruments already give exposure to the "compute-as-commodity" thesis (HBM-makers, hyperscaler-capex beneficiaries, power-infra plays, exchange operators), and what's the highest-quality basket for a long-only RDCO investing position? Context: OCPI compute futures listing on ICE (NYSE: ICE) was flagged as a derivative play on compute-becoming-a-commodity, a 12-month thesis. The vault tracks hyperscaler capex and has a Cerebras IPO breakdown but no consolidated exposure basket.
What we already know (from the vault)
- The seed note frames ICE itself as "a derivative play on the Singularity" — the venue capturing transaction-fee economics from compute's commoditization, but explicitly "not a clean thesis yet, needs the futures to actually clear and gain volume." The essay is sponsor-tainted (Wissner-Gross advises Ornn); the historical analogs (oil→NYMEX, gas→Henry Hub) are sound, the "this product will be the Brent-of-compute" claim is the load-bearing self-interested part [[2026-05-19-innermost-loop-first-major-exchange-compute-futures]].
- The memory/HBM bottleneck is RDCO's most-instrumented live thesis: an executable v1.1 doc with a MU + SMH + SNDK + INTC basket, tranche-accumulate entry, no mechanical exits, and anchor-based exit triggers. Korean primaries (SK Hynix, Samsung) are noted but excluded for Alpaca tradability [[2026-05-18-memory-cycle-v1.1]].
- As of July 2026 the cycle reads deep Phase 2 with the first flicker of a top forming: DRAM contract-price gains decelerated from +58–63% QoQ (2Q26) to +13–18% (3Q26), but inventories are still at supercycle-low single-digit weeks and new greenfield bit supply is dated 2029–2030. Zero Phase-3 exit triggers fired; earliest plausible turn window Q3 2026–H1 2027 [[2026-07-07-dram-hbm-phase2-phase3-early-signals]].
- Smart-money 13F corroboration by tier: memory held by 2/8 managers (MU, Seagate), power by 3/8 (Vistra, GE Vernova via Tepper/Druckenmiller/Tiger), ai-infra-chips by 4/8 (NVDA, AVGO, TSM, AMD), hyperscalers by 7/8. 2026 Q1 saw Druckenmiller + Tepper open SANDISK and Druckenmiller + Tiger open INTEL [[2026-05-17-aggregate-2yr-backfill-summary]].
- The vault explicitly rejects Cerebras as a core position despite it being the purest wafer-scale IPO: 86% revenue from two related Abu Dhabi entities, upside-down cap table, single-foundry TSMC dependency, and a disclosed material weakness in revenue-recognition controls. Filed as "why we don't chase the wafer-scale story," not a candidate [[2026-05-12-mostlymetrics-cerebras-ipo-s1-breakdown]].
What the web says
- Hyperscaler 2026 capex is ~$725B combined, up ~77% YoY (Amazon ~$200B, Google ~$175–185B, Meta ~$125–145B, Microsoft ~$110–120B). Goldman models ~$5.3T combined FY25–30. As-of Feb 2026 analyst compilations; guidance ranges, not primary-filing-verified here (Tom's Hardware, CNBC Feb 6 2026).
- The exchange-operator tier is now a race, not a single name. ICE + Ornn announced GPU compute futures on OCPI (May 19 2026, pending regulatory approval); CME Group plans compute futures with Silicon Data (also pending review); Architect's AX exchange plans GPU and DRAM futures on Ornn's benchmark. Ornn raised $33M led by a16z. All contracts $-denominated, cash-settled, none live yet (ICE IR, Markets Media).
- Power is the hardest physical bottleneck. Data-center electricity demand is projected to roughly double from ~485 TWh (2025) to ~950 TWh (2030); ~$1.4T of electrification spend needed by 2030 (24/7 Wall St Mar 2026).
- Vistra (VST): signed a 20-yr PPA with AWS (up to 1,200MW) and a 2,600+MW PPA with Meta from its Comanche Peak nuclear plant (24/7 Wall St May 2026).
- GE Vernova (GEV): Q1 2026 orders $18.3B (+71% organic), with $2.4B of electrification equipment orders for data centers in the quarter — more than all of last year. Eaton (ETN): record Q1 2026 Electrical Americas sales $3.6B, segment backlog +44% YoY, data-center backlog described as ~11 years of 2025 construction. Switchgear/transformer lead times have stretched to multi-year (Yahoo Finance). All figures per company reporting, as-of Q1 2026.
- Memory capex is discipline-shaped, not glut-shaped: Micron ~$13.5B (+23%), SK Hynix ~$20.5B (+17%), Samsung ~$20B, going to node migration + HBM packaging, not greenfield wafers. A Korea-government-backed ~$575B mega-buildout was announced (Jul 2026) but bit supply lands 2029–2030 (TrendForce Jul 3 2026, Motley Fool Jul 6 2026).
Convergences and contradictions
- Convergence — the demand engine is real and accelerating. Vault ($750B+ committed 2026 hyperscaler capex) and web (~$725B, +77% YoY) agree the money is being spent; power, memory, and foundry are the three physical chokepoints it flows through, and smart money is already positioned in all three.
- Convergence — the exchange-operator play is the truest expression of the literal thesis but the longest fuse. Vault ("not a clean thesis yet, needs volume") and web (all three venues pending regulatory approval, none live) agree ICE/CME are call-options-on-commoditization layered on already-great businesses, not pure-plays.
- Contradiction to hold — memory is simultaneously the best-corroborated tier and the one closest to its cycle top. The vault's own tripwire brief shows the first deceleration flicker (a top forming), while the $575B greenfield announcement raises the Phase-3 prior without moving the clock (bit supply 2029–30). Power carries no equivalent rollover signal, which is why it may be the more durable long-only expression despite memory being the more instrumented thesis.
Synthesis for RDCO
This is thesis research and an exposure map, not a trade order and not a paper-trade authorization. Nothing here says "buy X now." It says: here are the existing instruments that already express compute-as-commodity, ranked by how cleanly they capture a real bottleneck for a long-only, capital-cycle-horizon holder.
The organizing insight: "compute-as-commodity" is not one trade, it is a stack of bottlenecks. Furthest downstream is the marketplace where the commodity clears (exchange operators). Above it sit the two hard physical constraints that resist commoditization because they have multi-year lead times — power (generation + electrical distribution) and memory/HBM (the packaging-gated bottleneck inside the chip). Feeding all of it is hyperscaler capex, whose single cleanest chokepoint is the foundry. The wildcard (Cerebras) is a pure compute-supplier the vault has already tested and rejected. A long-only basket should overweight the bottlenecks that are hardest to arbitrage away and underweight the ones that are either commodities themselves (raw compute suppliers) or immaterial near-term (exchange fees on a market that isn't live).
The basket (5–10 names, by tier):
| Tier | Ticker | One-line thesis (bottleneck captured) | Bear / invalidation |
|---|---|---|---|
| 1 — Exchange operators | ICE | Owns the venue where OCPI GPU compute futures will clear; earns fees on volume regardless of who is long/short compute | Compute futures are immaterial to ICE revenue for years; regulatory approval pending; thesis is a call-option on an already-diversified exchange, not a pure-play |
| 1 — Exchange operators | CME | Second horse in the compute-futures race (with Silicon Data); same fee-on-commoditization logic, different index partner | Same near-term immateriality; whichever venue wins liquidity is unknowable now — this tier is a duopoly bet, not a single winner |
| 2 — HBM / memory bottleneck | MU | Only US-listed DRAM/HBM pure-play; direct HBM-tightness exposure; vault's live thesis anchor | Cyclical; closest tier to its Phase 2→3 turn (window Q3 2026–H1 2027); the $575B greenfield raises the down-cycle prior |
| 2 — HBM / memory bottleneck | SMH | Semis-basket ETF; captures SK Hynix + Micron + NVDA HBM tailwind without single-name or Korea-listing risk | Diluted by non-memory semis; broad-market beta; not a pure bottleneck expression |
| 2 — HBM / memory bottleneck | SNDK | NAND tier-2; captures the crowd-out cascade from HBM into NAND; fresh 2026 Q1 smart-money entry (Druckenmiller + Tepper) | Signal is <2 quarters old — confirm it held next 13F; NAND tightness is a second-order effect, weaker than HBM |
| 3 — Power infrastructure | VST | Merchant + nuclear generation with signed hyperscaler PPAs (AWS, Meta); electricity is the true binding constraint | Merchant power-price and regulatory sensitivity; already run hard; PPA execution risk |
| 3 — Power infrastructure | GEV | Turbines + grid electrification equipment; DC-electrification orders exceeding prior full-year run-rate | Order cyclicality; valuation; gas-turbine slot backlog could cap near-term deliverable growth |
| 3 — Power infrastructure | ETN | Electrical distribution / switchgear — the multi-year-lead-time layer that can't be commoditized away | Supply-chain normalization compresses the lead-time premium; industrial-cyclical; valuation |
| 4 — Hyperscaler-capex beneficiary | TSM | The single foundry chokepoint under NVDA, AVGO, Cerebras and custom silicon alike; the vault's flagged "next position after Cerebras" | Geopolitical (Taiwan) concentration; capex-cycle exposure; already widely held (3/8 smart-money managers) |
| 5 — Wildcard (flagged AVOID) | Cerebras | Purest public wafer-scale / raw-compute-supplier bet on the commodity itself | Vault verdict is AVOID: 86% customer concentration, upside-down cap table, single-foundry TSMC, disclosed material weakness. Post-IPO ticker/price UNVERIFIED here |
Accessibility flags: SK Hynix (000660.KS) and Samsung (005930.KS) are the actual HBM leaders but are Korea-listed and not cleanly tradable on RDCO's current broker (Alpaca) — SMH, or the Korea ETF EWY, are the accessible proxies. The hyperscalers themselves (GOOGL/AMZN/MSFT/META) are the demand engine and are held by 7/8 tracked managers, but as compute buyers their exposure is diluted across enormous businesses; they belong in a portfolio for the "cash-is-the-ultimate-commodity" reason ([[2026-06-02-stratechery-google-capital-company]]) more than as a bottleneck play. NVDA / AVGO are the obvious substrate names but the vault treats them as structural holders, not fresh entries, and they are the most picked-over.
Highest-conviction call, honestly calibrated. For a long-only, capital-cycle-horizon holder, Tier 3 (power) edges out Tier 2 (memory) as the most durable expression, for one reason: power/electrical infrastructure has the longest lead times (transformers, switchgear, generation interconnect queues measured in years) and is the hardest bottleneck to arbitrage or glut away, and it currently shows no rollover signal — whereas memory, the better-instrumented and better-corroborated thesis, is closest to its cycle top (the first ASP-deceleration flicker has already appeared). Memory is the higher-conviction thesis but carries the highest timing risk. Tier 1 (exchange operators) is the truest to the literal "compute-as-commodity" framing but the lowest near-term conviction — a small, patient call-option allocation at most, sized for the multi-year payoff if compute futures actually clear and gain volume, not for 2026 earnings. Tier 4 (TSM) is the diversifying anchor that sits under every other tier. Cerebras stays a flagged wildcard, not a holding.
If this were to be sized, the shape would be: overweight the two physical bottlenecks (power + memory), anchor with the foundry, hold a small long-fuse position in the exchange-operator duopoly, and keep Cerebras as a watch-item only. But sizing is a separate decision gated by /investing:build-thesis and a founder deploy-gate — this brief is the exposure map that feeds that, not the order.
Open follow-ups
- Run
/investing:build-thesison the power tier (VST / GEV / ETN) — it is the one high-conviction tier without an executable RDCO thesis doc yet; memory already has v1.1. - Decide whether the exchange-operator tier warrants its own micro-thesis or stays a watch-item until OCPI/CME compute futures actually clear and post volume (the vault's stated trigger).
- Add ICE/CME/compute-futures launch to the smart-money-watch surface — position data becomes a new signal channel once the contracts trade (already queued in the seed note).
- Verify Cerebras post-IPO ticker + trading price before any wildcard sizing; the vault only has the S1-filing snapshot (~$33B implied, ~May 2026).
- Pull the next hyperscaler-capex pulse via
/investing:edgar-watchafter Q2 2026 10-Qs to confirm the $725B guidance is holding in actual spend, not just guidance. - Cross-check the power names against the Q2 2026 13F cron (~Aug 2026) to see if the 3/8-manager power exposure is still building.
Why this is in the vault
This brief closes the gap between RDCO's instrumented memory/HBM thesis (v1.1, with a live MU + SMH + SNDK + INTC basket) and the uninstrumented power and exchange-operator tiers — giving the Markov phase-tracker pipeline a full four-tier exposure map to work against. It is the direct input to the next /investing:build-thesis run targeting VST / GEV / ETN, which is the one high-conviction tier the vault currently lacks an executable doc for.
Related
- [[2026-05-19-innermost-loop-first-major-exchange-compute-futures]] — the OCPI/ICE seed note that flags exchange operators as the compute-commodity venue play
- [[2026-05-18-memory-cycle-v1.1]] — executable memory/HBM thesis; the Tier 2 basket and exit anchors
- [[2026-07-07-dram-hbm-phase2-phase3-early-signals]] — current phase read (deep Phase 2, first top-forming flicker) for the memory tier
- [[2026-05-12-mostlymetrics-cerebras-ipo-s1-breakdown]] — the wildcard's AVOID verdict and reasoning
- [[2026-05-17-aggregate-2yr-backfill-summary]] — smart-money 13F cross-reference across all tiers
- [[2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] — parent AI-infrastructure thesis + hyperscaler-capex anchor framing
- [[2026-06-02-stratechery-google-capital-company]] — "cash is the ultimate commodity" framing for the hyperscaler-buyer tier
- [[2026-05-27-markov-equities-pipeline-spec]] — the phase-tracker pipeline the memory tier feeds
Sources
Vault
- [[2026-05-19-innermost-loop-first-major-exchange-compute-futures]] —
~/rdco-vault/06-reference/2026-05-19-innermost-loop-first-major-exchange-compute-futures.md - [[2026-05-18-memory-cycle-v1.1]] —
~/rdco-vault/01-projects/investing/theses/2026-05-18-memory-cycle-v1-1.md - [[2026-07-07-dram-hbm-phase2-phase3-early-signals]] —
~/rdco-vault/06-reference/research/2026-07-07-dram-hbm-phase2-phase3-early-signals.md - [[2026-05-12-mostlymetrics-cerebras-ipo-s1-breakdown]] —
~/rdco-vault/06-reference/2026-05-12-mostlymetrics-cerebras-ipo-s1-breakdown.md - [[2026-05-17-aggregate-2yr-backfill-summary]] —
~/rdco-vault/01-projects/investing/anchors/smart-money/2026-05-17-aggregate-2yr-backfill-summary.md - [[2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] —
~/rdco-vault/06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis.md - [[2026-06-02-stratechery-google-capital-company]] —
~/rdco-vault/06-reference/2026-06-02-stratechery-google-capital-company.md - [[2026-05-27-markov-equities-pipeline-spec]] —
~/rdco-vault/01-projects/investing/2026-05-27-markov-equities-pipeline-spec.md
Web
- ICE and Ornn to Launch GPU Compute Futures Contracts — ICE IR, May 19 2026 — https://ir.theice.com/press/news-details/2026/ICE-and-Ornn-to-Launch-GPU-Compute-Futures-Contracts/default.aspx
- ICE, Ornn to Offer GPU Compute Futures (notes CME + Silicon Data, Architect AX) — Markets Media — https://www.marketsmedia.com/ice-ornn-to-offer-gpu-compute-futures/
- Big Tech's AI spending plans reach $725 billion in 2026 — Tom's Hardware — https://www.tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion
- Tech AI spending approaches $700B in 2026 — CNBC, Feb 6 2026 — https://www.cnbc.com/2026/02/06/google-microsoft-meta-amazon-ai-cash.html
- How AI Data Centers Are Reshaping the Power Market (Vistra PPAs) — 24/7 Wall St, May 2026 — https://247wallst.com/investing/2026/05/04/how-ai-data-centers-are-reshaping-the-power-market-and-the-4-plays-investors-are-making/
- $1.4 Trillion Needed for AI Data Center Electrification by 2030 — 24/7 Wall St, Mar 2026 — https://247wallst.com/investing/2026/03/03/1-4-trillion-needed-for-ai-data-center-electrification-by-2030-chief-investment-officer/
- 5 Stocks Solving the AI Power Crisis (GE Vernova, Eaton Q1 2026) — Yahoo Finance — https://finance.yahoo.com/markets/stocks/articles/5-stocks-solving-ai-power-152000530.html
- AI Server Demand Continues to Support Memory Prices in 3Q26, but Gains Moderate — TrendForce, Jul 3 2026 — https://www.trendforce.com/presscenter/news/20260703-13134.html
- Micron's Strategy Will Face SK Hynix and Samsung's $575 Billion Spending Plans — Motley Fool, Jul 6 2026 — https://www.fool.com/investing/2026/07/06/microns-strategy-face-sk-hynix-samsung-spend/