publish date 2026-07-16
Welcome back! This is Canaan's weekly newsletter on Bitcoin mining, energy, and compute infrastructure.
Last year, Elon Musk told CNBC that after the chip shortage and the transformer shortage, the next constraint on AI would be electricity itself: "My guess is people are going to start hitting challenges with power generation maybe by the middle of next year, end of next year."
The middle of next year is now.
The prediction itself was not controversial. By mid-2025, almost everyone in energy and compute expected a power wall. That was why many bitcoin miners were repriced as power companies, why hyperscalers started signing nuclear deals, and why "energized megawatts" became a metric that moved stocks. The industry saw the bottleneck coming. But has it actually arrived? And does it look the way people expected?
On the first question, the evidence is not subtle. The largest hyperscalers are committing more than $750 billion in capital expenditures for 2026, most of it destined for AI infrastructure. The EIA projects U.S. electricity consumption will set records in both 2026 and 2027, with commercial demand overtaking residential demand for the first time this year. During the most recent heat wave, PJM reached a preliminary all-time peak load of approximately 168 GW, surpassing the previous record set in 2006. The Department of Energy responded by issuing its third emergency order of the year, directing data centers and other large loads to switch to their own backup generation as a last resort before rolling blackouts.
So yes, energy is still the bottleneck. It arrived roughly on schedule. But the shape of the bottleneck is different from what most people were picturing, and that difference matters.
The immediate constraint in many major markets is not simply how much electricity can be generated. Delivery seems to be the bigger issue. Industry analysis projects that 30–50% of the data center capacity planned for 2026 will slip, with interconnection queues now exceeding 2,100 GW nationwide and grid connection timelines running three to seven years in the major markets. Large power transformers that took two years to procure before 2020 are now quoted at four to five years. Gas turbine order books at the major OEMs stretch into 2029 and 2030. The United States does not lack electrons, but it does lack the ability to plug in hundred-megawatt loads within the timelines that capital expects.
Beneath the supply chain sits an institutional layer that responds to capital even less: interconnection study backlogs, transmission siting, state utility commission dockets, environmental permitting. You cannot spend your way through a permit. That procedural stack, as much as any piece of hardware, is why a lot of planned capacity slips, and why headline capex converts into energized megawatts far more slowly than the number suggests. Anyone underwriting "energized megawatts" should be discounting the timelines rather than just the costs.
So now roughly 30% of planned new data center capacity is designed around on-site generation, up from effectively zero a year ago. Brookfield's recent decision to expand its fuel-cell financing framework with Bloom Energy from $5 billion to $25 billion is capital buying its way around the permitting queue, not merely adding generation.
On the second question, flexible, interruptible loads are rapidly becoming the price of admission for any large load that wants a grid connection this side of 2030.
The ability to absorb surplus power in one hour and disappear during the next hour is the operating model many bitcoin miners have run for a decade. That’s the theme Canaan's Gwyn Lauber explored yesterday in a conversation with Blockchain Association EVP Dan Spuller, and the question she'll take up on stage at the Energy Investors Forum in Dallas next week.
One thing worth noting is that distillation, smaller task-specific models, and the shift from training toward inference are all pushing compute-per-task down. This means that, if efficiency outruns adoption, the power wall could ease sooner than the capex commitments imply. So far, cheaper inference has meant more total usage, a pattern consistent with the Jevons paradox. The demand curve can be less certain than the headlines make it look, which is itself an argument for flexibility over fixed bets.
Overall, it’s unlikely that the bottleneck is going away anytime soon. And, for the years ahead, the advantage belongs to operators who treat power as something to be scheduled, shaped, and shared rather than merely consumed.
In the News
Network at a Glance
Project Spotlight
By late 2019, the question hanging over ASIC design was whether the efficiency curve could keep bending. Canaan's answer was the Avalon 10 series. The flagship AvalonMiner 1066, released in November 2019, packed 342 of Canaan's 16nm A3205 chips into a single unit, delivering 50 TH/s at 3,250W — (roughly 65 J/TH) with the 1066 Pro pushing to 55 TH/s at 60 J/TH. It was the first Avalon to clear the 50-terahash mark, more than doubling the output of the Avalon 9 generation in a single product cycle.
The timing gave the machine a place in industry history beyond its spec sheet. Weeks after the A10 series launched, Canaan completed its IPO on Nasdaq — the first bitcoin mining manufacturer to list on a major U.S. exchange. The machine that had started as a boxy 66 GH/s desktop experiment in 2013 had become, six years and roughly 750x later, the flagship product of a publicly traded semiconductor company. The A10 went on to become a workhorse of the industrial farms, then scaling across North America and Central Asia — the generation where Avalon's four-fan, die-cast aluminum design language settled into the form factor the fleet still recognizes today.
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 year, Elon Musk told CNBC that after the chip shortage and the transformer shortage, the next constraint on AI would be electricity itself: "My guess is people are going to start hitting challenges with power generation maybe by the middle of next year, end of next year."
The middle of next year is now.
The prediction itself was not controversial. By mid-2025, almost everyone in energy and compute expected a power wall. That was why many bitcoin miners were repriced as power companies, why hyperscalers started signing nuclear deals, and why "energized megawatts" became a metric that moved stocks. The industry saw the bottleneck coming. But has it actually arrived? And does it look the way people expected?
On the first question, the evidence is not subtle. The largest hyperscalers are committing more than $750 billion in capital expenditures for 2026, most of it destined for AI infrastructure. The EIA projects U.S. electricity consumption will set records in both 2026 and 2027, with commercial demand overtaking residential demand for the first time this year. During the most recent heat wave, PJM reached a preliminary all-time peak load of approximately 168 GW, surpassing the previous record set in 2006. The Department of Energy responded by issuing its third emergency order of the year, directing data centers and other large loads to switch to their own backup generation as a last resort before rolling blackouts.
https://x.com/SmallCapSnipa/status/2028495005896053017?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E2028495005896053017%7Ctwgr%5Eaee903a2aabd9d0056e246afbd2b956fd06e79b4%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fcanaan-newsletter.beehiiv.com%2Fp%2Felon-was-right-750-billion-is-looking-for-a-socket
So yes, energy is still the bottleneck. It arrived roughly on schedule. But the shape of the bottleneck is different from what most people were picturing, and that difference matters.
The immediate constraint in many major markets is not simply how much electricity can be generated. Delivery seems to be the bigger issue. Industry analysis projects that 30–50% of the data center capacity planned for 2026 will slip, with interconnection queues now exceeding 2,100 GW nationwide and grid connection timelines running three to seven years in the major markets. Large power transformers that took two years to procure before 2020 are now quoted at four to five years. Gas turbine order books at the major OEMs stretch into 2029 and 2030. The United States does not lack electrons, but it does lack the ability to plug in hundred-megawatt loads within the timelines that capital expects.
Beneath the supply chain sits an institutional layer that responds to capital even less: interconnection study backlogs, transmission siting, state utility commission dockets, environmental permitting. You cannot spend your way through a permit. That procedural stack, as much as any piece of hardware, is why a lot of planned capacity slips, and why headline capex converts into energized megawatts far more slowly than the number suggests. Anyone underwriting "energized megawatts" should be discounting the timelines rather than just the costs.
So now roughly 30% of planned new data center capacity is designed around on-site generation, up from effectively zero a year ago. Brookfield's recent decision to expand its fuel-cell financing framework with Bloom Energy from $5 billion to $25 billion is capital buying its way around the permitting queue, not merely adding generation.
On the second question, flexible, interruptible loads are rapidly becoming the price of admission for any large load that wants a grid connection this side of 2030.
The ability to absorb surplus power in one hour and disappear during the next hour is the operating model many bitcoin miners have run for a decade. That’s the theme Canaan's Gwyn Lauber explored yesterday in a conversation with Blockchain Association EVP Dan Spuller, and the question she'll take up on stage at the Energy Investors Forum in Dallas next week.
https://x.com/canaanio/status/2074152841300586913?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E2074152841300586913%7Ctwgr%5Eaee903a2aabd9d0056e246afbd2b956fd06e79b4%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fcanaan-newsletter.beehiiv.com%2Fp%2Felon-was-right-750-billion-is-looking-for-a-socket
One thing worth noting is that distillation, smaller task-specific models, and the shift from training toward inference are all pushing compute-per-task down. This means that, if efficiency outruns adoption, the power wall could ease sooner than the capex commitments imply. So far, cheaper inference has meant more total usage, a pattern consistent with the Jevons paradox. The demand curve can be less certain than the headlines make it look, which is itself an argument for flexibility over fixed bets.
Overall, it’s unlikely that the bottleneck is going away anytime soon. And, for the years ahead, the advantage belongs to operators who treat power as something to be scheduled, shaped, and shared rather than merely consumed.
In the News
- CleanSpark Signs $6.6B Data Center Lease
- Bitcoin Difficulty Slides 5% as Hashrate Retreats to 908 EH/s
- Galaxy Adds Former Xerox CEO to Board
- Meta Expands Louisiana AI Data Center to 5 GW in $50 Billion Buildout
- Solo Bitaxe Miner Beats 18,000-Year Odds to Mine Full Bitcoin Block
- Nscale’s £2 Billion UK AI Data Center Reportedly Faces Grid Delay
Network at a Glance
- BTC price (USD): ~$65,324
- Network hashrate: 907 EH/s
- Difficulty: 127.17T
- Hashprice: ~$32.49 / PH / day
Project Spotlight
By late 2019, the question hanging over ASIC design was whether the efficiency curve could keep bending. Canaan's answer was the Avalon 10 series. The flagship AvalonMiner 1066, released in November 2019, packed 342 of Canaan's 16nm A3205 chips into a single unit, delivering 50 TH/s at 3,250W — (roughly 65 J/TH) with the 1066 Pro pushing to 55 TH/s at 60 J/TH. It was the first Avalon to clear the 50-terahash mark, more than doubling the output of the Avalon 9 generation in a single product cycle.
The timing gave the machine a place in industry history beyond its spec sheet. Weeks after the A10 series launched, Canaan completed its IPO on Nasdaq — the first bitcoin mining manufacturer to list on a major U.S. exchange. The machine that had started as a boxy 66 GH/s desktop experiment in 2013 had become, six years and roughly 750x later, the flagship product of a publicly traded semiconductor company. The A10 went on to become a workhorse of the industrial farms, then scaling across North America and Central Asia — the generation where Avalon's four-fan, die-cast aluminum design language settled into the form factor the fleet still recognizes today.
Events and Media
- Today, Canaan announced it was granted an 180-day extension to regain compliance with Nasdaq’s minimum bid price requirement.
- Canaan released its monthly production update, recording its strongest monthly increase in BTC in its treasury this year.
- Yesterday, Gwyn Lauber joined speakers from NVIDIA and Samsung at PowerUp Data Centers in Austin, Texas, for the following panel: “How Processor, Server, and Data Center Energy Management Advances Are Affecting Onsite Power and Grid Connection Needs”
- Yesterday, Canaan hosted an X Space with Blockchain Association EVP Dan Spuller on U.S. digital asset policy, bitcoin mining, energy infrastructure, and the future of compute.
- On July 23, Canaan will be represented at the Energy Investors Forum in Dallas on the panel "Balancing Act: How Digital Infrastructure Can Stabilize the Grid", alongside Brad Cuddy (Cholla), Arseniy Grusha (Dataprana), JohnPaul Baric (MiningStore), and Taras Kulyk (Synteq).
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/.
