basicmi/AI-Chip: a curated index of AI accelerator ICs and IP
A list of ICs and IPs for AI, Machine Learning and Deep Learning.
At a glance
- What is it?
- The basicmi/AI-Chip repository tracks AI accelerator silicon and licensable IP across vendors, cloud providers and startups. It is a reading list, not a benchmark suite, and the README is the product.
- Who is it for?
- Adopt it as a starting map when you need to know which companies build AI accelerators and which IP blocks exist, and when you want one page that links out instead of twenty vendor pages. Do not adopt it as a performance reference: there are no benchmark numbers, no prices and no per-chip specifications, and MLPerf results are handled as external links.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 76 days ago.
- What is it written in?
- Mainly PHP, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 30, 2026, and from our analysis. They are not legal advice.
Editorial analysis
What basicmi/AI-Chip actually is
This is a Markdown index. The repository holds README.md, _config.yml and a resource directory, and the README is where the content lives. It groups AI accelerator silicon into four buckets: IC vendors such as Intel, Qualcomm, Nvidia, Samsung, AMD, IBM and Marvell; tech giants and HPC vendors such as Google, Amazon AWS, Microsoft, Apple, Alibaba, Tencent Cloud, Baidu, Fujitsu, Nokia, Facebook and Tesla; IP vendors such as ARM, Synopsys, Imagination, CEVA, Cadence and VeriSilicon; and a long startup list running from Cerebras and Graphcore through to Neureality, Analog Inference, Quadric, EdgeQ, Innatera and Aspinity. A shortcut table at the top links into each of those sections, so the README doubles as its own navigation.
The intended reader is someone who needs to know the shape of the field rather than the details of one part. An engineer writing a competitive brief, a student picking a thesis topic, or a product manager trying to work out whether an edge inference problem belongs to Hailo, Kneron or a licensable NPU from Synopsys. The list answers the question of who exists. It does not answer what any of them costs, how fast they run, or whether you can buy one today.
How the README is organised and updated
There is no build step, no database and no generated output. The README is hand-edited Markdown with anchor links, and the anchors are what hold it together: the shortcut table points at IDs such as #IC_Vendors, #IP_Vendors and #Startups_Worldwide, and each vendor section carries a matching anchor. The homepage at basicmi.github.io/AI-Chip/ is a GitHub Pages site driven by _config.yml, so the published page and the repository README are two views of the same text.
The Latest updates block at the top is the changelog, and it is written as a flat list of one-line additions: a piece of SambaNova news, a Groq update, a d-Matrix entry, an IBM AIU item, a Tesla Dojo item, and repeated entries pointing at the MLPerf Results from MLCommons. That block is the best signal of what the editor has been tracking, but it is not a dated changelog. There are no timestamps on individual entries, so you cannot tell from the README alone when a given vendor line was added or last checked. The commit history on master is the only place that information exists.
The resource directory holds the images the README embeds, including an AI chip landscape diagram and the header graphic. Those diagrams are versioned by filename, and the README references a landscape image with a v0p7 suffix, which suggests the diagram is revised in place rather than regenerated from data.
Reading the list from your machine
There is no package to install. The README gives no installation instructions because there is nothing to install: the deliverable is a text file and a static page. The practical first step is to open the repository so you can follow the shortcut table and the vendor sections, and to use the anchor IDs the README defines to jump between them.
The README's own navigation is built from anchors such as #IC_Vendors, #IP_Vendors and #Startups_Worldwide. Those are the entry points the editor maintains, and following them in order is the fastest way to see the four-way split before reading any single vendor entry.
If you want the rendered page rather than the source, the homepage is the GitHub Pages build of the same content, and _config.yml controls that build. The README does not document a local preview command for the Pages site, so treating the Markdown as the source of truth is the simpler path.
What the list does not tell you
The most important limitation is that this is an index, not a datasheet. There are no TOPS figures, no memory bandwidth numbers, no process nodes, no power envelopes and no prices anywhere in the README. If your question is which accelerator gives the best throughput per watt for a specific model, this repository will not answer it, and the MLPerf entries it links to are pointers to MLCommons rather than results reproduced in the repository.
Coverage is also uneven by design. The Latest updates block shows repeated additions for the same handful of names, Cerebras, Groq, SambaNova, Nvidia, Tesla and d-Matrix, while many entries in the shortcut table appear once and are never revisited. A vendor that stopped publishing news may still be listed, and a vendor that started shipping last month may be missing. Nothing in the README marks an entry as stale, defunct or acquired.
Finally, the repository is Markdown plus images, and the primary language reported for the repository is PHP, which is a configuration artefact rather than a codebase. There is no parser, no schema and no validation. Nothing checks that an anchor still resolves or that a link still works, so broken links accumulate silently until someone edits the file.
How it compares with Awesome lists and vendor documentation
The obvious alternative is a general Awesome-style list of machine learning resources. Those tend to be broader and shallower: frameworks, papers, courses and datasets in one file, with hardware as one section among many. basicmi/AI-Chip goes the other way. It is narrow and deep on silicon and IP, with the four-way split between IC vendors, hyperscalers, IP licensors and startups, and it keeps a running news block that a general list would not maintain. If your question is specifically about accelerator hardware, the narrower scope saves you filtering.
The other alternative is going straight to vendor documentation and MLPerf result tables. That is the only source for actual numbers, and for anyone making a purchasing or design decision it is the correct place to end up. The difference in approach is that vendor documentation is authoritative but fragmented across dozens of sites with no shared taxonomy, while this list imposes one taxonomy and then hands you off. Use the list to build the candidate set, then leave it.
Maintenance, licence and the cost of keeping it current
The last push to master was on 2026-07-16, which is recent enough that the repository is not abandoned, but the README carries no per-entry dates, so the freshness of any individual vendor line is unknown. The upgrade cost is close to zero in the software sense: there is no dependency to bump and no API to track. The real cost is editorial. Every entry is a claim about a company, and companies in this sector rename products, get acquired and shut down. Keeping the list accurate means periodically opening the links and removing what no longer resolves.
The repository does not state a licence. The README, the _config.yml and the images are all in the repository, and without a licence file the default position is that the author retains rights, which matters if you intend to republish the landscape diagram or mirror the list inside internal documentation. The README does link the editor's WeChat blog and LinkedIn profile, which suggests the list is a personal curation project rather than a formally governed dataset. If reuse matters to you, ask before copying the images; linking to the repository is the safe path.
Where this list earns its place
The strongest part of the repository is the startup section. It names companies that do not appear in mainstream hardware coverage, including Koniku, Adapteva, Leepmind, Gyrfalcon Technology, GreenWaves, Anaflash, Optalysys, Eta Compute, Areanna AI, Neuroblade, Luminous Computing, AISTORM, GrAI Matter Lab, Rain Neuromorphics, Applied Brain Research, XMOS, DinoPlusAI, Furiosa, Perceive, SimpleMachines, TeraMem, Ceremorphic and Aspinity. For anyone mapping the field or looking for acquisition targets, that breadth is the reason to open the page.
The four-way taxonomy is the second useful thing. Splitting licensable IP from shipping silicon is a distinction that most hardware roundups blur, and it changes the question you ask: an IP vendor sells a block you integrate, while an IC vendor sells a part you socket. The shortcut table makes that split visible in one screen.
What you should not expect is depth on any single entry. The README is a set of pointers, and the value is in the pointer set, not in the prose around it.
Editorial conclusion
Adopt it as a starting map when you need to know which companies build AI accelerators and which IP blocks exist, and when you want one page that links out instead of twenty vendor pages. Do not adopt it as a performance reference: there are no benchmark numbers, no prices and no per-chip specifications, and MLPerf results are handled as external links. Before relying on it, open the vendor entries that matter to you and confirm each link still resolves, and check the commit history on master to see which sections were touched most recently.
Frequently asked questions
What exactly is an AI chip, according to basicmi/AI-Chip?
The repository treats AI chips as a broad category covering ICs and IPs for AI, machine learning and deep learning, and splits them across IC vendors, tech giants and HPC vendors, IP vendors, and startups. That range includes licensable NPU blocks as well as packaged accelerators.
Does basicmi/AI-Chip say which AI chip is the best?
No. The README is a list of vendors, IP blocks and news links, and it contains no benchmark numbers, no prices and no comparative rankings. The only performance-related pointers are links to the MLPerf Results from MLCommons, which live outside the repository.
Does basicmi/AI-Chip list AI chip prices?
It does not. Nothing in the README gives a price for any accelerator or IP block, and the entries are vendor names and news links rather than product listings. Pricing has to come from the vendor.
What is the use of an AI chip in a laptop?
The README does not discuss laptop implementations or consumer devices. Its scope is data centre accelerators, licensable IP and startup silicon, so questions about laptop AI hardware fall outside what the repository documents.
What does an AI chip do in the products listed by basicmi/AI-Chip?
The repository does not describe what any individual chip does. It lists IC vendors, IP vendors, hyperscalers and startups with links to their news, and leaves the function of each part to the linked source.
Is basicmi/AI-Chip about AI chip design?
Partly. The IP vendors section covers licensable blocks from ARM, Synopsys, Imagination, CEVA, Cadence and VeriSilicon, which is the design side of the list, while the IC vendors and startup sections cover companies shipping or developing their own silicon.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/basicmi-ai-chip)