The four venues are not four versions of the same product. They are four different answers to the question "how do you build a forward curve for a commodity that has never had one?"

• Kalshi (live today). Binary threshold ladders — "will the Ornn print for B200 exceed $K on date T?" — at weekly and month-end expiries. The forward isn't quoted anywhere; it's the strike where the implied CDF crosses 50%. This is reading a forward off an options ladder before the futures exist, and it only reaches out a few weeks. But it trades, now, on a CFTC-regulated venue.
• Architect (two tracks, both pending). Offshore: AX, a Bermuda-regulated institutional exchange, announced compute and DRAM perpetual futures in January — funding-rate-anchored continuous prompt exposure, settling to Ornn Data indices, up to 25× leverage, 23/7. Onshore: Architect acquired IMX Health, a CFTC Designated Contract Market, in May and plans the "American Innovation Exchange" — cash-settled compute futures and options at SKU-level index granularity, with an exchange-for-physical bridge to physical compute marketplaces. The EFP ambition is the differentiator: nobody else is trying to connect the paper curve to deliverable capacity.
• CME × Silicon Data (announced May 12, pending). Cash-settled futures on Silicon Data's quote-based daily assessments (SDH100RT and family). Silicon Data separately already publishes a 36-month term-rate curve and a derived no-arbitrage forward curve, built from observed 1–36-month rental contract rates — a Platts-style assessment stack, extended into the tenor dimension.
• ICE × Ornn (announced May 19, pending). USD cash-settled, cleared futures on OCPI — the index built only from printed transactions, volume-weighted, regionally weighted, with period-average settlement mechanics modeled on power monthlies (per our prior read; the averaging detail is not yet in published specs).

Status matters as much as design. Kalshi is the only venue where a compute forward price is executable today — which means that for however long the CFTC reviews take, the event-ladder market is the entire live term structure of the asset class. Everyone building a settlement nowcast, marking an OTC swap, or pre-positioning for the ETF wave is, for now, marking to Kalshi.

Two independent axes organize the methodology comparison, and collapsing them into one is the most common analytical error in the coverage so far.

Axis one: how the curve is constructed. Kalshi's curve is market-implied — traders bet on threshold events, and the forward falls out of the probability ladder, the way dealers read forwards off caps/floors in young rates markets. Silicon Data's curve is observed — it collects actual 1–36-month rental contract rates and derives the no-arbitrage forward from the term structure, the way swap curves get bootstrapped from deposits and FRAs. Architect's AX curve is funding-implied — a perpetual's basis and funding rate encode the carry, the way crypto desks back out forward curves from perp funding. ICE and CME futures, once listed, are discovered — outright price discovery at each tenor. Four construction technologies; each fails differently. Implied curves die when ladders go stale (see the B200 monthly below). Observed curves embed the lag and selection of whoever signs term contracts. Funding-implied curves whip with leverage demand. Discovered curves need liquidity that doesn't exist yet.

Axis two: what settles. This is the SD-vs-OCPI split from the Compute Complex piece, now with a scoreboard. Quote-based assessment (Silicon Data: listed and quoted rates, ~50 chipsets, hyperscaler tier included) versus executed prints (Ornn: transaction-only, volume-weighted, neocloud-sampled). In the March squeeze the quoted ceiling spiked past $8 while the printed floor sat near $2.85 — the two philosophies can disagree by more than the price of the underlying. Every structural conclusion from that analysis carries over; what's new is that the market has been voting, and it's voting Ornn — three venues, including the only live one, settle on Ornn-family prints.

The 2×2 that falls out: Kalshi is implied-construction on print-settlement. Architect is funding/discovered-construction on print-settlement. ICE is discovered-construction on print-settlement. CME is discovered-construction on quote-settlement — alone in its column. If the CME contract lists into that position, the cross-index basis (A5 in the monitor below) becomes the loneliest, and therefore possibly the richest, spread in the complex.

Prices, captured from Kalshi's public API today. The implied forward is the strike where the mid-quote CDF crosses 50%; confidence varies enormously across the four chips, and the variation is itself informative.

• H100 ≈ $2.52 for end-July. A clean, tight ladder. Note what this does to the stale "$1.70 H100" figure still circulating: the Ornn print already settled above $2.17 on May 31 (every strike through $2.17 paid YES), and the market now prices another ~14% on top by July 31. The squeeze that started in Q1 on Blackwell has spread to the mature grade.
• B200 ≥ $7.00 near-spot — and the venue disagrees with itself. The July 17 weekly, the deepest compute market on Kalshi, prices the Ornn B200 print at or above $7. The July 31 monthly ladder tops out at a $5.36 strike with asks untouched since July 2, bid-implying ~$5.2. That's a 26% collapse priced over 14 days, or — more plausibly — a stale, strike-capped ladder on a venue whose listing cadence hasn't caught up with a violent repricing. Either reading is tradeable (A1 below).
• H200 ≈ $3.1–4.3, and incoherent. The mids are non-monotonic in strike — the implied probability of settling above $3.15 exceeds the implied probability of settling above $3.05, which no valid distribution allows. The widths block the riskless version, so this is a market-making opportunity rather than free money (A2).
• A100 ≈ $1.02. The best-behaved ladder of the set, on the oldest chip — exactly what the vol-scales-with-grade-age result from the Q1 tape predicts. Two-generation-old hardware trades like deferred power; the young grade trades like prompt.

Against the other venues: Silicon Data's March SDH100RT print (~$2.55, neocloud tier) sits almost exactly on today's Kalshi-implied H100 — but the vintages differ by three months, and SD's current level is portal-gated, so treat the apparent convergence as suggestive, not settled. The structural gap is unchanged: SD's hyperscaler H100 index prints ~3× the neocloud tier on identical silicon, and no listed contract on any of the four venues hedges that tier.

The dashboard carries the piece from here. Four venues is the methodology matrix in interactive form. Ladder → curve renders the actual launch-day strike ladders and computes each implied forward, flagging the no-arb violations. Tenor model runs the decay-vs-scarcity fight out 36 months against the observable marks. Arb monitor is the eight-entry map from Section 4. Power link converts any deployment into zonal grid load and puts the compute spark spread against real forward curves.

The desk plan organized the complex into futures-native trades (F1–F6) and ETF-flow trades (E1–E8). Kalshi's launch re-sequences that map: some of what was "at listing" is executable today, and the settlement-index nowcast the plan budgeted infrastructure for is now partially published, for free, as market prices.

• Executable today (A1–A3). The intra-Kalshi B200 calendar — weekly at ≥$7.00 vs a monthly bid-implied at ~$5.3 — is the cheapest expression of "the squeeze persists two more weeks" anywhere. The H200 ladder's monotonicity violations are a standing invitation to market-make coherence into the implied CDF. And the grade ratio — B200/H100 at ~2.8× traded vs ~2.2× dense-FP8 parity — is the desk plan's F4 grade-spread trade, expressible in event space before any future lists.
• At listing (A4–A5). The Ornn index concentration creates something the May announcements didn't promise: a same-index, cross-venue complex. Kalshi ladders, ICE futures, and Architect perps settling on one index family means their residual spreads are pure microstructure — funding vs carry vs ladder discreteness — the lowest-basis-risk RV in the complex. Against that stands the true cross-index basis, Ornn-family vs CME × SD, the original F1 trade, now upgraded: the Ornn side has a live short-end mark printing daily.
• Structural (A6–A7). The tenor gap — Kalshi's executable curve ends at ~6 weeks; Silicon Data's observed curve runs 36 months; nothing reconciles them, and whoever stitches the two owns the middle of the term structure for OTC swaps and lease pricing. And the congealed-electricity spread: at $7 B200 and $65/MWh power, energy is ~2.3% of GPU-hr revenue — both legs of the fattest spark spread in any commodity market now quote.
• Flow (A8). The thirteen registered compute-futures ETFs remain pending (Roundhill's GPUX legend was still up this morning), keyed to the futures actually listing. Nothing about the flow book changes except its inputs improve: the mechanical cohort's roll and reweight trades price off the settlement index, and the settlement index's forward now has a public, market-implied nowcast.

Everything in the compute-power framework survives contact with the new curve, and one thing improves. The conversion stack is unchanged: chip TDP × count × PUE gives zonal grid load (a 100k-GPU B200 deployment is ~250 MW post-PUE; GB200, ~209 MW); load growth is visible in interconnection queues 24–36 months before commercial operation; and the zones hosting compute — Dominion, ERCOT North, AEP — carry forward curves that already price that growth. Northern Virginia's 16.8 GW pipeline against ~4 GW of existing inventory is still the single most legible supply curve in the complex, and it prints in PJM forwards, not in any compute index.

What's new: the compute spark spread now has two quoted legs. Before July 14, the power leg quoted (zonal forwards, decades of liquidity) and the compute leg was marked to assessments or private term sheets. Now the compute leg has an executable, if short-dated, market price. For a neocloud deciding between locking power and locking compute revenue — or a desk structuring the take-or-pay lease optionality the plan's F6 trade prices — the hedge-ratio math finally has two observable inputs. The desk plan dropped standalone power positions by design; that logic holds. The power book's role is informational: the queue is the supply model, and the supply model dates the resolution of every scarcity-premium trade in the tenor model.

Changed. The Ornn index family now has three committed venues and the only live market; the benchmark race has a leader before the starting gun. The settlement-index nowcast — the desk plan's Phase-0 infrastructure build — is partially commoditized: the market itself now publishes an implied forward on the Ornn print. Price discovery has begun at the short end, and it's saying things: H100 above every stale spot estimate, B200 holding a violent squeeze, the young grade rich to FLOP parity in exactly the pattern the Q1 tape taught.

Unchanged. The curve beyond six weeks is still model, not market. Event ladders are thin — tens to a few hundred dollars of notional per strike in places, one deep market (the B200 weekly, ~$20k OI at the $4 strike). A binary ladder settling on a point-in-time print is game-able in ways a volume-weighted monthly average is not; the settlement-integrity questions from the Compute Complex piece apply with more force here, not less. And the hyperscaler tier — the largest differential in the complex at ~3× — remains unhedgeable on every venue, including the new one.

The honest frame: Kalshi didn't build the compute forward curve. It built the short end of one candidate curve, on one index family, and in doing so handed every desk in the complex a free, live calibration point that didn't exist last week — plus a scoreboard that shows the settlement-index war leaning Ornn's way. The first trade of the compute complex is still the cross-index basis. It just acquired a second live leg.