Hopper

Also discussed in

• Dated test Chapter 12, claim 5, the first fleet reading, rent

Sources that use it

• NVIDIA, “NVIDIA Spectrum-X Makes Colossal NVIDIA Hopper 100,000-GPU System Possible,” press release, 28 October 2024 (“xAI chose to pair the Spectrum-X SN5600 switch with NVIDIA BlueField-3 ® SuperNICs”; “It took 19 days from the time the first rack rolled onto the floor until training began”). In the Sources
• NVIDIA, Form 10-Q for the quarter ended 26 July 2026, filed 26 August 2026 (China, including Hong Kong, revenue $7.880 billion, 8.19% of $96.221 billion total quarterly revenue; approximately $0.4 billion of H200-related inventory and purchase-obligation charges in the first half of fiscal 2027; NVIDIA said it had been unable to deliver a competitive China product approved by both governments; Note 10’s separate table of ‘the maximum gross exposure related to our guarantees’ — ‘Land, power, and shell guarantees for AI clouds’ $3.5 billion plus ‘SB Energy Corp. guarantees’ $105.0 billion, total $108.5 billion, the latter ‘to provide credit support on a land, power, and shell buildout with affiliates of SB Energy Corp. (SB Energy) on behalf of a customer, an affiliate of OpenAI Group PBC (OpenAI)’; Exhibit 10.1, the Form of Residual Value Guaranty, §20(e): “The parties agree that the Premises will exclusively install and deploy NVIDIA reference architecture for accelerated computing hardware and compute technologies … for a period of not less than twenty (20) years from the Commencement Date of the Lease” — an exclusivity covenant inside the guaranty; its consideration is a separate, redacted letter agreement, Recital F, and the landlord’s S-1 says the compensation “is provided separately by OpenAI under a separate agreement”; Note 10 on the SB Energy guarantees — “limited to defined portions of lease and power payments and not the full cost of the site or all of the tenant’s obligations”, terminating on events “including OpenAI achieving a satisfactory credit rating or after each respective lease term has completed”, and “In exchange for the guarantees, the site will exclusively host NVIDIA AI infrastructure, subject to limited exceptions”; Note 10, Commitments and Contingencies, on the supply agreements: “in certain instances, these agreements may be cancelable, rescheduled, or adjustable for our business needs prior to placing firm orders”; Note 10, Commitments and Contingencies: cloud service agreement commitments of $36 billion at 26 July 2026, typically six years, which the AI clouds “can unilaterally stop providing to us and sell to third-party customers at more advantageous rates”, set out as an ‘AI cloud agreements’ line in the Additional Commitments table by fiscal year — $6 billion in 2028, $8 billion in 2029, $7 billion in 2030, $6 billion in 2031, $9 billion in ‘2032 and thereafter’ — and classed as a commitment, the word ‘backstop’ appearing nowhere in the filing; one direct customer at 16% of second-quarter revenue; the Groq tail being paid down: a “Groq, Inc.” line of $2,944 million in cash used in financing activities for the first half of fiscal 2027, and “Accrued purchase consideration” of $986 million at 26 July 2026 against $3,921 million at 25 January 2026, footnoted “Related to the Groq, Inc. non-exclusive license agreement”; supply-and-capacity commitments of $279 billion: $92 billion for the rest of fiscal 2027, $87 billion in fiscal 2028, $88 billion in fiscal 2029; primarily memory, manufacturing facilities and data-center systems; supplier split not disclosed; the July 10-Q’s own comparison — “increasing supply commitments from $119 billion last quarter to $279 billion as of July 26, 2026. These supply commitments are for our data center infrastructure systems, primarily memory and manufacturing facilities” (Note 10, p. 18); the July 10-Q’s maturity table runs the $279 billion to “2032 and thereafter”: $92, $87 and $88 billion in fiscal 2027 to 2029 and $6, $5 and $1 billion in 2030, 2031 and 2032 and thereafter, so $12 billion falls after the three fiscal years the book’s table names; a risk factor on the guarantees: “Although OpenAI has agreed to reimburse and indemnify us for certain losses, we may not recover amounts promptly or in full”; the census inputs at 26 July 2026: debt at a net carrying amount of $33,366 million, Note 9; in Note 10, data center leases not commenced $25 billion, cloud service agreements $29 billion, equity investments $25 billion and capital expenditures $8 billion, and in its additional commitments $20 billion of data center leases not commenced for third parties, which NVIDIA expects to reassign; of the AI cloud agreements, in the management discussion: “Our commitments, which are typically six years in duration, totaled $36 billion as of July 26, 2026, and decrease as capacity is used by third-party customers or by us for our research and development efforts”; of the leases signed for others, in Note 10: “We have entered into data center leases with terms of approximately fifteen years that are expected to commence between fiscal year 2028 and fiscal year 2029” and “We expect to reassign these data center leases to third parties”; on China: “under the current rules and geopolitical landscape” it is “unable to create and deliver a competitive product for wide distribution in China’s data center market with the approval from both the USG and the Chinese government” (Part II, Item 1A, Risk Factors, p. 37); $25.0 billion of senior unsecured notes in seven tranches issued in June 2026, 4.25 to 5.625 percent, due 2028 to 2056; the notes’ net carrying amount $33.4 billion at 26 July 2026 against $8.5 billion at 25 January 2026; Note 10 on when the SB Energy guarantees run: “Each guarantee generally becomes effective upon commencement of the applicable lease, with corresponding guarantee amounts increasing, as each of the nine phases of data center construction is completed, the first of which is expected in fiscal year 2029”; the census’s cells at 26 July 2026, the sums the book’s: debt 33.4, the notes’ net carrying amount of $33,366 million (Note 9, Debt, p. 17); leases not yet begun 45.0, data center leases not commenced of $25 billion in Note 10’s commitments table plus data center leases not commenced for third party of $20 billion in its additional commitments table (p. 18), $45 billion; purchase and other commitments 377.0, supply and capacity $279 billion, cloud service agreements $29 billion, equity investments $25 billion and capital expenditures $8 billion in the commitments table, plus AI cloud agreements of $36 billion in the additional commitments table (Note 10, p. 18), $377 billion, which is the first table’s own total of $366 billion less its $25 billion of leases, plus the $36 billion; on the H200 licenses, “such sales were restricted by the PRC government, and we have been unable to sell all the products for which we have licenses”; “we have made a fraction of the allowed shipments under the USG’s H200 licensing program. Those shipments account for less than 1% of Data Center revenue in our most recent quarter”; “We have been unable to pass along any of the tariff to our customers”; in the management discussion, “Shipments of Data Center Hopper products to China during the second quarter of fiscal year 2027 were less than 1% of Data Center revenue”; “our effective foreclosure from the China market helped our competitors build larger developer and customer ecosystems to challenge us worldwide”; “The Chinese government has encouraged customers to purchase from our China-based competitors and discouraged customers from purchasing, importing, or using our data center products”). In the Sources
• Bloomberg Open Interest, Bloomberg Television and Radio, 31 August 2026, Ed Ludlow’s interview with Jensen Huang and Rick Tsai of MediaTek, on a three-way split screen with Tsai in a separate location, and, in the same program, Dani Burger and Sally Bakewell’s with Chris Peterson, chief executive of Newell Brands (Huang on NVIDIA’s economics per gigawatt by generation, at about 22:28, offered as an extrapolation by his own opening, “inside a gigawatt, if we were just to extrapolate”; “if we were just to extrapolate, in a gigawatt, Hopper was about $18 billion of economics for Nvidia per gigawatt. Grace Blackwell is about 25. And the Vera Rubin generation, it’s about 40-plus billion dollars per gigawatt of total Nvidia compute economics,” because “there’s like seven different types of Nvidia chips that has to go into an AI factory”, at 23:01; Huang on the $3.5 billion MediaTek convertible-bond investment, “not circular”; NVHBM custom memory extended to NVLink Fusion partners including Amazon; Peterson, in the same program: digital-content creation cost down 80 percent year on year against volume up 500 percent, “And in many cases we’re choosing to maintain that headcount and just do more to get the company back to stronger core sales growth” at 1:16:52–1:17:00 and “largely we’re using it as capability build as opposed to headcount reduction”, at 1:17:06, and “There are a few areas where we’ve made some reductions” at 1:17:01 as the caption and one of six local decode cells read it, the other five reading “few areas” and the loudness showing no voiced peak where the caption puts “a”, unsettled and kept as printed under the set’s standard; the Huang passage at 23:01 and the Peterson passage quoted from the recording and read against the auto caption track on 13 September 2026, Bloomberg’s own caption lossy at the Peterson sentence; the caption’s uh at 1:16:54, between to and maintain, at 1:17:03, between where and where, and at 1:17:06, between largely and we’re, carried by neither local decoding of the audio on 25 September 2026 and dropped from Peterson’s quotations, and its restart at 1:17:03, where we’re where on four of six decode cells, rendered once, read on the word-timed track on 24 September 2026 and on the three windows of 25 September 2026 at medium.en and large-v3 held in book 2’s Quote audio checks folder of that day with their md5 in its manifest, the speaker settled by the frames and the anchors by Bloomberg’s own caption labels; the Huang opening read on medium.en’s 20:30–24:30 slice held in book 1’s Quote audio checks folder of 5 September 2026 and on Bloomberg’s caption). In the Sources
• Chase Lochmiller (co-founder and chief executive, Crusoe), interviewed by Harry Stebbings, “Crusoe CEO: Why Everyone Gets GPU Depreciation & AI Energy Costs Wrong,” the publisher’s title, though everyone is the host’s word and the tape gives no cost of energy, 20VC, published 3 October 2026 (every figure is his own account of a private firm that files no accounts, and no contract or depreciation schedule is shown; at 35:24: “we are renting capacity on a long term basis”, “call it five year contracts to” “credit quality customers”; at 41:54: “we use a six year depreciation cycle” and “that’s sort of like the standard across the industry”; at 45:38: “earlier stage startups that are being asked to commit to compute in 2028” and “that is an eternity from now”; at 17:54: “we had sort of committed to doing this in one year and the next closest bid was two and a half years”; at 18:34, of a medium-voltage power distribution center: “the lead time for this one component was 100 weeks”, and by his account his own team then made it in 28 weeks; at 25:46 and 26:00: “use pretty close to zero water” and, of one of the first eight buildings at Abilene, “about the same water annually as about 10 single family homes”, with no volume given; at 27:01 and 27:18: “typically energy prices for communities have come down” and “more megawatts being amortized over the same transmission and distribution infrastructure”, with no market or study named; at 28:52 and 29:07: “we’re going to account for more than a third of the tax revenue in Taylor County and Abilene, the town” and “we’re more than doubling the tax receipts that are going to the schools”, with no abatement mentioned on the tape; at 43:14: “will ultimately, we believe, extend the depreciation cycle beyond six years”; at 43:40 and 43:54, of the prices charged for Hopper chips and with no price given: “higher than the rates that were being charged three years ago” and “payback very, very quickly”; decoded at two models). In the Sources
• Jensen Huang, GTC Washington, D.C. keynote, NVIDIA, 28 October 2025 (caption track, 1:43:13; at 1:01:32 “visibility into half a trillion dollars of cumulative Blackwell and early ramps of Rubin through 2026”, at 1:02:26 “This doesn’t include China and … Asia. So this is just … the West” and 20 million Blackwell GPUs against Hopper’s 4 million; the ellipses in the China sentence standing for the caption’s “and um and” and “uh”; both read against two decodings of the audio, medium.en and large-v3, of 25 September 2026, two slices held in the set’s Quote audio checks folder of that day with their md5 in its manifest: Asia on the caption and medium.en where large-v3 reads an agent, kept as printed and unsettled; the half trillion is “visibility” and “business … on the books … so far” in his words, where the slide behind him reads “more than half a trillion dollars in cumulative Blackwell and Rubin revenue through CY2026”; “This doesn’t include China” is said at the chart’s Hopper bar and the slide footnotes the Blackwell–Rubin bar “Excluding China”, so which figure this names the frames cannot settle; the 20 million counted as he counts them, “each one of the Blackwells has two GPUs in it in one large package” at 1:02:41, so about two and a half times Hopper’s 4 million in packages, Blackwells on the caption and medium.en where large-v3 writes black walls; the same half trillion named forward at 1:02:12, “the next five quarters, there’s $500 billion”, where the slide’s cumulative run includes 2025, the base of the figure open; at 46:25, “Nvidia pays for every license of Cursor. And we gladly do it. We gladly do it because Cursor is helping a several hundred thousand dollar employee software engineer or AI researcher be many, many times more productive”, dollar on both decoders at two widths, 46:10–47:25 and 46:36–46:42, held in the set’s Quote audio checks folder of 25 September 2026, where both caption layers drop it; at about 47:00, in the slice 46:10 to 47:25, read on two decodings of the audio, medium.en and large-v3, of 25 September 2026, “we now have two exponentials” and “one is the exponential compute requirement of the three scaling law”). In the Sources

And 6 more: search the Sources.

Related terms

• H100
• Blackwell
• H20
• H200