High-signal people to follow on the AI industry: models, chips, memory, supply chains, policy, and the money behind them. No hype accounts, no repost-only feeds, no engagement farming. Every entry is an account a human reads regularly and can explain what it adds.
Seeded from a deal breakdown by @P_Bonnet (see Origin) that broke down who actually makes money when an AI company gets acquired.
| Handle | Focus | Output | Links |
|---|---|---|---|
| jukan05 | semis, AI hardware, China AI policy | scoops, research memos, conference notes | Substack RSS |
| zephyr_z9 | AI and chips, memory, compute economics | analysis threads | |
| divBy_zero | inference numerics, mixed-precision matmul, accelerators | technical deep dives | |
| QingFengJianZX | chip packaging materials, ABF substrates, concept stocks | market news and stock roundups (Chinese) | WeChat: qingfengfinance |
| earnings_guy | earnings coverage across the market, AI-adjacent names included | charted earnings summaries | Support |
| trendforce | memory pricing, HBM, CSP capex | price data and research summaries | Site Substack WeChat |
| xu159411209 | materials choke points, export controls | supply chain updates with stock picks (Chinese) | |
| antiAIvo | passive components (MLCC), price cycles | component market intelligence (Chinese) | |
| vikramskr | semiconductor physics, optics, nodes | long-form research and explainers | Substack RSS Podcast YouTube |
| lithos_graphein | EUV lithography, High-NA, fab timelines | daily commentary, threads, Chip War Dashboard | Site Substack RSS |
| pequityresearch | AI capex, earnings, financial modeling | earnings breakdowns, capex forecasts, deep dives | Substack RSS |
| tphuang | Chinese AI stack, chips, export controls, embodied AI | analysis threads, podcast, Substack | Substack RSS Podcast |
| __d_1ggin | AI infrastructure, data centers, grid, AI labor | infrastructure investment research | Substack RSS |
| P_Bonnet | who actually makes money in AI deals | deal-breakdown threads | |
| ChipsandWafers | packaging and substrate supply chains | analysis naming the tickers behind the flows | Podcast |
| Schulz_Research | AI supply chain: optics, memory, power | long-form teardowns naming the chokepoints | |
| jasonschips | optics/photonics, InP substrates, chip theses | stock pitches with the math attached | Substack RSS |
Website: https://www.semiconsam.com/
Semiconductor and AI hardware analyst on the Citrini Research team. Publishes long-form research on Substack as SemiconSam.
What you get: chip supply chain and memory semiconductor news, China AI policy scoops (multiple Reuters exclusives, including Beijing weighing curbs on overseas access to China's top models), model intel like GLM-5.x specs before launch, and first-hand notes from ICML. Skeptical of national AI hype cycles, and posts sources alongside the news.
Independent AI and chips analyst.
What you get: supply chain analysis of the hardware side of AI, built on a running thesis that memory is becoming the center of the AI stack now that Nvidia has lost the perf-per-dollar and perf-per-watt crown. Threads on model training economics (Meta's LLaMA-scale spending versus DeepSeek R1) and the business reality behind compute buildouts. Data-backed and consistently skeptical of hype.
Eric Quinnell. Hardware-focused engineer posting deep dives on inference numerics and accelerator engineering.
What you get: low-precision inference numerics explained at the level of what the silicon actually does. Recent posts break down the MXFP8 x MXFP4 mixed-precision trick (fp8 weights, fp4 activations, faster than a plain fp8 matmul) and the implementation cost: 2x activation fetches plus a scale sidecar running concurrently on the fp8 datapath to match native fp4 element sizes. Also tracks new accelerator silicon like Jalapeno versus VR200 and evaluates it honestly, crediting good engineering and calling out what is not trivial.
Chinese-language semiconductor supply chain news account (display name Qingfeng Finance News).
What you get: supply chain signals from the China market side: material shortages and the stocks tied to them. Recent posts cover the ABF substrate shortage (the laminate used in advanced chip packaging) and a roundup of the ABF concept stocks. Useful for tracking the packaging materials and substrate layer of the supply chain, where shortages surface in prices before Western coverage.
Support: https://ko-fi.com/earnings_guy
The Earnings Correspondent. Independent account posting earnings coverage for public companies.
What you get: fast, charted earnings summaries the moment results drop: $TICKER (Company) #earnings are out, with TradingView charts and the key numbers. Covers the AI-adjacent names (ASML, IREN, monday.com) alongside the rest of the market, which makes it a useful filter for the business reality behind AI capex. Also posts a weekly lookahead of which companies report next.
Website: https://www.trendforce.com/
Market intelligence firm founded in 2000, covering semiconductors, memory, and displays. Corporate account, not an individual analyst.
What you get: the memory price data everyone else quotes. Posts DRAM and NAND pricing moves, HBM contract price forecasts (another 70-140% up in 2027), and CSP capex allocation analysis, with links to the full research. Where the individual analysts on this list give you takes, TrendForce gives you the numbers underneath.
Chinese-language semiconductor supply chain analyst posting under the display name Zhaowen.
What you get: supply chain updates with stock picks attached. Recent posts cover the materials choke points: high-purity quartz, germanium, and high-end NdFeB magnets, with mainland export review tightening since late 2025 slowing Taiwan equipment deliveries by months, per Nikkei Asia. Reads like a sector research note: what is tightening, who is exposed, which names benefit.
Chinese-language supply chain watcher posting under the display name Nano AI.
What you get: component market intelligence with an AI server angle. Recent posts cover the MLCC price cycle: Samsung Electro-Mechanics leading Q4 OEM price hikes (consumer X5R up 25-30%, AI-server X6S up 10-20%), with Murata, Taiyo Yuden, and Kyocera holding back for now and some capacity utilization nearing 90%. Tracks passive components the way others on this list track memory: who hikes, who follows, what it signals about the AI server buildout.
Website: https://www.viksnewsletter.com/
Vikram Sekar. Independent semiconductor researcher, physics-first as he puts it. Writes Vik's Newsletter on Substack and co-hosts the Semi Doped podcast.
What you get: long-form physics-first semi research, plus short posts that land the idea. Examples from his feed: "Optics is the next Memory. You just don't know it yet," and the reminder that a 1.8nm node has 20-30nm physical transistors, so angstrom-class nodes are a figure of speech. 172 articles and roughly 400K words over 2.5 years on Substack. The account that the individual analysts and the price-data firms on this list both read.
Website: https://lithosgraphein.com/ Substack: https://lithosgraphein.substack.com/
Lithos Graphein. Three decades in semiconductor lithography, per the bio. Runs the Chip War Dashboard (lithosgraphein.com), a live feed of semi news and market data, and guest-contributes to SemiAnalysis.
What you get: daily commentary and hot-takes on the chip war from a lithography veteran: EUV and High-NA economics, ASML, fab construction timelines ("at least 2 years from fab construction start to production"), and the process-technology layer the supply chain accounts on this list skim over. Long threads on ASML Investor Day and SPIE coverage.
Website: https://pequityresearch.substack.com/
P Equity Research. Equity researcher with an accounting background (B.S. and M.S.) who publishes on Substack under the same name.
What you get: AI capex and earnings analysis with financial modeling underneath. Recent posts cover the J.P. Morgan hyperscaler capex forecast (top 7 hyperscalers spending close to $1.5 trillion in 2027) and Meta's Q2 earnings breakdown (revenue up 28%, costs up 55%, operating margin from 43% to 31%). Subscriber-gated deep dives on top of the free analysis, including collaborations like the substrate supply chain piece with Arvind.
Substack: https://tphuang.substack.com/
TP Huang. Former Salesforce VP turned technology investor, and one of the better China-watchers on AI. Hosts the China Tech Talk podcast and writes a Substack under the same name.
What you get: direct-source analysis of the Chinese AI stack - chips, export controls, and the diverging supply chain - from someone who reads the Chinese coverage and primary material. Recent work (the Multipolarity Dialogues episode among it) argues the post-2022 export controls have set up a supply chain divergence that is building toward a space race over chips and software, and that China's edge is cost, not speed: it operates at a fraction of the cost of US rivals, so the perpetual six-to-twelve-month gap may not be decisive. Also covers embodied AI - models embedded into factories and robots - where China is integrating AI into physical systems rather than chasing California's AGI narrative.
Substack: https://leoinai.substack.com/
Leo (Leonardo Boquillon). Senior software engineer writing on AI and equities, running the Leonardo Boquillon Briefs Substack under the tagline "Thinking about AI when the incentives point the other way."
What you get: research on the AI infrastructure buildout as an investment theme. Recent posts break down GPU depreciation economics in the neoclouds (is the thesis overstated?), the power grid's 26 consecutive quarters of growth, the 500-plus-company data center supply chain from high-voltage transformers to server racks, and the AI labor shift as a way to invest in the buildout without owning the model names. Cites the actual infrastructure vendors behind the capex - CoreWeave, Nebius, IREN, Koncar among them.
Reference post: https://x.com/P_Bonnet/status/2092946471989465470
Breaks down who actually makes money in AI deals. The post that seeded this list: Nvidia acquires Hugging Face for about $12.9bn, with investors sharing roughly $7.3bn of profit on under $400m invested (about 20x blended) and Lux Capital the single biggest winner.
Reference post: https://x.com/ChipsandWafers/status/2093999509747413158
Chips & Wafers tracks the semiconductor supply chain, with a lens on packaging and substrate sourcing. The post that seeded this list: Singapore ramping its packaging capacity and importing more substrates, and where those substrates come from - Malaysia - tied to AT&S (ATS.VIE) and Simmtech (222800.KQ).
What you get: supply-chain analysis that names the listed names behind the physical flows (packaging, IC substrates), with a focus on where the wafers and substrates actually get made.
Reference post: https://x.com/Schulz_Research/status/2094071237916193258
Independent AI supply-chain research: optics, memory, power, Asia and US filings read in the original and translated into pricing - "the chokepoints nobody prices in". The post that seeded this list: a teardown of the indium phosphide (InP) stack - the metal (not the binding constraint), the wafer (the real scarcity), and the PRC export permits that catch AXT (AXTI) rather than its Japanese peers.
What you get: long-form supply-chain teardowns that separate the material, the wafer, and the paperwork, and name the companies actually caught by each.
Website: https://jasonschips.ai/
Jason's Chips. Buyside AI stonks investor and writer - "20 & college dropout", all views his own, not financial advice. First-principles logic, economics, and humor on semiconductor and AI investing, with a heavy optics/photonics bent: InP substrates, co-packaged optics, and the names behind them.
What you get: stock theses with the math attached - AXTI (his "biggest position"), the Soitec photonics-substrate series, Sivers as the most overvalued stock in optics - plus earnings reviews of Nvidia, ARM, Lumentum, CoreWeave, and Tower Semiconductor.
Curated by hand. An account qualifies when it meets all of:
- Original analysis with sources, not reposts or link drops
- A track record: months of consistent output, not one viral thread
- Stable identity, verifiable as a real person or team
- Active: posted within the last month
Exclude: engagement farming, AI-generated content without substance, accounts that mostly reply to others, and teasers that lead nowhere without the paywall clearing.
Want to add someone? Open a PR with the handle and a short note on what the account adds. The criteria above apply.