Hardware
Anker unveils Eufy MindBase hub with on-device LLM for home security
Device processes camera video without sending data to the cloud, using an AI model trained by Anker itself.
Sections
Hardware
Everything this paper has published on Hardware, newest first.
26 stories · By Max · Hardware
Hardware
Fujitsu Monaka stacks full cache on separate 5nm die
Fujitsu's Japanese processor stacks all cache on a dedicated die, arriving in 350W and 500W versions with samples available and volume production planned for 2027.
Hardware
Headphones are becoming AI. Nobody's saying who pays for it
Live translation in your ear burns hundreds of calls per conversation. While the model runs on someone's server, someone is billing.
Archive
August 2026
Issue 15
August 4–9, 2026HP, Asus and Acer buying Chinese memory isn't drama, it's engineeringThe shortage bit hard and manufacturers accepted a second source. It's the most basic practice of anyone running something critical.Issue 15
August 4–9, 2026Optics inside the package: the fight is over who owns the bottleneckTSMC, Intel, Samsung and GlobalFoundries are racing to swap copper for light. Whoever controls the interconnect controls the next decade.Issue 15
August 4–9, 2026Memory prices return to 2007 levels, erasing twenty years of declineThe AI build-out pushed memory to prices the market hasn't seen since before the first iPhone. A billion dollars of phone silicon waits on a shelf.Issue 15
August 4–9, 2026Portable data centres tested as an answer to the space shortageThe proposal assembles capacity in transportable modules, bringing processing closer to available power.Issue 14
July 27 – August 2, 2026Musk announces a 100-million-square-foot chip plant in TexasTesla and SpaceX are putting in $16.8 billion to start. The stated reason is that their combined demand already exceeds what the market can supply.Issue 14
July 27 – August 2, 2026AMD buys the company that etches the model into the siliconThe Taalas acquisition targets inference performance an order of magnitude higher, with the weights baked into the chip itself.
August 4–9, 2026HP, Asus and Acer buying Chinese memory isn't drama, it's engineeringThe shortage bit hard and manufacturers accepted a second source. It's the most basic practice of anyone running something critical.Issue 15
August 4–9, 2026Optics inside the package: the fight is over who owns the bottleneckTSMC, Intel, Samsung and GlobalFoundries are racing to swap copper for light. Whoever controls the interconnect controls the next decade.Issue 15
August 4–9, 2026Memory prices return to 2007 levels, erasing twenty years of declineThe AI build-out pushed memory to prices the market hasn't seen since before the first iPhone. A billion dollars of phone silicon waits on a shelf.Issue 15
August 4–9, 2026Portable data centres tested as an answer to the space shortageThe proposal assembles capacity in transportable modules, bringing processing closer to available power.Issue 14
July 27 – August 2, 2026Musk announces a 100-million-square-foot chip plant in TexasTesla and SpaceX are putting in $16.8 billion to start. The stated reason is that their combined demand already exceeds what the market can supply.Issue 14
July 27 – August 2, 2026AMD buys the company that etches the model into the siliconThe Taalas acquisition targets inference performance an order of magnitude higher, with the weights baked into the chip itself.
July 2026
Issue 13
July 20–26, 2026The memory price increase now shows up in the largest budgetsTwenty-five billion of Microsoft's capex increase is price alone. The squeeze runs through the whole chain.Issue 12
July 13–19, 2026Micron adds $50 billion, reaching $250 billion planned through 2035The investment covers fabs in New York, Idaho and Virginia. The timeline is a decade; the shortage is now.Issue 12
July 13–19, 2026Meta puts its own silicon into productionThe company's in-house chip leaves the design stage and enters the line. It's the same path other large players announced this month.Issue 12
July 13–19, 2026Nvidia introduces switching at 102.4 terabits per secondDouble the previous generation. The connection between machines became the bottleneck deciding overall performance.Issue 11
July 6–12, 2026Nvidia doubles switching capacity and targets the bottleneck between machinesThe new platform moves 102.4 terabits per second. In a large cluster, the cable decides performance.Issue 11
July 6–12, 2026Micron adds $50 billion in US fabsThe planned total rises to $250 billion through 2035. Relief, if it comes, arrives at the turn of the decade.Issue 11
July 6–12, 2026Intel commits €5 billion to European productionThe announcement is part of the geographic redistribution of manufacturing that marked the month.Issue 11
July 6–12, 2026India enters commercial chip packagingThe final manufacturing stage became a visible bottleneck this month, with finished dies waiting for assembly.Issue 10
June 29 – July 5, 2026US government plans $300 million to speed data movement between processorsThe award targets interconnect technology, the bottleneck the industry has been naming for months.Issue 10
June 29 – July 5, 2026A manufacturer swaps waiting for launching on the current chipNot waiting for the next processor suggests the intended differentiator sits elsewhere.Issue 10
June 29 – July 5, 2026Interconnect becomes a target of industrial policyPublic funding aimed at connections between processors shows where the bottleneck was officially recognised.
July 20–26, 2026The memory price increase now shows up in the largest budgetsTwenty-five billion of Microsoft's capex increase is price alone. The squeeze runs through the whole chain.Issue 12
July 13–19, 2026Micron adds $50 billion, reaching $250 billion planned through 2035The investment covers fabs in New York, Idaho and Virginia. The timeline is a decade; the shortage is now.Issue 12
July 13–19, 2026Meta puts its own silicon into productionThe company's in-house chip leaves the design stage and enters the line. It's the same path other large players announced this month.Issue 12
July 13–19, 2026Nvidia introduces switching at 102.4 terabits per secondDouble the previous generation. The connection between machines became the bottleneck deciding overall performance.Issue 11
July 6–12, 2026Nvidia doubles switching capacity and targets the bottleneck between machinesThe new platform moves 102.4 terabits per second. In a large cluster, the cable decides performance.Issue 11
July 6–12, 2026Micron adds $50 billion in US fabsThe planned total rises to $250 billion through 2035. Relief, if it comes, arrives at the turn of the decade.Issue 11
July 6–12, 2026Intel commits €5 billion to European productionThe announcement is part of the geographic redistribution of manufacturing that marked the month.Issue 11
July 6–12, 2026India enters commercial chip packagingThe final manufacturing stage became a visible bottleneck this month, with finished dies waiting for assembly.Issue 10
June 29 – July 5, 2026US government plans $300 million to speed data movement between processorsThe award targets interconnect technology, the bottleneck the industry has been naming for months.Issue 10
June 29 – July 5, 2026A manufacturer swaps waiting for launching on the current chipNot waiting for the next processor suggests the intended differentiator sits elsewhere.Issue 10
June 29 – July 5, 2026Interconnect becomes a target of industrial policyPublic funding aimed at connections between processors shows where the bottleneck was officially recognised.
June 2026
Issue 09
June 22–28, 2026Apple prepares a touchscreen laptopThe devices are expected to ship with the current processor generation rather than waiting for the next.Issue 08
June 15–21, 2026The open question in China's plan is interconnectConnecting data centres at national scale is a different problem from filling one.Issue 06
June 1–7, 2026An in-car assistant reaches vehicles across two hardware generationsCompatibility with older hardware is the hard part and the most valuable to existing owners.
June 22–28, 2026Apple prepares a touchscreen laptopThe devices are expected to ship with the current processor generation rather than waiting for the next.Issue 08
June 15–21, 2026The open question in China's plan is interconnectConnecting data centres at national scale is a different problem from filling one.Issue 06
June 1–7, 2026An in-car assistant reaches vehicles across two hardware generationsCompatibility with older hardware is the hard part and the most valuable to existing owners.
April 2026
Issue 01
April 20–26, 2026AI data centres consume the equivalent of a US state's peak demandThe annual index puts global draw at 29.6 gigawatts, with nearly every leading chip coming from one country.Issue 01
April 20–26, 2026A model co-designed with the newest accelerator generationDesigning model and hardware together left the lab and entered commercial launches.Issue 01
April 20–26, 2026The electrical grid became the sector's limit on expansionA 29.6 gigawatt draw explains why capacity contracts began being negotiated in power.
April 20–26, 2026AI data centres consume the equivalent of a US state's peak demandThe annual index puts global draw at 29.6 gigawatts, with nearly every leading chip coming from one country.Issue 01
April 20–26, 2026A model co-designed with the newest accelerator generationDesigning model and hardware together left the lab and entered commercial launches.Issue 01
April 20–26, 2026The electrical grid became the sector's limit on expansionA 29.6 gigawatt draw explains why capacity contracts began being negotiated in power.