Hardware
The photomask holding back big AI dies gets bigger in 2031
The 6x12 photomask transition for High-NA EUV has ASML, TSMC, Samsung and Intel aligned. Pilot line in 2031, production around 2033. Here is why it ...
Memory is sold out until 2027, and SSDs are up 147 percent
RAM prices 2027 outlook: contract prices nearly doubled in Q1 and rose 58 percent in Q2, SSDs are up 147 percent, and capacity is booked out. What to ...
How much RAM you need in 2026, now that it costs this much
DRAM prices are up 171 percent year over year, so how much RAM you need in 2026 is now a budget question, not a spec sheet question. Here are the ...
A 2nm chip has nothing on it that is 2 nanometres wide
A 2nm chip has no feature 2 nanometres wide. Node names are foundry labels now, and the real gain from N2 is roughly 15 percent in one direction you ...
Threadripper Halo Station has no price, but the parts add up to 150,000 dollars
AMD Threadripper Halo Station price is unannounced, but 96 cores, dual MI350P accelerators and 2TB of DDR5 put the parts bill close to 150,000 ...
vLLM vs Ollama is the wrong question if you own spare machines
vLLM vs Ollama is the usual local LLM argument, but LocalAI and exo solve different problems. Here is how to pick a serving layer for your hardware.
CXMT G5 DRAM is in mass production, and export controls did not stop it
CXMT G5 DRAM has reached mass production at 11.95nm using quadruple patterning, adding a fourth supplier to a memory market that is sold out into ...
Nvidia GPUs, Google TPUs and AWS Trainium all run AI, but only one is rentable
A plain comparison of GPU vs TPU vs Trainium vs custom ASIC in 2026, including the part most explainers skip: three of the four are not for sale.
Amazon will buy up to 60 billion dollars of Qualcomm AI chips
The Qualcomm Amazon AI chip deal covers custom inference silicon and 1.6T optical interconnects, and turns a phone chip company into a data centre ...
RAM is up 89 percent this year. Buy now, not in 2027.
Should you buy RAM now in 2026 or wait for prices to fall? DRAM is up over 400 percent since 2024 and every published forecast says 2027 is not the ...
Everyone quotes qubit counts. Almost nobody quotes the one that matters.
Logical qubits versus physical qubits in 2026: the overhead ratio fell roughly tenfold this year across several unrelated hardware platforms at once.
Nvidia promised to double chip shipments. Memory and substations may not let it.
Nvidia chip sales 2027 guidance doubles unit shipments. But $279 billion in supplier commitments now runs into tight DRAM supply and grid transformer ...
A 1500VA UPS for your home server is really a 900W UPS
Sizing a UPS for a home server or NAS comes down to watts, not VA, and runtime you will actually use. Here is the maths, plus five units worth buying.
Wi-Fi 8 is not faster than Wi-Fi 7. That is the entire point.
Wi-Fi 8 vs Wi-Fi 7 comes down to reliability, not speed. Same 46 Gbps ceiling, same spectrum, but 25% lower latency and far fewer dropouts when ...
Huawei just moved its Nvidia rival forward by nine months
Huawei's Ascend 960 now lands in Q1 2027, nine months early. Here is how the Ascend 960 vs Nvidia comparison actually reads, and what still caps it.
The VRAM tier list, and exactly which AI models your GPU can run
Local LLM VRAM requirements in 2026, by tier. What 8GB, 16GB, 24GB and 48GB actually run, plus which model to pick at each level and why.
Gartner says memory prices climb 130% this year and AI is eating 70% of supply
RAM prices will not go down in 2026. Gartner forecasts a 130% surge, AI data centres take 70% of supply, and the earliest real relief is late 2027.
Cornelis put processors inside the network to stop GPUs sitting idle
Cornelis Networks raised 205 million dollars for Active Compute Fabric, embedding RISC-V cores in NICs and switches to cut GPU time spent waiting on ...
What you can actually run locally in 2026, by how much RAM you have
Local LLM hardware requirements by tier, from an 8GB laptop to a 128GB tower, plus the two rules that decide which models will actually fit on your ...
Nvidia Is Raising AI Server Prices by 15 Percent, and Memory Is the Reason
The Nvidia AI server price increase for 2027 shipments runs past 15 percent on Vera Rubin and Grace Blackwell systems. Memory costs are driving it ...
Enflame IPO Was Oversubscribed 4,073 Times by Retail Investors
The Enflame IPO closed 4,073 times oversubscribed on retail demand, a Shanghai listing that says more about China
iPhone Ultra Price: Apple Is Reportedly Asking 2,400 Dollars
The iPhone Ultra price is reported at 2,400 dollars, which would make Apple's first foldable the priciest iPhone ever sold. What lands on 9 September.
When Will RAM Prices Drop? The Honest Answer Is 2027
When will RAM prices drop? Analyst forecasts point to Q1 or Q2 2027, and the reason for the delay is structural rather than a normal supply spike.
AMD Helios vs Nvidia Vera Rubin: The Memory Number Decides This One
AMD Helios vs Nvidia Vera Rubin comes down to memory, not FLOPS. AMD claims 31TB of HBM4 per rack against Nvidia
What Is HBM4, and Why Your AI Accelerator Is Starving Rather Than Thinking
What is HBM4 memory explained in plain terms: how stacked memory works, why bandwidth beats raw FLOPS for AI inference, and why only three firms can ...
Snapdragon C Puts Arm Windows Laptops in the Budget Aisle
Snapdragon C debuts in the Acer Aspire Go 15, pushing Arm Windows laptops into the entry tier with up to 8GB RAM, Wi-Fi 6E and dual USB-C ports.
Nvidia RTX Spark: A Petaflop And 128GB Of Memory On A Desk
Nvidia RTX Spark specs: a 20 core Grace CPU, a Blackwell RTX GPU, up to 1 petaflop of AI compute and 128GB of unified memory in a desk-side box this ...
Wigner Crystal Explained: Physicists Just Watched One Move
Wigner crystal explained: electrons locking into a repeating pattern through mutual repulsion alone, and the Basel and Munich team who finally watched ...
HBM vs GDDR vs DDR: Which Memory Does What, and Why One Costs 100 Dollars a Stack
HBM vs GDDR vs DDR explained: three memory types, three different jobs. Bandwidth, latency, power and cost compared, plus what actually goes in your ...
High Bandwidth Flash Just Got Its First Open Spec, and Samsung Answered With zHBM
High Bandwidth Flash has its first open spec: 512GB stacks and up to 3.0TB/s over UCIe. Here is how HBF differs from HBM, and where Samsung zHBM fits.
Nvidia's H200 Reached China and HBM Memory Prices Jumped 20 Percent
Nvidia H200 China export approval pushed HBM3E prices up nearly 20 percent for 2026. Six memory stacks per GPU explain the whole move.
Google's Pixel 11 line lands with Tensor G6, the first 2nm phone chip
Google's Pixel 11 series launched 12 August 2026 with the first 2nm smartphone chip. Prices, specs and the honest take.
Nvidia squeezes a data centre chip into a laptop with RTX Spark
Nvidia's RTX Spark Superchip fuses a Grace CPU with a Blackwell GPU for laptops arriving fall 2026.
Global Chip Sales Just Smashed Every Record With a $403 Billion Quarter
Global semiconductor sales hit $403.3 billion in Q2 2026, up 35% from Q1 with a 123% year-over-year spike in June driven by AI demand.
Sony and TSMC Are Building a $6.4 Billion Factory for Machine Eyes
Sony and TSMC signed a $6.4 billion deal to build an image sensor fab in Kumamoto, Japan, targeting smartphones, autonomous vehicles and robotics.
Lumilens Raised 700 Million to Bring Optical Networking to the Data Center
Lumilens raised 700 million dollars to bring optical networking to the data center, swapping electrical wiring for light to break the AI bandwidth ...
Tesla and SpaceX Break Ground on Terafab, a 16.8 Billion Dollar Texas Chip Plant
Tesla and SpaceX are building Terafab, a 16.8bn dollar Texas chip factory to feed their own AI, robotics and space compute. Here is what it means.
SK Group and NVIDIA's $500bn pact just made memory the most important chip in AI
SK Group and NVIDIA signed a partnership worth more than 500 billion dollars to build Vera Rubin AI infrastructure powered by SK hynix HBM4 memory ...
Why RAM prices are climbing fast in 2026, and AI is the reason
RAM prices have jumped sixfold in a year as memory makers chase AI demand. Here is why DDR5 costs more in 2026 and what it means for your next build.
Arm laptops go mainstream: what a phone chip in your PC actually changes
Arm laptops are going mainstream in 2026. Here is what a phone-style chip in your PC changes for battery life, speed, app support and price.
AMD Advancing AI 2026: EPYC Venice and the Helios rack aim at Nvidia
AMD launched EPYC Venice on TSMC's 2nm process and the Helios AI rack with 31TB of HBM4, plus the MI455X accelerator, aimed squarely at Nvidia's lead.
Google Is Building a Server Chip It Claims Is Up to 10 Times More Efficient Than Its TPUs
Google is reportedly developing a server chip code-named Frozen v2, built around the Gemini architecture, with internal claims of 6 to 10 times the ...
Budget phones are about to get a lot more expensive
The budget smartphone market is under real pressure this year, and the cause isn't demand, it's memory.
Samsung's foldables just got real specs, days before Unpacked
Samsung's Galaxy Unpacked lands on Tuesday 22 July, streamed from London at 2pm BST (9am EDT), and in the days before it the guessing game has mostly.
MSI's new handheld beats AMD's best chip, at a very steep price
MSI's Claw 8 EX AI+ is now on sale, and it's the first handheld gaming PC built around Intel's brand new Arc G3 Extreme processor, a 14-core chip ...
Google's Fitbit Air Is a Screenless, 12-Gram Band With a Week of Battery
Google's Fitbit Air ditches the screen for a 12-gram band with about a week of battery life.
Best budget 4K monitors in 2026: the ones actually worth buying
The best budget 4K monitors in 2026, tested and compared. Real prices, real specs, and which panel wins for work, gaming, PS5, and Mac.
Meta just launched its cheapest smart glasses yet, and the price is the point
Meta introduced its most affordable smart glasses to date, a clear push to move the category out of early-adopter territory and onto ordinary faces.
AMD RX 9060 XT: the $299 GPU that actually has enough VRAM
AMD's Radeon RX 9060 XT launched June 5 at $299 with 16GB of VRAM. It outperforms the RTX 4060 Ti by 15-20% and makes the RTX 5060 look overpriced.
Samsung Is Betting $648 Billion on AI Chips Over the Next Decade
Samsung Group is committing 1,000 trillion won ($648bn) to AI chip factories, data centres, and displays. Here's what the scale of this bet actually ...
Qualcomm wants to buy Tenstorrent for up to $10 billion to take on Nvidia
Qualcomm is in early talks to acquire AI chip startup Tenstorrent for $8 to $10 billion. Jim Keller's RISC-V bet could be the most credible challenge ...
IBM crams 100 billion transistors onto one chip
IBM's NanoStack architecture puts 100 billion transistors on a single chip with 50% more performance and 70% less energy, via vertical stacking.
OpenAI built its own AI chip and called it Jalapeno
OpenAI and Broadcom unveiled Jalapeno, OpenAI's first custom AI chip. Prototypes by end of 2026, full scale by 2028. Here is why it matters.
Intel's 18A-P node hits a milestone, with Apple and Google reportedly circling
The advanced process enters risk production, and a tie-up with Taiwan's UMC takes aim at TSMC's foundry lead.
Microsoft's 2026 Surface line is its best in years, and its most expensive
OLED screens, new haptics and a two-stage chip rollout. The catch is a price hike that pushes Surface above equivalent Macs.
Nvidia just walked into the PC chip market, and Intel should be worried
The RTX Spark Superchip puts a 20-core CPU and a Blackwell GPU on one piece of silicon. It lands in laptops this autumn.
OpenAI built its own chip to stop paying Nvidia so much
OpenAI revealed Jalapeno, its first custom inference processor built with Broadcom, as it signs giant compute deals and tries to cut the cost of ...
China is spending $295 billion on AI infrastructure. Here is what that actually means.
China's National Development and Reform Commission is finalising a five-year, $295 billion plan to build a nationwide AI data centre network. Nvidia ...
The US Government Thinks ASML's Most Advanced Chip Tool Is Inside China
The US government has raised the alarm that one of ASML's most powerful extreme ultraviolet lithography machines, the kind that's essential for.
Quantum computers just got dramatically more reliable, and almost nobody noticed
Microsoft reports a 1,000x reliability jump in its Majorana qubits, while error correction across the industry moves from lab trick to engineering ...