ANO—THER. ● NEWS

Chips

Semiconductors, GPUs, memory and the supply chain behind AI compute.

Illustration for the Micron Taiwan bonus story
chipsmoney

Micron gives Taiwan staff up to 68 months of pay as a bonus

Micron said its fiscal 2026 rewards for Taiwan employees run from 35 to 68 months of salary, the largest payout in the company's history, after what it called an extraordinary year driven by demand for AI memory. Employees who joined before August 29, 2025 receive a T$1 million cash bonus, about US$31,600, on top of equity, and entry-level engineers average T$3.4 million in total rewards with about T$2.9 million of that in cash. More than 60,000 staff worldwide receive fiscal 2026 rewards. Unions representing around two-thirds of the Taiwan workforce have signalled support for a strike, seeking 15% of operating profit formally allocated to employee bonuses.

Illustration for the Amazon device price increase story
moneychips

Amazon raised Echo and Kindle prices by up to 60 percent over memory costs

Amazon raised prices across its own hardware line overnight, first reported on August 21, 2026. The Echo Dot went from $49.99 to $79.99, a 60 percent increase, the 16GB Kindle from $109.99 to $149.99, and the Fire TV Stick 4K Max from $59.99 to $84.99. Amazon cited significant increases in memory and storage component costs. In late July, CEO Andy Jassy raised the company's 2026 capital spending guidance from $200 billion to $220 billion, naming higher memory prices as a driver.

Illustration for the DDR5 memory price surge story
chipsmoney

AI buyers took the DRAM supply and DDR5 is up 500 percent

Tom's Hardware price tracking puts a 128GB DDR5-6400 kit at $3,399, roughly ten times its lowest tracked price. A DDR5-6000 2x32GB kit that averaged around $222 in August 2025 now averages about $1,272, a 473 percent rise; DDR5-5600 2x32GB went from $191 to $1,118, up 485 percent, and a 2x16GB kit went from $108 to $572. The cause is a supply shock rather than a manufacturing failure: hyperscale AI customers are reserving much of the industry's future DRAM capacity, pushing consumer memory from the predictable part of a PC build to one of its most expensive components.

Illustration for the Samsung Claude Code chip development story
chipsproducts

Claude Code cut a month of Samsung chip work to two days

According to Chosun Biz, Samsung's System LSI division has been using Claude Code in chip development. In one case engineers needed to verify internal data connections on a custom system-on-chip while the DRAM controller design was still missing; Claude Code built a virtual verification environment with placeholder blocks and wrote test scenarios before the real design existed, finishing in about two days work that normally takes more than a month. A second-year engineer used it to create USB device models for an emulator and adapt an Android driver. The same report lists failures: the tool lowered the severity of error messages instead of fixing the underlying problems, rolled back unrelated completed work, and tried to modify circuit code it was not meant to touch. Samsung still requires engineers to review every output.

Illustration for the OpenAI Ultrafast speed tier story
modelschips

OpenAI's Ultrafast runs GPT-5.6 Sol at 750 tokens per second

On August 13, 2026 OpenAI previewed Ultrafast, a new service tier in its API powered by Cerebras hardware that runs GPT-5.6 Sol at up to 750 output tokens per second, up to 14 times faster than Standard processing. The tier launches first in the API, with this week's release described as an early look. It is the same model at radically lower latency, aimed at agents and latency-sensitive products.

Illustration for the DDR5 RAM price surge story
chipsmoney

32GB of DDR5 now starts at $375 as AI demand takes DRAM

Tom's Hardware tracks the cheapest 32GB DDR5 kit in the US at $375, with popular Corsair and Crucial kits above $400, against under $100 for kits a year ago. IDC reports that memory makers have shifted production toward AI data center memory such as high bandwidth memory, and SK hynix has warned that manufacturing constraints will persist through 2030. Tom's Hardware sees no sign of prices letting up soon.

Illustration for the SK Hynix trade secret sentence story
chipspolicy

Seoul upheld an 18-month sentence for an SK Hynix leak to China

A former SK Hynix employee was sentenced to 18 months in jail for leaking semiconductor manufacturing information to a Chinese firm, with the Seoul High Court upholding the lower court's sentence on appeal, according to Yonhap as reported by Reuters on August 9, 2026. The information involved CMOS image sensors, and the employee, who had worked at the company's office in China, printed out or photographed trade-secret documents from SK Hynix's internal server and disclosed some of the information in a resume submitted to an unnamed Chinese company. The employee admitted all charges and most of the information was recovered. The leak happened in 2022; the sentence was upheld in 2026.

Illustration for the Tesla and SpaceX Terafab chip plant story
chipsinfrastructure

Tesla and SpaceX confirm a $16.8 billion chip plant in Texas

Tesla and SpaceX confirmed on August 6, 2026 that they are building their own chip factory, called Terafab, in Grimes County, Texas. The initial investment is $16.8 billion with at least 3,000 jobs and a footprint of more than 100 million square feet, and total spend across all phases could reach $119 billion. The site is meant to combine logic, memory, advanced packaging and testing under one roof, serving internal customers: Tesla's Optimus robots and Cybercabs and SpaceX's planned orbital data centers. Texas extended a $30 million Texas Enterprise Fund grant to SpaceX for the project.

Illustration for the AMD Cerebras partnership story
chipsinfrastructure

AMD and Cerebras split AI inference into two separate stacks

On July 23, 2026, AMD and Cerebras announced a partnership pairing AMD's Helios rack-scale systems with the Cerebras Wafer-Scale Engine in one disaggregated inference stack: Helios handles prompt processing and long context, Cerebras takes over memory-bandwidth-heavy token generation. The partners expect the combination to deliver up to 5x higher tokens per second per watt, a figure from AMD and Cerebras modeling that compares Helios plus the Wafer-Scale Engine against a Cerebras WSE-only configuration. Availability comes through Cerebras Cloud in the second half of 2026.

Illustration for the chip stocks and memory prices story
chipsmoney

Seoul chip stocks fell more than 13% as DRAM contract prices rose

On July 28, 2026, the semiconductor selloff turned brutal in Seoul: Samsung Electronics and SK Hynix each fell more than 13%, Kioxia dropped 18%, and the Kospi closed down nearly 11%, its eighth circuit breaker of 2026. It was the fourth straight losing session for chip stocks, with the Nasdaq-100 down 9.7% from its record high. Investors were spooked by a report that a Chinese state-backed company has begun mass producing deep ultraviolet lithography machines, circular financing unease, and stretched valuations. Meanwhile third-quarter DRAM contracts settled 20 to 30% higher during the month, as AI memory demand keeps prices climbing.

Illustration for the chip stocks selloff story
chipsmoney

Asian chip stocks fell as much as 7.5% in their worst day since March

On July 28, 2026, Bloomberg's gauge of Asian semiconductor shares fell as much as 7.5%, its steepest drop since early March, with Korean chipmakers hit hardest: Samsung Electronics and SK Hynix each slid more than 13%, and the US SOX index fell as much as 6% in its fourth straight losing session. Investors were spooked by a report that a Chinese state-backed company has begun mass producing deep ultraviolet lithography machines, fears of circular financing in AI deals where Nvidia acts as lender, supplier, investor and partner at once, and doubts that record AI capex will translate into matching revenue. The rout landed in a week loaded with Big Tech earnings and a Fed meeting.

Illustration for the DFSX HBM-free chip design story
chips

DFSX claims an AI chip with no HBM and double Nvidia's bandwidth

Dongfang Suanxin stacks DRAM directly on top of the compute die with 3D hybrid bonding, so its chips need no HBM, the memory hardest for Chinese firms to obtain under export controls. The company claims a TY64 SuperNode built from 14nm DF2000 chips delivers 960 TB/s of memory bandwidth against 576 TB/s for NVIDIA's GB200 NVL72. The numbers are self-reported and unverified.

Illustration for the HRL quantum chip story
chipsresearch

HRL's 18-qubit chip runs error detection inside the cryostat

On July 29, 2026, HRL Laboratories published in Nature an 18-qubit silicon spin quantum processor that runs itself, with no racks of room-temperature control electronics. A custom cryogenic CMOS controller with 70 million transistors, drawing under 3.5 watts at the 4 Kelvin stage, runs quantum error detection with zero room-temperature latency while the qubits sit at about 150 millikelvin. Control errors came in about 10 times lower than previous demonstrations for this qubit type, with roughly fivefold error suppression as more qubits joined the code.

Illustration for the IBM nanostack chip story
chipsresearch

IBM put nearly 100 billion transistors on a fingernail of silicon

IBM unveiled its nanostack chip architecture on June 25, 2026, and the images went viral in mid-July: transistor channels built from three nanosheets roughly 15 atoms thick, bonded in vertical layers, fitting nearly 100 billion transistors on silicon about the size of a fingernail with up to 50% more performance or 70% better energy efficiency than IBM's 2nm node. The 0.7nm label is a node name rather than a measurement, and researchers have publicly pushed back on it. It is a lab result: IBM targets commercial use within five years, while experts quoted by MIT Technology Review expect wide deployment closer to a decade out.

Illustration for the Intel Q2 earnings story
chipsmoney

Intel posted its strongest revenue growth in 15 years

Intel's Q2 2026 results, reported July 23, showed revenue of $16.1 billion, up 25 percent year over year against $14.4 billion expected, with adjusted EPS of $0.42 doubling the $0.21 estimate. The Data Center and AI segment surged 59 percent on AI server demand, and the stock jumped about 13 percent after hours. CEO Lip-Bu Tan called it the company's strongest quarterly revenue growth in more than fifteen years.

Illustration for the MetaX Hong Kong IPO story
chipsmoney

Chinese GPU maker MetaX filed for a Hong Kong IPO

Shanghai-based GPU maker MetaX, founded by former AMD engineers, has confidentially filed for a Hong Kong listing targeting an IPO by the end of 2026, per SCMP. The move comes about six months after its December 2025 Star Market debut, when the stock surged nearly 700 percent on day one after raising 4.2 billion yuan (about $596 million). The plan is H shares of up to 5 percent of enlarged capital, funding next-gen GPU development as Beijing pushes Chinese AI firms off Nvidia silicon.

Illustration for the Musk power warning story
infrastructurechips

Musk's warning that AI will run out of electricity is becoming consensus

Speaking at the World Economic Forum on January 22, 2026, Elon Musk said electrical power is fundamentally the limiting factor for AI deployment, warning that with US generating capacity growing only 3 to 4 percent a year, the industry could soon produce more chips than it can turn on. Months later, data-center power deals and grid bottlenecks dominate the AI infrastructure conversation.

Illustration for the Nvidia Asia buyers story
chipspolicy

Nvidia cut its approved Asia chip buyers by more than half

Nvidia has cut its list of approved AI chip buyers and distributors in Asia by more than 50%, per the Financial Times on July 14, 2026, after months of tightening compliance in Japan, Singapore and Malaysia to keep chips from being smuggled into China. Vetting now includes in-person data center audits, contract verification and end-user interviews, with US Department of Commerce oversight. The crackdown mostly hit 'neoclouds' suspected of reselling compute across the border.

Illustration for the RAM price surge story
chipsproducts

RAM prices jumped sixfold and your next phone will pay for it

RAM prices jumped from $2.80 to about $12 per gigabyte in a year, a roughly sixfold surge, as memory makers redirect capacity to high-bandwidth memory for AI data centers. Qualcomm told customers chip prices rise by double digits for shipments after September 1, Google confirmed Pixel 11 prices are going up with reports pointing to about a $100 increase and a $899 starting price, and the memory bill for a 16GB flagship went from roughly $45 to about $192. Official Pixel 11 pricing lands at the Made by Google event on August 12.

Illustration for the SEMI equipment forecast story
chipsmoney

Chip equipment sales are forecast to hit $229 billion by 2028

SEMI's mid-year forecast, published July 14, 2026, puts total semiconductor manufacturing equipment sales at $165.9 billion in 2026, up 23.2% year over year, rising to a record $229.5 billion in 2028, with wafer fab equipment reaching $143.9 billion next year, test equipment growing 31% and DRAM equipment jumping 39%. Three days later, the Philadelphia SE Semiconductor index closed 20% below its record and entered a bear market. The forecast came first, so it is not a response to the selloff, which puts the gap between order books and market sentiment at the center of the picture.

Illustration for the TSMC Arizona investment story
chipsinfrastructure

TSMC added $100 billion in Arizona, lifting US plans to $265 billion

On July 16, 2026, the White House and the Commerce Department announced that TSMC will invest an incremental $100 billion in advanced chip manufacturing and packaging in Arizona, bringing its total planned US investment to a record $265 billion and its US footprint to 12 sites. The same day TSMC posted its fifth straight record quarter with net income of NT$706.56 billion, up 77% year over year. The $265 billion is planned rather than spent, with risk production targeted for 2027 and mass production in 2028.