publish date 2026-07-14
Welcome back! This is Canaan's weekly newsletter on Bitcoin mining, energy, and compute infrastructure.
Last week, Palantir CEO Alex Karp said on CNBC's Squawk Box that "something has gone completely wrong" with how AI is sold. He argued that enterprises paying metered token fees to frontier labs like OpenAI and Anthropic are getting little return while handing over the proprietary data (the "alpha") that makes their businesses defensible. In fairness, two days earlier, the company had announced an expanded NVIDIA partnership to deploy open-weight models in customer-controlled environments, so the critique conveniently sells Palantir’s product. But the sentiment he tapped is real, and it has been building for a while.
The trust problem he referred to has a known case study. Anthropic's chief product officer resigned from Figma's board three days before the company launched Claude Design, a product that competes directly with Figma. Meanwhile, the economics are doing their own persuading: Coinbase said it cut internal AI spending by nearly half by routing engineers to open-weight models, and open-weight architectures have closed much of the capability gap while activating a fraction of the parameters per query.
Put economics, IP protection, and vendor risk together, and the direction of travel is toward models that enterprises own, running on hardware those enterprises control.
The infrastructure implications are yet to be clearly understood. So far, the frontier-lab model concentrates demand by training clusters and API inference live in gigawatt-scale campuses. The grid's job has been to feed a few enormous, firm loads.
Sovereign AI inverts that. Training stays centralized, but inference fragments into on-premise clusters, colocation cages, and regional facilities in the 5–50 MW range. Inference does not require every workload to live inside a gigawatt campus and could push capacity toward smaller, modular, distributed facilities.
That becomes a very different grid problem (and opportunity). Thousands of mid-size interconnections instead of a queue of mega-loads. Latency and data-residency pulling compute toward demand centers and jurisdictions. Utilization could follow business hours rather than running flat. If this is where enterprise AI goes, the winners will be operators who can deploy modular capacity quickly, in unglamorous locations, at megawatt rather than gigawatt scale.
And that is the playbook that many Bitcoin miners have run for a decade. It is also the thesis Canaan's CEO and founder, NG Zhang, recently laid out: the AI buildout has barely begun, and it looks like the early years of Bitcoin mining.
Most likely, the hyperscalers are not going away anytime soon because of the unique nature of the frontier training economics. But if Karp is right that enterprises want to own the means of production, the next phase of the buildout will be measured in thousands of small interconnection requests, not a handful of headline campuses.
In the News
Network at a Glance (July 8, 2026)
Project Spotlight: The A3206 Chip
By 2018, the easy wins in Bitcoin ASIC design had narrowed. Process shrinks from 55nm to 40nm to 28nm to 16nm had each delivered step-change efficiency gains, but the next node (7nm) was the bleeding edge of global semiconductor manufacturing, territory usually reserved for flagship smartphone silicon.
Canaan got there anyway. The A3206 was the company's first 7nm chip, and it powered the ninth generation of Avalon miners. The flagship AvalonMiner 921 packed 104 of them into a single unit, delivering 20 TH/s at roughly 1,800W — an efficiency of 0.09 J/GH, which was very competitive for the 2018 transition to 7nm. The 9-series also carried forward the ROQ ("Reliable Open Quality") branding and the Airforming cooling design, with up to twenty units manageable from a single AvalonMiner Controller - 400 TH/s under one roof, a figure that would have represented a meaningful share of the entire network just five years earlier.
Events and Media
Follow and Contact Us
Disclaimer: This newsletter shares industry commentary and third-party news for informational purposes only. The views and opinions expressed by third-party sources are those of their respective authors and do not necessarily reflect the views of Canaan Inc. For official news, please refer to Canaan’s press releases and SEC filings at https://investor.canaan-creative.com/.
Last week, Palantir CEO Alex Karp said on CNBC's Squawk Box that "something has gone completely wrong" with how AI is sold. He argued that enterprises paying metered token fees to frontier labs like OpenAI and Anthropic are getting little return while handing over the proprietary data (the "alpha") that makes their businesses defensible. In fairness, two days earlier, the company had announced an expanded NVIDIA partnership to deploy open-weight models in customer-controlled environments, so the critique conveniently sells Palantir’s product. But the sentiment he tapped is real, and it has been building for a while.
https://x.com/PalantirTech/status/2072326189079757277?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E2072326189079757277%7Ctwgr%5Ed1d9e9d93f639d7f5bed6fe408a24603e2d4b4b6%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fcanaan-newsletter.beehiiv.com%2Fp%2Fwhat-would-an-enterprise-ai-revolt-mean-for-the-grid
The trust problem he referred to has a known case study. Anthropic's chief product officer resigned from Figma's board three days before the company launched Claude Design, a product that competes directly with Figma. Meanwhile, the economics are doing their own persuading: Coinbase said it cut internal AI spending by nearly half by routing engineers to open-weight models, and open-weight architectures have closed much of the capability gap while activating a fraction of the parameters per query.
Put economics, IP protection, and vendor risk together, and the direction of travel is toward models that enterprises own, running on hardware those enterprises control.
https://x.com/ihtesham2005/status/2073512349651374092?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E2073512349651374092%7Ctwgr%5Ed1d9e9d93f639d7f5bed6fe408a24603e2d4b4b6%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fcanaan-newsletter.beehiiv.com%2Fp%2Fwhat-would-an-enterprise-ai-revolt-mean-for-the-grid
The infrastructure implications are yet to be clearly understood. So far, the frontier-lab model concentrates demand by training clusters and API inference live in gigawatt-scale campuses. The grid's job has been to feed a few enormous, firm loads.
Sovereign AI inverts that. Training stays centralized, but inference fragments into on-premise clusters, colocation cages, and regional facilities in the 5–50 MW range. Inference does not require every workload to live inside a gigawatt campus and could push capacity toward smaller, modular, distributed facilities.
That becomes a very different grid problem (and opportunity). Thousands of mid-size interconnections instead of a queue of mega-loads. Latency and data-residency pulling compute toward demand centers and jurisdictions. Utilization could follow business hours rather than running flat. If this is where enterprise AI goes, the winners will be operators who can deploy modular capacity quickly, in unglamorous locations, at megawatt rather than gigawatt scale.
And that is the playbook that many Bitcoin miners have run for a decade. It is also the thesis Canaan's CEO and founder, NG Zhang, recently laid out: the AI buildout has barely begun, and it looks like the early years of Bitcoin mining.
https://x.com/canaanio/status/2071619690099536006?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E2071619690099536006%7Ctwgr%5Ed1d9e9d93f639d7f5bed6fe408a24603e2d4b4b6%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fcanaan-newsletter.beehiiv.com%2Fp%2Fwhat-would-an-enterprise-ai-revolt-mean-for-the-grid
Most likely, the hyperscalers are not going away anytime soon because of the unique nature of the frontier training economics. But if Karp is right that enterprises want to own the means of production, the next phase of the buildout will be measured in thousands of small interconnection requests, not a handful of headline campuses.
In the News
- TeraWulf jumps on $19 billion data center lease deal with Anthropic
- Blockfusion Touts Up to $5.4B From AI Lease as SPAC Deal Advances
- DOE Lets PJM Tap AI Data Center Backup Power During Heat Wave
- Ionic Raises $400M as AI Revenue Tops Bitcoin Mining Ahead of Nasdaq Listing
- Cipher CEO Files to Sell $2.4M of Stock as AI Rally Triggers Trading Plan
- Sphere 3D Seeks DarkHorse Rebrand as It Pivots Toward AI Infrastructure
Network at a Glance (July 8, 2026)
- BTC price (USD): ~$61,940
- Network hashrate: 920 EH/s
- Difficulty: 133.87T
- Hashprice: ~$29.28 / PH / day
Project Spotlight: The A3206 Chip
By 2018, the easy wins in Bitcoin ASIC design had narrowed. Process shrinks from 55nm to 40nm to 28nm to 16nm had each delivered step-change efficiency gains, but the next node (7nm) was the bleeding edge of global semiconductor manufacturing, territory usually reserved for flagship smartphone silicon.
Canaan got there anyway. The A3206 was the company's first 7nm chip, and it powered the ninth generation of Avalon miners. The flagship AvalonMiner 921 packed 104 of them into a single unit, delivering 20 TH/s at roughly 1,800W — an efficiency of 0.09 J/GH, which was very competitive for the 2018 transition to 7nm. The 9-series also carried forward the ROQ ("Reliable Open Quality") branding and the Airforming cooling design, with up to twenty units manageable from a single AvalonMiner Controller - 400 TH/s under one roof, a figure that would have represented a meaningful share of the entire network just five years earlier.
Events and Media
- Leo Wang joined Soluna’s CEO, John Belizaire, on the Clean Integration podcast to discuss the Project Dorothy partnership and what 1 exahash behind-the-meter actually looks like.
- Canaan’s vice president of corporate affairs, Gwyn Lauber, will join speakers from NVIDIA and Samsung on July 14 at PowerUp Data Center in Austin, Texas, for the following panel: “How Processor, Server, and Data Center Energy Management Advances Are Affecting Onsite Power and Grid Connection Needs”
Follow and Contact Us
- X: @canaanio
- Youtube: Canaanmining
- LinkedIn: Canaan Inc.
- Website: canaan.io
- Support: Customer Care
Disclaimer: This newsletter shares industry commentary and third-party news for informational purposes only. The views and opinions expressed by third-party sources are those of their respective authors and do not necessarily reflect the views of Canaan Inc. For official news, please refer to Canaan’s press releases and SEC filings at https://investor.canaan-creative.com/.
