AI Wiki

SemiAnalysis

AI CompaniesAI HardwareAI Infrastructure
20 min read
Updated
Suggest editHistoryTalk47 citationsFact-checked
RawGraph

SemiAnalysis is an independent research and analysis firm covering the semiconductor and artificial intelligence industries. It describes itself as "an independent research and analysis company specializing in the Semiconductor and AI industries" and works under the tagline "Bridging the Gap Between business and the world's most important industry" [1]. Founded in 2020 by Dylan Patel as a one-person newsletter, it now sells subscription research, industry forecasting models, consulting, and benchmarking data to chipmakers, cloud operators, hyperscalers, and investors [2][3]. Much of what it publishes rests on supply-chain sourcing and independent measurement rather than company disclosure, and the firms it writes about frequently decline to confirm its findings or dispute them outright.

Origins and growth

SemiAnalysis dates its founding to 2020, when Dylan Patel began publishing semiconductor analysis on his own. The company's profile page for Patel says that "Since 2020, SemiAnalysis has transformed its business from a solo venture into a cohesive and focused team" [2]. The publication ran for years primarily as a Substack newsletter and still distributes through that platform, alongside its own site at semianalysis.com.

By 2026 the company described itself as "a global team of analysts and engineers, publishing more than 100 articles a year and participating in 40+ conferences" [3]. Its careers page lists roles based in San Francisco, Chicago, New York City, Hillsboro (Oregon), and Singapore, with additional remote positions in Korea and internationally, spanning research, consulting, sales, legal, business operations, hardware engineering, and lab technician work [3]. The Hillsboro postings for a hardware engineer and a lab equipment and facilities technician indicate the firm operates its own physical test hardware rather than relying only on rented capacity.

Published subscriber figures are inconsistent across SemiAnalysis's own properties. Its about page carries both "Over 180,000+ Subscribers" and a reference to "200,000+ AI and Semiconductor industry professionals" [1]; the careers page says "over 200k subscribers" and calls the newsletter "the #1 technology publication on Substack" [3]. As of 1 August 2026 the SemiAnalysis Substack profile showed "306K+ subscribers" and a "#1 in Technology" leaderboard badge [4]. The company has not explained the gap.

Patel said in a February 2026 interview that "we're 60 people now" and that the firm was "in eight or 10 different countries" [45]. Business Insider reported in August 2024 that SemiAnalysis had grown "from a team of three analysts in two countries to a team of 12 subject matter experts in the US, Japan, Singapore, Taiwan, and France" [46]. Business databases and recruiting aggregators give unsourced figures ranging from roughly 70 to over 100, which this article does not adopt.

What SemiAnalysis publishes, and how much of it is confirmed

The output splits into three broad categories: free posts on the public newsletter, paid deep dives on architecture, supply chain, and economics, and institutional data products sold separately from the newsletter subscription [5].

The distinction between what SemiAnalysis reports and what is established fact matters for reading almost any of its major work. Its highest-profile pieces are built from anonymous supply-chain contacts, permit and property records, satellite imagery, teardown-adjacent inference, and its own benchmarking. Subjects rarely confirm the resulting numbers. The Center for Strategic and International Studies, citing the firm's January 2025 DeepSeek work, described it as "the semiconductor consulting firm SemiAnalysis, citing anonymous industry sources" and reproduced its GPU-count estimates as claims rather than facts [6]. That framing is the appropriate one for most SemiAnalysis figures repeated elsewhere: they are estimates from a paid research firm, sometimes later corroborated by events, sometimes contradicted by the companies involved.

Notable reports

ReportDateBylined authorsCore claim
Google "We Have No Moat, And Neither Does OpenAI"4 May 2023Dylan Patel, Afzal AhmadRepublication of a leaked internal Google document arguing open-source AI would outpace both Google and OpenAI [7]
GPT-4 Architecture, Infrastructure, Training Dataset, Costs, Vision, MoE10 July 2023Dylan Patel, Gerald WongGPT-4 is a sparse mixture-of-experts model of roughly 1.8 trillion parameters [8][9]
AI Datacenter Energy Dilemma13 March 2024Dylan Patel, Daniel Nishball, Jeremie Eliahou OntiverosGlobal datacenter critical IT power demand rising from 49 GW in 2023 to 96 GW by 2026, about 40 GW of it AI [10]
100,000 H100 Clusters: Power, Network Topology, Ethernet vs InfiniBand, Reliability, Failures, Checkpointing17 June 2024Dylan Patel, Daniel NishballA 100,000-GPU training cluster needs more than 150 MW and over $4 billion in server capex [11]
AI Neocloud Playbook and Anatomy3 October 2024Dylan Patel, Daniel NishballBill of materials, deployment, and unit economics of GPU rental providers [12][13]
MI300X vs H100 vs H200 Benchmark Part 1: Training, CUDA Moat Still Alive22 December 2024Dylan Patel, Daniel Nishball, Reyk KnuhtsenFive months of testing found AMD software defects severe enough to block practical training use [14][15]
DeepSeek Debates31 January 2025Dylan Patel and othersDeepSeek has access to roughly 50,000 Hopper-generation GPUs and about $1.6 billion in server capex [6][16][17]
Huawei AI CloudMatrix 384: China's Answer to Nvidia GB200 NVL7216 April 2025Dylan Patel, Daniel Nishball, Myron Xie, Patrick Zhou, Ivan Chiam, AJ Kourabi, Christopher Seifel, Doug O'LaughlinHuawei's CloudMatrix 384 rack system beats NVIDIA's GB200 NVL72 on some system-level metrics despite weaker individual chips [18]

The GPT-4 report is the single most-repeated piece of SemiAnalysis work. Coverage at the time summarized its claims as roughly 1.8 trillion parameters across 120 layers, 16 experts of about 111 billion parameters each with two routed per forward pass, around 13 trillion training tokens, and a training cost near $63 million [9]. OpenAI has never confirmed any of it. The contemporaneous write-up noted plainly: "No statement from OpenAI, but the rumors are credible" [9]. The 1.8 trillion figure has since been repeated widely as though settled, which it is not.

The May 2023 "We Have No Moat" post is unusual in the firm's catalogue because SemiAnalysis wrote almost none of it. Its introduction stated: "The text below is a very recent leaked document, which was shared by an anonymous individual on a public Discord server who has granted permission for its republication. It originates from a researcher within Google. We have verified its authenticity." The editors added: "We do not agree with what is written below, nor do other researchers we asked" [7].

The 2024 neocloud work put a name to a market segment. Trade coverage of the October 2024 report described it as a look at "AI Neoclouds", a new class of GPU rental providers including CoreWeave, Crusoe, and Lambda Labs [13]. The term is now standard industry vocabulary.

Research products and tools

Institutional products sit behind a separate login from the newsletter and are not included in a paid Substack subscription [5]. The catalogue as listed in mid-2026 included the following forecasting models and interactive tools, alongside an Industrials Model, a VR NVL72 BoM and Power Model, and a Taiwan Monthly Dashboard [19]:

ProductTypeWhat it covers
Accelerator & HBM ModelModelShipments, ASPs, and revenue by SKU, company, and customer for AI accelerators and high bandwidth memory
AI Cloud TCO ModelModelTotal cost of ownership for building and pricing an AI cloud
AI Networking ModelModelSwitches, optics, cables, and AEC/DAC interconnect
ChipBookTrackerMonthly AI and semiconductor data tracker with dashboards
Core ResearchSubscriptionWeekly and ad hoc institutional research stream
Datacenter Industry ModelModelCritical IT power, capacity, and constraints across the buildout
Energy ModelModelPower supply inputs to the datacenter buildout
Foundry Industry ModelModelWafer shipments, ASPs, capacity, and utilization
Memory ModelModelDRAM, NAND, and HBM supply and demand
Space DC TCO ModelModelCost of orbital datacenter concepts
Tokenomics ModelModelDemand-side framework tying hardware inputs to inference output
Wafer Fab Equipment ModelModelBottom-up equipment demand from node roadmaps
Die Yield Calculator, Fab MapToolYield estimation and global fab locations
GPU Pricing Index and APIData feedRental pricing for GPU capacity
Inference SimulatorToolModeled inference performance

The Datacenter Industry Model illustrates the firm's method. SemiAnalysis says it tracks "over 5,000 datacenters and their deployments with publicly available information" including "property records, permits, power usage, FOIA requests, and satellite images", and that it uses computer vision models to process satellite imagery and track construction size, capacity, and timelines. The same model tracks accelerator deployment plans for more than 50 companies, and the firm names its clients as "hyperscalers and major semiconductor companies for competitive analysis and supply chain planning, as well as large investors in public and private markets" [20].

ClusterMAX

ClusterMAX is a rating system for GPU cloud providers, introduced in March 2025 and updated in November 2025 and April 2026 [21]. Providers are scored across ten categories (security, lifecycle, orchestration, storage, networking, reliability, monitoring, pricing, partnerships, and availability) and placed in one of six rating states: Platinum, Gold, Silver and Bronze, plus a Not Recommended band split into Underperforming and Unavailable. Bronze is described as "the last tier we directly recommend" [21]. Evaluation combines hands-on deployment of real workloads, documentation review, and customer interviews.

The November 2025 ClusterMAX 2.0 round rated 84 providers out of 209 tracked, up from 26 in version 1.0. CoreWeave was the only Platinum-rated provider in both rounds; the Gold tier included Nebius, Oracle, Azure, Crusoe, and Fluidstack [22]. The April 2026 ClusterMAX 2.1 update added new entrants without publishing a provider count, with CoreWeave still the sole Platinum and the same five Gold providers [21].

InferenceX

InferenceX, launched in October 2025 under the name InferenceMAX, is an open-source continuous inference benchmark [23][24]. Rather than publishing a fixed result set, it re-runs benchmark suites on live hardware through public GitHub Actions workflows, so each dashboard data point traces back to a reproducible run. Coverage spans NVIDIA H100, H200, B200, B300, GB200, GB300, and RTX 6000 Pro parts alongside AMD MI300X, MI325X, and MI355X, and reports throughput, performance per dollar, and tokens per megawatt [23]. It is a rare case of a SemiAnalysis product whose underlying evidence is fully public, which distinguishes it from the sourced reporting that carries most of the firm's influence.

Business model

Revenue comes from at least five distinct lines.

The free newsletter is the top of the funnel. A paid Substack subscription unlocks "every newsletter article available on the SemiAnalysis website" and the "conclusions and further insights in the deep dives", but explicitly excludes "institutional content such as models" [5]. SemiAnalysis offers a 20% discount for groups and states that "corporate subscriptions are available at a greater discount", with payments processed through Stripe. Individual list prices are not published on the FAQ page [5].

Institutional data products and models are sold separately to enterprise clients, with pricing by quote [19][20]. Consulting is a distinct line; the careers page lists dedicated consultant and technical consultant roles serving semiconductor and AI clients [3].

Events form a fourth line. SemiAnalysis hosts, co-hosts, partners on, and attends industry conferences, including sessions it ran in Paris in July 2026 and a hackathon it co-hosted with Fluidstack in San Jose in March 2026 [25].

The fifth is venture investing, and it is the newest. On 29 July 2026, SemiAnalysis Capital Fund I, LP filed a Form D with the U.S. Securities and Exchange Commission for a $400,000,000 offering, with none sold as of filing. The fund is a Delaware limited partnership classified as a venture capital fund, and Dylan Patel signed the filing as manager of the general partner [26]. The Information had reported earlier in 2026 that SemiAnalysis was weighing a venture fund and then that Patel was targeting $400 million [27].

On overall scale, The Information reported in April 2026, in a briefing by Abram Brown, that SemiAnalysis projected more than $100 million in 2026 revenue from subscriptions and AI supply-chain research [28][29]. That figure is a company projection relayed by one publication, not an audited result.

People

The company does not publish a full staff roster, but individual analysts have bylined author pages on semianalysis.com. Verified roles include:

NameRoleCoverage
Dylan PatelFounder, CEO, and Chief AnalystSemiconductor supply chain, cloud, machine learning models [2]
Doug O'LaughlinPresidentSemiconductor strategy, market intelligence, competitive analysis [30]
Dan NishballDirector of ResearchNetworking and optical, AI systems, neocloud systems and economics [31]
Jeremie Eliahou OntiverosManaging Director, Datacenter, Energy, IndustrialsHyperscaler and neocloud infrastructure, cooling and electrical systems, power sources [32]
Ivan ChiamAnalystEconomics of AI accelerators, HBM, and GPU infrastructure [33]
Wega (Wei-Chia) ChuAnalystSemiconductor supply chain, sourcing, OSAT dynamics, Asian fab operations [34]

Other names recur on bylines without published role titles, including Afzal Ahmad, Gerald Wong, Myron Xie, Patrick Zhou, AJ Kourabi, Christopher Seifel, Jordan Nanos, and Michelle Shen [7][8][18][22].

Reception and influence

Companies SemiAnalysis rates have begun treating its assessments as credentials in regulated disclosures. CoreWeave's April 2026 investor presentation, filed with the SEC as an exhibit to a Form 8-K, cites "SemiAnalysis ClusterMAX rating" alongside MLPerf results and NVIDIA Exemplar Cloud status as evidence of its platform's standing, and dedicates a slide to being the "Only Player to Achieve Top-Ranked Platinum Status in Each of SemiAnalysis' GPU Cloud ClusterMAX Ratings" [35]. A private research firm's tier label appearing in an SEC-filed deck is an unusual degree of adoption.

Vendors also promote the firm's benchmark results. NVIDIA published a developer blog in October 2025 on its results in InferenceMAX v1, describing the exercise as an "independent third-party evaluation from SemiAnalysis" [24].

The December 2024 AMD benchmark report produced a documented change in a chip vendor's engineering practice. SemiAnalysis reported that AMD's software stack made MI300X impractical for training relative to NVIDIA H100 and H200 systems, writing that "AMD's Out of the Box Experience is very difficult to work with and can require considerable patience and elbow grease" and recommending bluntly: "Make the out-of-the-box experience usable!" [14][15]. AMD CEO Lisa Su met Patel for about 90 minutes the following day and, per contemporaneous reporting, acknowledged the gaps in AMD's software stack and took the recommendations seriously [15]. In an April 2025 follow-up, SemiAnalysis wrote that "before SemiAnalysis published the AMD article in December, AMD had zero MI300X participating in PyTorch CI/CD" and that "AMD has since added MI300 into PyTorch CI/CD", while noting that dynamic-shapes torch.compile pass rates still trailed NVIDIA [43]. That account of its own impact is the firm's, not AMD's. ROCm remains the point of comparison in most subsequent coverage of the CUDA software advantage.

Its work is cited in policy analysis of AI chip export controls and Chinese compute capacity. The CSIS analysis of DeepSeek by Gregory C. Allen, published in April 2025, reproduces SemiAnalysis's GPU-inventory and capex estimates in its assessment of Chinese AI capability [6]. Its reporting on Huawei Ascend production, HBM bottlenecks, and CloudMatrix systems has been a recurring input to the export-control debate [18].

Market reaction to its publications is measurable. After the July 2026 report that NVIDIA's Kyber NVL144 rack had slipped more than twelve months to 2028 because of a 78-layer PCB midplane that could not be produced reliably [47], Asian PCB suppliers fell sharply on the day: Ibiden dropped as much as 10%, Kingboard Laminates 18%, Elite Material 10%, and Samsung Electro-Mechanics 11% [36].

Criticism and disputes

Most of the firm's substantive analysis sits behind a paywall, and its sourcing is anonymous, so readers generally cannot check a claim against the underlying evidence [5]. The disputes below are the cases where the disagreement became public.

Denials by NVIDIA

NVIDIA has publicly contradicted SemiAnalysis at least twice. In August 2024, after reports of Blackwell delays that SemiAnalysis attributed to TSMC CoWoS-L packaging problems including bridge-die redesign and substrate warping, an NVIDIA spokesperson said: "As we've stated before, Hopper demand is very strong, broad Blackwell sampling has started, and production is on track to ramp in 2H. Beyond that, we don't comment on rumors" [37]. In July 2026, responding to the Kyber report, an NVIDIA spokesperson told Yahoo Finance: "Our roadmap remains intact" [38]. The firm's Rubin Ultra and rack-scale roadmap claims therefore stand as contested reporting, not confirmed fact.

The IBM ClusterMAX dispute

SemiAnalysis stated publicly that IBM's analyst relations team said it "would only be willing to participate if guaranteed a silver rating or higher", that IBM staff called SemiAnalysis engineers on personal phones to question their usage, blocked payments, and shut off the company's IBM Cloud account, and that "IBM actively worked against us during our testing period" [39]. IBM Cloud is rated Bronze on the current ClusterMAX roster [21]. IBM has not publicly responded to the allegations. The episode drew criticism in both directions on the SemiWiki forum in November 2025, where the thread starter argued that benchmarking a provider's service after being denied permission could expose SemiAnalysis to legal risk, while other participants argued the underlying problem was a vendor demanding a guaranteed grade [40].

Conflict-of-interest concerns

Jon Stevens, CEO of the GPU cloud provider Hot Aisle, published a critique in December 2025 arguing that the combination of critical research and paid consulting creates a pay-to-play incentive, pointing to the AMD sequence as his central example. Stevens presented no contracts or payment evidence and acknowledged the limits of the argument, writing that "from the outside, it's often impossible to know whether that happens in specific cases, but the overall incentives are troubling" [41]. Stevens also raised attribution disputes and the firm's handling of a compromised social media account that was used to promote a cryptocurrency scam, and argued that Patel's living arrangements created an appearance-of-influence problem. Hot Aisle is an AMD-focused provider and Stevens is not a disinterested observer.

Litigation with a former employee

In March 2026, SemiAnalysis and former employee Wei Zhou filed competing suits in San Francisco County Superior Court. Zhou, who according to reporting built the firm's Tokenomics product, alleges he was fired in retaliation for refusing to incorporate material non-public information into models sold to clients, and claims that SemiAnalysis President Doug O'Laughlin neglected compliance duties while raising money on company time for a separate $50 million investment vehicle managed by Patel. His causes of action include breach of contract, California Labor Code retaliation, and wrongful termination in violation of public policy. SemiAnalysis filed first, on 27 March 2026, pre-empting a suit Zhou had told it he intended to bring. Its complaint (SemiAnalysis LLC v. Wei Zhou, case CGC-26-635328) says Zhou was terminated for workplace misconduct, characterizes his allegations as "an attempt to extort money", and pleads nine causes of action: declaratory relief, trade secret misappropriation, conversion, breach of contract, breach of fiduciary duty, intentional interference with prospective economic advantage, intentional interference with contractual relations, defamation, and unfair competition [44]. Patel said: "SemiAnalysis unequivocally denies these claims, and we will defend against them vigorously" [42]. None of the allegations on either side have been tested in court.

Investment positions and disclosure

Patel has invested personally and through special purpose vehicles in AI infrastructure companies, including a reported $50 million SPV tied to Fluidstack, a company that SemiAnalysis rates in ClusterMAX and has co-hosted an event with [22][25][27]. The July 2026 Form D formalizes the investing arm as SemiAnalysis Capital Fund I [26]. Traditional financial research firms typically carry standing disclosure language on positions held; SemiAnalysis's published articles generally do not.

References

  1. ^SemiAnalysis, "About." semianalysis.com/about (accessed 1 August 2026)
  2. ^SemiAnalysis, "Dylan Patel: Founder, CEO, and Chief Analyst, SemiAnalysis." semianalysis.com/dylan-patel
  3. ^SemiAnalysis, "Explore Career Opportunities at SemiAnalysis." semianalysis.com/semianalysis-careers
  4. ^Substack, SemiAnalysis publication profile. substack.com/@semianalysis (accessed 1 August 2026)
  5. ^SemiAnalysis, "FAQ." semianalysis.com/faq
  6. ^Gregory C. Allen, "DeepSeek: A Deep Dive," Center for Strategic and International Studies, 8 April 2025. csis.org/...deepseek-deep-dive
  7. ^Dylan Patel and Afzal Ahmad, "Google 'We Have No Moat, And Neither Does OpenAI'," SemiAnalysis, 4 May 2023. newsletter.semianalysis.com/...no-moat-and-neither
  8. ^Dylan Patel and Gerald Wong, "GPT-4 Architecture, Infrastructure, Training Dataset, Costs, Vision, MoE," SemiAnalysis, 10 July 2023. newsletter.semianalysis.com/...ture-infrastructure
  9. ^The Decoder, "GPT-4 architecture, datasets, costs and more leaked," July 2023. the-decoder.com/...-datasets-costs-and-more-leaked
  10. ^Dylan Patel, Daniel Nishball and Jeremie Eliahou Ontiveros, "AI Datacenter Energy Dilemma: Race for AI Datacenter Space," SemiAnalysis, 13 March 2024. newsletter.semianalysis.com/...energy-dilemma-race
  11. ^Dylan Patel and Daniel Nishball, "100,000 H100 Clusters: Power, Network Topology, Ethernet vs InfiniBand, Reliability, Failures, Checkpointing," SemiAnalysis, 17 June 2024. newsletter.semianalysis.com/...sters-power-network
  12. ^SemiAnalysis, "AI Neocloud Playbook and Anatomy," 3 October 2024. semianalysis.com/...ai-neocloud-playbook-and-anatomy
  13. ^Techmeme, item on the AI Neocloud Playbook, 5 October 2024. techmeme.com/...p12
  14. ^SemiAnalysis, "MI300X vs H100 vs H200 Benchmark Part 1: Training, CUDA Moat Still Alive," 22 December 2024. newsletter.semianalysis.com/...ark-part-1-training
  15. ^The Decoder, "AMD's software woes leave Nvidia unchallenged in AI chip market, study finds," 23 December 2024. the-decoder.com/...d-in-ai-chip-market-study-finds
  16. ^Techmeme, item on the SemiAnalysis DeepSeek analysis, 31 January 2025. techmeme.com/...p9
  17. ^Tom's Hardware, "DeepSeek might not be as disruptive as claimed, firm reportedly has 50,000 Nvidia GPUs and spent $1.6 billion on buildouts." tomshardware.com/...nt-usd1-6-billion-on-buildouts
  18. ^SemiAnalysis, "Huawei AI CloudMatrix 384: China's Answer to Nvidia GB200 NVL72," 16 April 2025. semianalysis.com/...s-answer-to-nvidia-gb200-nvl72
  19. ^SemiAnalysis, "Models & Research." semianalysis.com/models-research
  20. ^SemiAnalysis, "Datacenter Industry Model." semianalysis.com/datacenter-industry-model
  21. ^ClusterMAX by SemiAnalysis, "ClusterMAX Overview: GPU Cloud Rating Methodology." clustermax.ai/overview
  22. ^SemiAnalysis, "ClusterMAX 2.0: The Industry Standard GPU Cloud Rating System," 6 November 2025. newsletter.semianalysis.com/...e-industry-standard
  23. ^InferenceX by SemiAnalysis, "About." inferencex.semianalysis.com/about
  24. ^NVIDIA Developer Blog, "NVIDIA Blackwell Leads on New SemiAnalysis InferenceMAX v1 Benchmarks," 13 October 2025. developer.nvidia.com/...is-inferencemax-benchmarks
  25. ^SemiAnalysis, "Events." semianalysis.com/semianalysis-events
  26. ^U.S. Securities and Exchange Commission, Form D, SemiAnalysis Capital Fund I, LP (CIK 0002145695), filed 29 July 2026. sec.gov/...primary_doc.xml
  27. ^The Information, "Exclusive: SemiAnalysis' Dylan Patel Targets $400 Million Venture Capital Fund." theinformation.com/...million-venture-capital-fund
  28. ^Abram Brown, "Dylan Patel's SemiAnalysis Projects Over $100 Million In 2026 Revenue," The Information. theinformation.com/...cts-100-million-2026-revenue
  29. ^Techmeme, item on The Information's SemiAnalysis revenue report, 19 April 2026. techmeme.com/...p2
  30. ^SemiAnalysis, "Doug O'Laughlin: President of SemiAnalysis." semianalysis.com/doug-olaughlin
  31. ^SemiAnalysis, "Dan Nishball: Director of Research." semianalysis.com/...danielnishball
  32. ^SemiAnalysis, "Jeremie Eliahou Ontiveros." semianalysis.com/...jeremieeliahouontiveros
  33. ^SemiAnalysis, "Ivan Chiam." semianalysis.com/...ivan
  34. ^SemiAnalysis, "Wega (Wei-Chia) Chu." semianalysis.com/...wegachu
  35. ^CoreWeave, Inc., Form 8-K Exhibit 99.3 (investor presentation), filed with the SEC 9 April 2026. sec.gov/...d136714dex993
  36. ^The Decoder, "Nvidia's Kyber NVL144 reportedly pushed back more than a year, Asian suppliers drop," 6 July 2026. the-decoder.com/...han-a-year-asian-suppliers-drop
  37. ^Dan Robinson, "Nvidia reportedly delays Blackwell GPUs," The Register, 5 August 2024. theregister.com/...dia_delays_blackwell_gpus_until
  38. ^Yahoo Finance, "Nvidia denies report its next-generation AI server faces delays, says roadmap is intact," July 2026. finance.yahoo.com/...s-roadmap-is-intact-183310296
  39. ^SemiAnalysis (@SemiAnalysis_), thread on IBM Cloud and ClusterMAX testing, November 2025. x.com/...1991923408913957196
  40. ^SemiWiki forum, "SemiAnalysis Gross Mafia Tactics?", thread started 21 November 2025. semiwiki.com/...analysis-gross-mafia-tactics.24073
  41. ^Jon Stevens, "Influence as a Service: SemiAnalysis Under the Microscope," 2 December 2025. jon4hotaisle.substack.com/...-service-semianalysis
  42. ^Ian Cutress, "Dylan Patel's SemiAnalysis Is Being Sued," More Than Moore, 4 April 2026. morethanmoore.substack.com/...emianalysis-is-being
  43. ^SemiAnalysis, "AMD 2.0: New Sense of Urgency, MI450X Chance to Beat Nvidia, Nvidia's New Moat," 23 April 2025. newsletter.semianalysis.com/...ia-nvidias-new-moat
  44. ^Superior Court of California, County of San Francisco, SEMIANALYSIS LLC VS. WEI ZHOU, case no. CGC-26-635328, filed 27 March 2026. docketalarm.com/...10130362
  45. ^Latent Space, "Dylan Patel: What's Cooking at SemiAnalysis," 28 February 2026. latent.space/...dylanpatel-cooking
  46. ^Emma Cosgrove, "How analyst Dylan Patel makes connections from Nvidia to the semiconductor industry and the tech it enables," Business Insider, 14 August 2024. businessinsider.com/...ia-chips-ai-research-2024-8
  47. ^Tom's Hardware, "Nvidia's Kyber rack for Rubin Ultra slips to 2028." tomshardware.com/...-for-rubin-ultra-slips-to-2028

Improve this article

Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.

1 revision · v2 · 4,016 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Company figures, product catalogue, SEC Form D filing, ClusterMAX version history, staff titles and all quotations confirmed against primary sources on 2026-08-01.

Cite this page: AI Wiki. "SemiAnalysis." aiwiki.ai, updated 1 Aug 2026, fact-checked 1 Aug 2026. CC BY 4.0. https://aiwiki.ai/wiki/semianalysis

Suggest edit