449 articles
MediaTek's 2nm AI chip cuts power 61% while running 30B-parameter models locally; Ambarella's modular X7 adds ARM AI to existing hardware at just 5 watts.
Groq's deterministic LPU chip predicts power demand to the clock cycle, cutting wasted voltage by 60% and delivering 35x more AI inference throughput per.
China has closed SK Hynix's HBM technology lead from five years to three, backed by an 873% revenue surge and an 82% operating margin.
NVIDIA is folding d-Matrix's inference chips into its MGX rack architecture, a rare move that targets a market projected to hit $546 billion by 2030.
AI data centers face multi-year grid delays, not power shortages; mobile battery storage lets operators go live in weeks while permanent upgrades queue up.
Trump dismisses AI safety warnings as a "SICK conspiracy" while Anthropic and OpenAI push for slowdowns; the Nasdaq dropped 0.8% amid the clash.
AI data centers are skipping years of permits by converting old industrial mills, using their existing power, water, and zoning to deploy faster.
DeepSeek will deploy 160,000 Huawei Ascend chips at a new data center, building a self-sufficient AI ecosystem designed to operate without Nvidia hardware.
Bloom Energy stock has hit Wall Street's exact consensus target at $276 just before its S&P 500 entry on September 21, raising the question: floor or.
A CBP supervisor stole over $105,000 in Intel chips from DHS computers, swapping them for cheaper parts and cashing out through Newegg's trade-in program.
Positron AI hit a $5 billion valuation by ditching HBM for laptop memory in its inference chips, a contrarian bet backed by $875 million in fresh funding.
Cerebras holds a $25.4 billion AI chip backlog, yet Wall Street favors IonQ for 2026 thanks to 287% revenue growth and a cleaner balance sheet.
The UAE is restructuring its $30 billion Stargate AI project, ditching a single Abu Dhabi campus for distributed data centers with air defense systems.
SpaceX's Starmind project will place up to 1 million AI satellites in orbit by 2028, using solar power and space cooling to slash data center energy costs.
71% of Americans oppose data centers near their homes, surpassing even nuclear plant fears, as politicians nationwide rush to capitalize on the backlash.
The data center construction market is set to reach $662 billion by 2033, driven by AI's unique demands for power, cooling, and specialized infrastructure.
Electricity, not chips, now limits AI growth; Fluidstack breaks ground on a $4B Texas data center campus as Microsoft plans to triple capacity to 38.
Nvidia's real AI dominance may lie in its rack infrastructure, not its chips, as even rivals like Groq and d-Matrix build around its architecture.
AMD's new Ryzen 5 5500F at $99 is 38% cheaper than the 5600, giving budget PC builders a rare win while memory prices stay high.
Samsung is partnering with Mistral AI to use artificial intelligence in semiconductor manufacturing, targeting defect detection and yield gains as chip.
The DOJ is investigating whether Nvidia's $20 billion Groq deal was structured as a license to dodge antitrust review that a direct acquisition would.
The DOJ is investigating Nvidia's $20 billion Groq deal, questioning whether the licensing agreement was structured to dodge federal merger notification.
Apple's M5 Ultra and M6 chips can run large AI models locally on your Mac, with 512GB unified memory and no cloud subscription required.
Google is locking in 450 megawatts of nuclear power for 22 years and spending €13 billion in Finland to secure AI's most critical resource: electricity.
On-device AI speed depends more on unified memory bandwidth than processor power, and a 4,096-token chat can silently consume 512 MB of RAM.
Google's €13 billion Finland deal secures nuclear power for AI data centers through 2048, marking its first nuclear energy agreement outside the United.
AI data centers could generate $200 billion in insurance premiums by 2030, but insurers warn they lack the models to price risks for $50 billion campuses.
Google is investing $15.1 billion in Finland AI data centers, securing nuclear power and creating up to 37,000 jobs in Europe's emerging AI hub.
Qualcomm's new Snapdragon chip could cut mobile AI's RAM demands with a redesigned NPU, enabling smarter on-device AI without cloud dependency.
AI data centers face a $660 billion buildout, but the real bottleneck is power: the US grid holds 2,600 gigawatts in pending requests, double its total.
China's AI chips just broke the export control thesis: Zhipu's 320B-parameter model trained entirely on Huawei silicon triggered a $1 trillion US chip.
AMD raised its AI market forecast to $3 trillion by 2030, but independent analysts estimate just $1.2 trillion, raising questions about whether the.
Spain is emerging as Europe's next AI data center hub, with a $1 billion hyperscale campus planned and over 1 gigawatt of capacity in the pipeline.
Google's 25-year nuclear power deal is financing the restart of Iowa's Duane Arnold plant, showing how AI data centers are reviving dormant reactors.
Arm's CSS for Mobile 2 platform will run AI agents directly on phones with a 70% speed boost, but consumers won't see devices until 2027.
India's data center capacity hit 2.1GW as US projects stall over power and community backlash, drawing $57 billion from Amazon, Microsoft, and Google.
AI data centers may outpace power grid capacity within years, making energy access, not chips, the true bottleneck in the global AI race.
Unified memory architecture lets consumer hardware run 200B+ parameter AI models locally; Apple, AMD, and Nvidia are now racing to lead on capacity and.
New York plans to nearly triple its nuclear capacity to 8.4 gigawatts, spending up to $23.9 billion to power AI data centers with carbon-free electricity.
SK Hynix trades at a forward P/E of 5 with aggressive buybacks, making it the defensive pick over Micron for AI memory exposure.
Chip stocks historically stumble in September, but AI semiconductor ETFs gained up to 12% last September, proving strong earnings can override seasonal.
Meta's new MTIA 400 AI chip delivers 46 times more memory bandwidth than its predecessor, targeting recommendation systems and generative AI inference at.
Hanwha Qcells is sending perovskite-silicon tandem solar cells to the Moon to stress-test next-gen technology that could power AI data centers on Earth.
AI data centers are getting less thirsty, but cutting water use often means burning more electricity, creating a hidden trade-off companies rarely.
HQVT's thermal AI platform detects data center electrical fires before ignition, with revenue surging 84.5% as edge AI cuts deployment costs for smaller.
Elon Musk built everything in-house except three suppliers he couldn't beat: TSMC, ASML, and Nvidia's CUDA ecosystem hold moats even he can't dig around.
A 1960s nuclear technology could displace AI data centers by delivering cheaper, cleaner power, protecting communities from dirty electricity and rising.
Virginia's 665 data centers face a wave of moratoriums, zoning rules, and a new statewide electricity tax as cities push back against the AI.
Gimlet Labs hit a $3 billion valuation after raising $300 million, betting that routing AI inference across multiple chip types beats single-GPU serving.
64GB of unified memory and 250GB/s bandwidth are becoming the new baseline for local AI, letting developers run 30B-parameter models without costly cloud.
