cleverhack.com
Spot an error or a missing company?
Let me know → cleverhack at duck.com
NVIDIA's Market Position and the Up-and-Coming AI Hardware CompetitorsNovember 2026
Where NVIDIA StandsNVIDIA is still the center of the AI hardware market, and the numbers aren't close. Per J.P. Morgan, it took 83% of AI chip revenue from 2023 to 2025 at 75% gross margins, and went from 1% of the S&P 500 in December 2022 to 8.5% by May 2026. Its last reported quarter was a record $96.2B, with $89.0B of that from data center.
What Changed This MonthOn November 12, Jensen Huang resigned as president, CEO and board member, effective immediately. CFO Colette Kress is interim CEO, and Harvey Jones is now the company's first-ever board chair. NVIDIA gave no reason. Press reports tie the exit to an internal investigation into Huang's ties to Megaspeed, a Singapore-based NVIDIA customer that Culper Research alleged in May is funded and controlled by Alibaba. Megaspeed denies wrongdoing. Treasury Secretary Scott Bessent wants an outsider in the job, while a group of investors has asked the board to keep Huang's return on the table. The stock sits 32% below its all-time high, and earnings land November 17.
The Share That Matters NextHere's the number I keep coming back to: 28%. That's NVIDIA's share of the roughly 44 GW of compute agreements the frontier labs have announced for 2026 or later, per J.P. Morgan. Broadcom custom chips have 23%, AMD 18%, Amazon Trainium 16%, Google TPUs 14% and Cerebras 2%. The competition was building well before the leadership news. Switching is getting cheaper too, as one frontier lab told J.P. Morgan that onboarding a model to AMD GPUs now takes three days, where a year ago it took 6–12 months.
The Knock-On EffectsNVIDIA doesn't just sell the chips. It has also been the AI industry's financier, putting 58% of its operating cash flow into equity investments and M&A over the last three quarters. That money backs the neoclouds and open-weight labs like Reflection, Thinking Machines and Poolside, and J.P. Morgan flags that the support could fade under new management. NVIDIA's roughly $12.9B acquisition of Hugging Face is expected to close in the first half of 2027, pending regulatory approval. On the China front, the U.S. House passed the Remote Access Security Act 369–22 in January, though the Senate hasn’t voted on it. Chinese labs are moving to domestic chips anyway. Z.ai says it trained GLM-Image entirely on Huawei Ascend, a research team including Huawei says it used Ascend 910C chips for post-training of DeepSeek-V4-Pro, and Meituan says it trained LongCat-2.0 on a domestic cluster without naming the chip.
Merchant SiliconThe established chip companies.
• AMD — Helios racks built on Instinct MI450-series GPUs; a 6 GW agreement with OpenAI, up to 2 GW from Anthropic, and a Meta commitment; Epoch AI's November 9 inference index ranks the MI455X above NVIDIA's VR200
• Broadcom — The custom-chip partner behind Google's TPU, Meta's MTIA and OpenAI's Jalapeño; $16.7B in AI semiconductor revenue in its fiscal Q3 2026, up 221% year over year
• Marvell — The other custom-chip house alongside Broadcom; targets more than $12B in custom silicon revenue by fiscal 2029, has a custom silicon agreement with Google, and closed its acquisition of optical interconnect startup Celestial AI in February 2026
• Qualcomm — Began shipping custom AI accelerators to two hyperscalers in October 2026, one of them Amazon; reiterated a $5B data center revenue target for fiscal 2027; also sells its own AI200 and AI250 inference racks
• Intel — Introduced the Intel GPU D-series (formerly Crescent Island) for AI inference in October 2026; CEO Lip-Bu Tan calls it a new market that will take time to scale; its custom-chip business more than doubled year over year
In-House Chips at the Labs and HyperscalersBuilt for their own workloads first.
• Google TPU — Now offered to outside customers, with Anthropic as the anchor and a reported multibillion-dollar Meta rental deal; SemiAnalysis modeling shows the TPU7x beating NVIDIA's GB300 on tokens per dollar at most speeds on Qwen 3.5 397B
• OpenAI Jalapeño — Inference chip co-designed with Broadcom, unveiled June 24, 2026; first deployment targeted for the end of 2026 as part of a 10 GW agreement running through 2029
• Amazon Trainium — 16% of announced frontier-lab compute agreements (7 GW), per J.P. Morgan
• Microsoft Maia 200 — Inference accelerator announced January 26, 2026
• Meta MTIA — Meta's in-house accelerator line, developed with Broadcom
Chip StartupsThe independent challengers, most of them aimed at inference.
• Cerebras — Wafer-scale chips; went public on Nasdaq (CBRS) on May 14, 2026, raising $5.55B; 0.75 GW of frontier-lab agreements
• Etched — Transformer-specific chips and racks; a reported $700M round led by Jane Street at a $21B valuation in August 2026, with Jane Street taking the first rack
• Fractile — UK inference chips that keep memory and compute together; a reported $720M Series C at a $7B valuation in November 2026 and a reported ~$250M agreement with Anthropic; first commercial chip planned for 2027
• Tenstorrent — RISC-V based AI chips and servers; $1.5B Series E at a $6.2B valuation, announced October 19, 2026
• Positron AI — Memory-first inference hardware; $875M at a $5B valuation, announced September 10, 2026
• MatX — Chips for large language models from ex-Google engineers; $500M Series B in February 2026, with shipping planned for 2027
• Groq — Fast-inference LPU chips; licensed its technology to NVIDIA in a roughly $20B non-exclusive deal announced December 2025, and continues as an independent company
• d-Matrix — Memory-centric compute for AI inference
• SambaNova — Inference chips and systems
Neolabs Building HardwareResearch labs going after the compute layer itself.
• Unconventional AI — Naveen Rao's lab building analog, brain-inspired chips; $475M seed at a $4.5B valuation in December 2025
• Ricursive Intelligence — Founded by AlphaChip creators Anna Goldie and Azalia Mirhoseini to have AI design chips; $300M Series A at a $4B valuation in January 2026, with NVIDIA's venture arm among the investors
• Callosum — Routes each step of an AI task to the best-suited model and chip; $100M seed led by Atomico in August 2026
Neolabs Backed by NVIDIAOpen-weight labs J.P. Morgan names as recipients of NVIDIA investment, support that could fade.
• Reflection — Open-weight frontier models
• Thinking Machines — Mira Murati's lab
• Poolside — Foundation models for software work
ChinaBernstein expects NVIDIA's share of China's AI chip market to fall to about 8% this year, with Huawei near 50%.
• Huawei Ascend — The domestic leader; Z.ai says it trained GLM-Image entirely on Ascend, and a research team including Huawei says it used Ascend 910C chips for post-training of DeepSeek-V4-Pro
• Cambricon — The largest domestic challenger after Huawei
• Restricted chips already in China — SemiAnalysis estimates 1.5 GW of restricted NVIDIA chips have ended up there
Exposed to NVIDIANeoclouds J.P. Morgan names as potential losers if NVIDIA changes its approach to vendor financing.
• CoreWeave — All-NVIDIA GPU fleet; NVIDIA is also an investor
• Nebius — AI cloud
• Applied Digital — AI data centers
• IREN — AI cloud and data centers
Sources• The Icarus Files: What's next for Nvidia and AI — Michael Cembalest, J.P. Morgan Eye on the Market, November 14, 2026
• NVIDIA Announces CEO Transition — NVIDIA, November 12, 2026
• Nvidia CEO Jensen Huang steps down — NPR, November 12, 2026
• Nvidia CEO exit: AMD, Broadcom and AI trade fallout — CNBC, November 13, 2026
• How Megaspeed built a network in Southeast Asia to move Nvidia chips to Chinese companies — Fortune, November 11, 2026
• NVIDIA financial results for Q2 FY27 — Jon Peddie Research
• Cerebras' wafer-scale AI bet delivers blockbuster IPO — The Register, May 15, 2026
• London's Fractile raises $720M and hits $7B valuation — The Next Web
• Tenstorrent raises $1.5bn at $6.2bn valuation — DCD
• Positron AI raises $875m at a $5bn valuation — DCD
• Chip startup MatX raises $500M — SiliconANGLE, February 24, 2026
• Jeff Bezos backs $475M seed round for Unconventional AI — SiliconANGLE, December 8, 2025
• Ricursive Intelligence raises $300M Series A — Converge Digest
• Etched raises $700M at a $21B valuation — Runtime Wire
• Nvidia's AI chip sales in China stall — AP via WTOP, June 2026
• NVIDIA Form 8-K on the Hugging Face acquisition — SEC, September 2, 2026
• Final Vote Results for Roll Call 13 (Remote Access Security Act) — U.S. House Clerk, January 12, 2026
• China's Z.ai Debuts First Major Model Trained Fully on Huawei's AI Chip Stack — eWeek, January 15, 2026
• Huawei chips refine DeepSeek model in major leap for China's AI self-reliance — SCMP, June 5, 2026
• Intel Q3 2026 earnings call prepared remarks — Intel, October 29, 2026
• Qualcomm Q4 2026 earnings call transcript — The Motley Fool, November 5, 2026
• Marvell Technology stock rallies following ambitious investor day targets — Yahoo Finance, October 6, 2026
|
|
|
|---|---|---|
|
|
|
|