B200

Where the book uses it

• Chapter 14 — The Convergence

  One campus, Meta’s Hyperion, run through four possible futures to show how science, power, money and permission meet in the price of an accelerator-hour.

Sources that use it

• Stephen Witt, The Thinking Machine: Jensen Huang, Nvidia, and the World’s Most Coveted Microchip (Viking, 2025; on the Nobu dinner, p. 194: “In late 2023, Huang met with Elon Musk and Oracle cofounder Larry Ellison for sushi at the Nobu restaurant in Palo Alto. Ellison and Musk had a collective net worth above $300 billion”) (Larry Ellison on a meal with Musk and Huang at the Nobu restaurant in Palo Alto in late 2023: “Elon and I were begging, I guess is the best way to describe it. An hour of sushi and begging”; a Caltech data-center administrator’s H100 delivery wait of “almost eighteen months” in early 2024, the professors unwilling to switch, “They’d rather wait for the hardware than switch away from CUDA”, ch. 16, p. 185; Witt’s per-chip power ladder for chips “in standard configurations”, p. 207 — A100 250 W, H100 350 W, B100 700 W, B200 1,000 W; the 2016 DGX box drew as much power as a clothes dryer and its 2024 successor enough for a single-family home, p. 207, the boxes “modular units that customers stacked into the long ‘superpods’” of the data center; a former US military official on seizing TSMC’s fabs — “These machines are insanely precise, and insanely fragile, down to a precision of one atom. What do you think’s going to happen to that machine if a missile goes off next to it?”, remote disabling, “I would say the Taiwanese engineer sabotage thing is not to be underestimated”; Deb Shoquist, who ran NVIDIA’s logistics, on what would happen in a war — orders transferred to Samsung and other partners, and “What would happen is everybody’s products would get dumbed down a click”; Morris Chang telling Nancy Pelosi in 2022 the US fab projects were “doomed to fail” and calling TSMC’s concentration Taiwan’s “silicon shield”, and his argument that “no national economy would suffer as much as China’s if TSMC went offline”, p. 218; “the problem that investors had once faced valuing Nvidia alone now applied to valuing the entire stock market”, p. 209, the last page of ch. 19; ch. 20, pp. 210–212: “Huang now had fifty-five” direct reports and “no designated successor”, p. 211, after the earnings of 21–22 February 2024; “a giant, GDP-sized bet on the capabilities of this single sixty-one-year-old man”, p. 212; on NVIDIA’s planning for Taiwan: Shoquist, “If something happens to Taiwan and TSMC, the ramifications are so large it’s almost like asking me what I’d do if California fell into the ocean”, and Huang, “I don’t want her spending one brain cell on trying to mitigate that, because it’s impossible for her to do so” (p. 217); on Morris Chang to TSMC employees at a TSMC function in late 2023: “in the semiconductor space, there is no globalization anymore. The priority is national security only” (p. 217, two quoted sentences either side of Witt’s attribution)). In the Sources
• CME Group and Silicon Data, “CME Group and Silicon Data to Launch Compute Futures on October 5 to Unlock New Way to Hedge AI Risks,” press release, 11 August 2026 (NYMEX-listed contracts, pending regulatory review; each a month’s rent on an Nvidia H100 or Blackwell B200, tracking Silicon Data’s GPU rental-price indexes; Keavey (“compute has become the currency of the AI age”) and Li (“a public, tradable reference price for the resource every AI system runs on”) quotes) In the Sources
• CME Group, Special Executive Report SER-9785R, “UPDATE: Initial Listing of Two (2) Compute Futures Contracts - Silicon Data H100 Rental Index Futures and Silicon Data B200 Rental Index Futures,” 25 September 2026 (no initial listing date, the specification table giving it as to be determined; “Pending all relevant CFTC regulatory review periods”; “The Exchange will further advise regarding the specific effective date”; 730 GPU-hours a contract, quoted in U.S. dollars and cents per GPU-hour, financially settled). In the Sources
• ICE Futures U.S., “New Compute Index Reference Data to Be Made Available on Monday, October 5, 2026,” exchange notice, 29 September 2026 (the first contracts set for listing, “OCPI H100 Futures (contract symbol HPR)” and “OCPI B200 Futures (contract symbol BKL)”; to be listed “pending completion of the request for comment period on compute futures by the Commodity Futures Trading Commission (CFTC).”; “While no listing date or timeline has been set at this time, the Exchange will continue to work cooperatively with the CFTC.”; from 5 October the index’s data on the exchange’s platform, with “hypothetical daily settlement prices”; the notice read as the exchange posts it, 23 October 2026) In the Sources
• Intercontinental Exchange and Ornn, “ICE and Ornn to Launch GPU Compute Futures Contracts,” press release via Business Wire, 19 May 2026 (plans for “a suite of GPU compute futures contracts based on Ornn’s Compute Price Index (OCPI)”, U.S. dollar denominated and cash-settled; “Contracts may reference the OCPI series covering H100, H200, B200, RTX 5090, and additional GPU types as the market develops”; “The contracts will be launched subject to regulatory approval.”). In the Sources

And 16 more: search the Sources.

Related terms

• Compute futures
• Accelerator
• Blackwell
• H100
• Perpetual
• Accelerator-hour
• Break fee
• Break-point
• Bridge power
• Capital recovery
• Clearing house
• Commodity infrastructure