TSMC's $2B Interposer Deal Is a Capacity Play
TSMC signed a five-year, $2 billion agreement with GlobalFoundries to make silicon interposers in Malta, New York. The engineering math explains why.
7 min read
Section
Hardware, software, startups, gadgets and the trends shaping the tech industry.
TSMC signed a five-year, $2 billion agreement with GlobalFoundries to make silicon interposers in Malta, New York. The engineering math explains why.
7 min read
Anthropic's rebuilt Cyber Verification Program gates the same Claude models with different classifiers. Its own benchmark shows what each tier trades away.
7 min read
Google paused open source bug bounty submissions on October 1 after invalid AI reports overwhelmed its engineers, and curl's data shows exactly why.
7 min read
Apple will require explicit user action before Mac apps get Full Disk Access, after Meta's Muse synced 187,462 rows of a journalist's private iMessages.
8 min read
Anthropic's IPO filing reveals Broadcom backstopping $42B in TPU leases. Inside the SPV, the $125.2B commitment, and the circularity question.
5 min read
A San Diego jury handed Taction $5.72B over Apple's Taptic Engine. The per-device royalty math and the non-willfulness finding are the real story.
7 min read
Google Project Suncatcher launches Oct. 1 with four TPUs, one kilowatt of solar power and 15-minute compute bursts. The thermal math is the real story.
7 min read

USD.AI's record $128.9M GPU-backed loan finances 32 rack-scale GB200 NVL72 systems. The GPU count, not the dollar figure, is the real signal.
8 min read

Alibaba unveiled the Zhenwu V900 chip and a 10-trillion-parameter model roadmap. The parameter count is vanity; the 500,000-card cluster is the claim.
7 min read

CXMT's G5 DRAM hits 11.95nm half-pitch with quadruple patterning, no EUV. The economics of multi-patterning explain why this changes DRAM pricing.
7 min read

DOE's Quantum Genesis Q pays only for verified logical qubits and fault-tolerant operations. The incentive design, not the $215M, is the story.
7 min read
TSMC's 2nm node is in volume production, but the true per-die cost hides behind yield and density, and it is far steeper than the headline wafer price suggests.
7 min read
SK Hynix is reportedly in talks with Intel to make memory chips in Ohio, and the tariff and trade pressure driving the top HBM maker toward its first real US fab.
6 min read
Meta is deploying its MTIA 450 Arke chip in the first half of 2027 to cut AI inference costs, but the savings claims rest on company numbers, not independent data.
6 min read
Omdia's 2Q26 report shows DRAM and NAND crossed 50% of all global chip revenue for the first time, and the cost is showing up in your next PC and TV.
6 min read
SpaceX hit its 700th Falcon 9 flight, landing a 29-flight booster. The reusability math inside turns spaceflight into a utility, and the numbers prove it.
8 min read
A $5,499 Mac Studio versus Claude Opus 5, GPT-6, and DeepSeek API pricing. We work through the actual break-even math for running AI models on your own hardware.
5 min read
DeepSeek V4.1 Flash uses 8B active parameters to beat a 1.6T model. This is the engineering behind its 890-byte KV cache and what it costs to serve.
11 min read
Quandela opened a photonic qubit pilot line, targeting 10,000 units a year and pushing yield past 70 percent. Here is what the fabrication means for quantum scale.
9 min read
OpenAI confirmed Samsung will manufacture its next custom chip, a supply chain move that splits fabrication away from TSMC and reshapes HBM economics.
7 min read
Why Samsung and SK hynix run under 10 days of inventory, how HBM4's 3:1 wafer math squeezes DDR5, and what the spot market tells builders.
10 min read
Apple unveiled the iPhone 18 Pro and the first foldable iPhone Duo. Every spec, price, and camera detail confirmed from Apple's official press release.
15 min read
Qualcomm filed SEC paperwork revealing Amazon warrants worth $4 billion and a multi-generation silicon roadmap as it cracks the AI data center market.
4 min read
Arm and Samsung are building a 2nm on-device AI accelerator to move inference off the cloud. A data scientist and engineer breakdown of why it matters.
6 min read
AMD's Threadripper Halo Station stacks 576GB of HBM3E to run trillion-parameter models locally. Here is the memory math and the cost trap nobody warns about.
7 min read
RTX 5090 vs MI350P vs RX 9070 XT: every GPU compared by specs, bandwidth, and GB-per-dollar. Includes the Threadripper Halo Station.
15 min read
AWS promised a December 2026 Saudi region and 50 MW of AI capacity, but turning LEAP announcements into live infrastructure is the harder test.
4 min read
TCS subsidiary HyperVault will spend up to Rs 70,000 crore to build India's first 1 GW liquid-cooled AI data centre in Hyderabad by mid-2028.
5 min read
Nvidia and Hugging Face signed a definitive $12.9 billion agreement that closes in the first half of 2027 while keeping the platform open for the entire ecosystem.
4 min read
Tesla launches its two-seat autonomous Cybercab on September 3. A data-scientist's read on the specs, pricing, and robotaxi economics.
4 min read
PLD Space extended its Series C by 108 million euros to 288 million total, with Mitsubishi Electric leading the tranche to fund commercial Miura 5 operations.
4 min read
Nissan and Honda signed a joint deal to standardize vehicle software and chip computers for 2029 cars, collapsing ECU counts from over 100 to fewer than 10.
5 min read
White House imposes steep tariffs on imported drones as Chinese manufacturers dominate humanoid robot shipments worldwide.
5 min read
Nvidia's Vera Rubin architecture pairs GPUs with custom CPUs and storage racks to master the emerging systems layer of AI infrastructure.
4 min read
Asahi Linux developer James Calligeros says M3 Mac support is nearly ready. Webcam, audio, USB 3.0 and Thunderbolt all working on Apple Silicon machines.
4 min read
Nvidia paid $6 billion to license Poolside's AI model platform and invested $1 billion, marking the third unconventional deal of its kind.
4 min read
IBM released Bob 2.0 with multi-agent architecture, subagents, parallel tool calling, and built-in cost analytics to help enterprises scale AI-driven development.
4 min read
NVIDIA announces Groq 3 LPX in full production, delivering 3,400 tokens per second at 100K context for agentic AI. Nebius is first to deploy.
4 min read
Apple refreshed the Mac Studio with the M5 Max and M5 Ultra — the latter is the company's most powerful chip ever, with up to 512GB of unified memory and 1.2TB/s of bandwidth. It's the desktop to end all local AI machines.
4 min read
Texas Gov. Greg Abbott turns on the AI data center boom, pausing 1,800 projects and demanding community approval, as Trump calls the move a 'mistake.'
3 min read
At Beijing's World Humanoid Robot Games, a Chinese humanoid sprinted faster than Usain Bolt in the 100m, shattered the standing high jump record and drew global attention.
3 min read
Microsoft's KB5121003 update is causing game crashes, reboots and access violations. Embark Studios confirms the issue and Microsoft is investigating.
2 min read
Nvidia will raise Grace Blackwell and Vera Rubin server prices up to 17% for 2027 shipments. Memory chip costs add billions to data center budgets.
3 min read
Samsung confirms August 27 reveal for the Galaxy S26 FE, featuring Exynos 2500 chip, larger display, and higher pricing. Here's everything leaked so far.
4 min read
Marvell issued Google a warrant for roughly 7% of the company, with shares vesting as Google's chip purchases accumulate. Google controls the pace entirely.
4 min read
Apple accidentally leaked a video of camera-equipped AirPods running Visual Intelligence, confirming the B790 project on track for 2027, per Bloomberg.
2 min read