Microsoft unveils Maia 300, aims to run inference from 2027
Microsoft says its Azure Maia chips will scale to multi-gigawatt deployments from 2027, with the third-generation Maia 300 becoming its primary inference system.
AI Chips News and Analysis
59 verified stories covering AI Chips, product updates and industry developments.
Former Groq Engineers Sue Board Over Nvidia Deal Payout
Two former Groq engineers are suing the company's board in Delaware, alleging a roughly $20 billion Nvidia deal favored insiders over ordinary shareholders.
Google's First TPU Satellite Operating Normally in Orbit
Google's first Project Suncatcher prototype satellite is in orbit, running Trillium TPUs on solar power to test space-based AI compute.
Tencent Leases 100,000 AI Chips From Oracle in $7B, Five-Year Deal
Tencent has signed a five-year, roughly $7 billion lease with Oracle for about 100,000 advanced AI chips housed in Southeast Asia, per a new report.
Google's TPU v7 Beats GB200 by 57% Running Kimi K3
Inferact's open-source megakernel lets 16 Google TPU v7 chips decode Kimi K3 57% faster than 16 Nvidia GB200 chips, even though GB200 leads on paper specs.
T-Head's Zhenwu V900 AI Chip Debuts, Mass Production Delayed to 2027
Alibaba's T-Head unveiled the Zhenwu V900 AI chip with 3x the compute of its predecessor, but mass production won't begin until 2027.
AMD trails Nvidia by up to 14.8x on DeepSeek's newest model
AMD shipped its DeepSeek V4.1 Flash image two days after CUDA's vLLM, and SemiAnalysis says it delivers up to 14.8x worse performance per dollar than H200, and up to 42x worse than B200 and B300.
OpenAI Used Its Own AI Models to Design a Chip in 9 Months
OpenAI's CFO says in-house AI models helped design its Jalapeno inference chip in just 9 months, as Luna's price cuts fuel a wider AI price war.
TSMC and ASML Push Photomasks From 6 to 12 Inches
TSMC and ASML will shift High NA EUV photomasks from 6 inches to 12 inches, aiming to eliminate stitching, raise throughput, and cut costs by 2033.
Hanxu Chips Keep AI Weights On-Die, Target 24TB/s Bandwidth
Peking University-linked startup Hanxu Technology unveils uHBM/uLPU chips that keep AI model weights on-die for 24TB/s readout bandwidth, but the approach caps models at around 4B parameters.
DRAM Price Surge Pushes CXMT's Profit to RMB77.6bn
CXMT's first half-year report shows revenue up 873.64% and net profit at RMB77.605 billion, as a DRAM price surge driven by AI data centers' HBM demand squeezes standard memory supply.
NVIDIA's $96.2B Quarter Comes With a $63B Receivables Jump
NVIDIA's Q2 FY27 revenue hit $96.2 billion with $108 billion guidance ahead, but receivables jumped 64% as payment terms stretched from 45 to 60 days.
GPT-5.6 Sol Designed a CPU, and It Runs DOOM
GPT-5.6 Sol built an RV32IM CPU gate by gate inside a puzzle-game sandbox, named it Codex-R32 itself, and got 1993's DOOM running on it — a serious test of cross-layer design consistency.
Nvidia's New Edge Module Hits 78 TOPS, Ships in 2027
Nvidia's Jetson Orin Nano 2 doubles edge inference at 78 TOPS with 40% lower power draw, but the module and dev kit won't ship until the first half of 2027.
OpenAI Shows Jalapeño Benchmarks, Beats GB300 at 700 Watts
OpenAI's Jalapeño inference chip claims up to 1.9x the throughput per kilowatt of Nvidia's GB300, running at less than half the rated power.
NVIDIA claims Vera Rubin NVL72 boosts throughput 30x per megawatt
NVIDIA's early tests show Vera Rubin NVL72 hitting 30x GB300 NVL72's per-megawatt throughput on agentic AI workloads, with independent review from SemiAnalysis still pending.
H20 Runs DeepSeek V4-Pro at 70% of B300 Speed
LMSYS's DeepSeek-V4-Pro benchmarks show Nvidia's China-market H20 chip closing the decode-speed gap with the flagship B300 to just 1.42x.
Amazon’s AI Run Rate Tops $25B
Amazon’s Q2 report showed AWS AI and chip businesses each above a $25 billion annual run rate, with a large Anthropic-related gain lifting profit.
AMD Gives Claude a 2GW Chip Roadmap
AMD will supply Anthropic with up to 2GW of MI450 GPUs and has committed up to $5 billion in future equity investment.
Moore Threads' Half-Year Revenue Tops 2025 Total
Moore Threads expects first-half 2026 revenue of RMB 1.65 billion to RMB 1.75 billion, putting delivery and repeat demand at the center of China's GPU story.
SambaNova raises $1 billion at an $11 billion valuation
SambaNova raises $1 billion at an $11 billion valuation is reframed for global technology readers, with the key numbers, product claims and open questions kept intact.
DeepSeek reportedly develops its own inference chip
DeepSeek reportedly develops its own inference chip is reframed for global technology readers, with the key numbers, product claims and open questions kept intact.
China’s AI chip shortage turns into a capacity problem
China’s AI chip shortage turns into a capacity problem is reframed for global technology readers, preserving the key numbers, claims and open questions.
OpenAI Prices GPT-5.6 in Three Tiers
OpenAI split GPT-5.6 into Sol, Terra and Luna, using Sol for top performance, Terra for mainstream work and Luna for cheaper high-volume use.
Keller says Tenstorrent is not for sale
Jim Keller denied that Tenstorrent is in acquisition talks with Qualcomm, while leaving the door open to a large partnership around RISC-V AI chips.
South Korea puts $518B behind four AI chip clusters
South Korea puts $518B behind four AI chip clusters
Nearfield raises $380M for AI-chip metrology
Nearfield raises $380M for AI-chip metrology
Upscale raises $190M for AI networking chips
Upscale AI raised $190 million at a $2 billion valuation, with Nvidia joining the round, as its SkyHammer switch chips pitch open Ethernet standards for large AI clusters.
Samsung gets a second look as TSMC fills up
AI chip demand has pushed TSMC capacity toward 2028, sending BYD, Google, AMD and Tesla to explore Samsung foundry capacity.
AMD-backed TensorWave raises 350 million dollars
TensorWave is building an AI cloud entirely around AMD Instinct GPUs, testing whether customers will pay for a serious Nvidia alternative.
SK hynix eyes a U.S. listing on HBM demand
SK hynix is preparing a U.S. ADR listing that could raise up to $14 billion as AI demand lifts high-bandwidth memory.
Bezos-backed Flourish raises $500 million for brain-like AI
The New York startup has no product or revenue yet, but raised at a $2.5 billion valuation on a bet that AI’s power problem needs brain-inspired computing.
Warren asks Nvidia’s Jensen Huang to testify on China chips
Senator Elizabeth Warren invited Jensen Huang to a June 11 Senate Banking Committee hearing on Nvidia’s China business and US export controls, with H200 shipments at the center.
AI chip stocks push India down market rankings
Taiwan and South Korea overtook India by market value as TSMC, Samsung and SK Hynix rallied on AI infrastructure demand.
SIA Says Chips Make Up 95% of an AI Server Rack’s Value
A Deloitte-SIA report projects AI data-center chip revenue could reach $1.2 trillion by 2028, reframing AI infrastructure as a full-stack semiconductor race.
Intel Bets Agentic AI Will Put CPUs Back at the Center
At Computex, Lip-Bu Tan framed Xeon 6+ and Intel 18A as a play for inference, agents and physical AI rather than GPU training.
Qualcomm launches Dragonfly to chase data center AI chips
Qualcomm used Computex 2026 to introduce Dragonfly, a data center brand covering CPUs, inference accelerators and ASICs, while leaving specifications for later.
Nvidia unveils RTX Spark, its first PC chip with 1 petaflop AI performance
Nvidia introduced the RTX Spark superchip at Computex, combining a 20-core Arm CPU and Blackwell GPU for Windows laptops and desktops, aiming to redefine local AI computing.
Nvidia's $20B 'Not-Acquisition' of Groq's Core Team and Technology
Nvidia pays $20 billion for Groq's LPU chip technology and hires its CEO and core engineering team, while the remaining company pivots to an AI inference cloud startup called Groq2.
Huawei Shifts to Folded Chip Design, Kirin to Debut Late 2025
Huawei unveils Tau Scaling Law and LogicFolding architecture to bypass EUV lithography restrictions, aiming for equivalent 1.4nm transistor density by 2030.
Broadcom, Meta, and Partners Invest $125M in UCLA Semiconductor Hub
Five companies including Broadcom and Meta jointly invest $125 million over five years to establish a semiconductor R&D center at UCLA, focusing on next-gen AI chips and workforce training.
AMD Packs 192GB Memory Into APU, Runs 300B LLMs Locally
AMD's Ryzen AI Max 400 'Gorgon Halo' APU supports up to 192GB unified memory, enabling local inference of 300B+ parameter models on a single chip.
Alibaba's AI writes its own chip driver, achieving 10x speedup
Alibaba demonstrated its Qwen3.7-Max model autonomously writing kernel drivers for its new Zhenwu M890 chip, achieving a 10x performance boost in 35 hours.
Nvidia Posts $81.6B Quarterly Revenue, Authorizes $80B Buyback
Nvidia reported Q1 FY27 revenue of $81.6 billion, beating estimates, and announced an $80 billion share repurchase authorization.
Intel and Qualcomm Both Eye Tenstorrent, Valuation Over $5B
Intel and Qualcomm are reportedly in talks to acquire AI chip startup Tenstorrent, which could be valued at over $5 billion.
US clears H200 chips for China, but Beijing stalls the orders
Washington reportedly approved H200 sales to about ten Chinese firms, yet Commerce Secretary Howard Lutnick said China is steering buyers toward domestic chips.
Fractile raises $220M to rethink inference chips
London startup Fractile raised a $220 million Series B led by Accel, Factorial Funds and Founders Fund to attack the inference bottleneck with a memory-centric chip design.
Cerebras lifts IPO range to $160 after 20x demand
Cerebras is raising its IPO price range and share count after orders reportedly reached more than 20 times the planned offering.
Micron surges as AI memory turns scarce
Micron's market value has pushed past $850 billion as AI demand locks up HBM supply and reframes memory as a strategic asset.
AMD Jumps 123% as Nvidia Gains 15%
AMD's first-quarter results pushed Wall Street to reconsider the idea that AI chips are only an Nvidia story.
Britain Pitches a Middle-Power Route Around AI Compute Giants
Liz Kendall says the UK will not outspend the five companies controlling most AI compute, and instead wants leverage through chips, standards and sovereign AI alliances.
DeepSeek V4 Only Released as Preview, Waiting for Huawei 950PR Mass Production
DeepSeek released V4 Flash and V4 Pro as preview versions on April 24, with the full release delayed until the second half of 2026 due to Huawei's Ascend 950PR chips not yet being ready for mass production.
Google splits TPU into two lines: training and inference
At Google Cloud Next, Sundar Pichai unveiled the eighth-generation TPU — not one chip, but two: the TPU 8t for training and the TPU 8i for inference, marking a strategic bet on specialized AI hardware.
OpenAI Bets $20 Billion on Cerebras
OpenAI commits over $20 billion to Cerebras for AI compute, with equity warrants and a $1 billion data center fund, as the battle shifts from training to inference. Nvidia counters by acquiring Groq for $20 billion.
Intel CEO invests in AI chip design startup Cognichip
Cognichip, a startup founded in 2024, has raised $60 million in new funding led by Seligman Ventures, with Intel CEO Lip-Bu Tan joining its board, aiming to slash chip development costs by over 75% and cut development time in half using a proprietary deep learning model trained on chip design data.
AMD MI400 with 432GB HBM4 takes on NVIDIA
AMD's upcoming MI400 series AI accelerators, led by the MI455X with 432GB HBM4 memory and 40 PFLOPS FP4 performance, challenge NVIDIA's dominance on paper, but software ecosystem hurdles remain.
Chip Giant Cerebras to List on Nasdaq
Cerebras Systems is targeting a Nasdaq IPO in the second quarter of 2026, aiming to raise $2 billion at a valuation between $22 billion and $25 billion. If successful, it would be the first pure-play AI chip alternative to Nvidia's GPU monopoly to go public.
Nvidia Blackwell Sold Out Through Mid-2026
Nvidia's B200 and GB200 NVL72 racks are sold out until mid-2026, with cloud giants holding over 3.6 million units in backlog. The article examines supply chain bottlenecks, the upcoming Blackwell Ultra (GB300), and what this means for AI companies.
DeepSeek V4 Runs on Huawei Chips
DeepSeek is set to release V4, a trillion-parameter MoE model running on Huawei's Ascend 950PR chips, marking the first frontier AI model built entirely for a domestic Chinese chip architecture and a potential shift away from NVIDIA's CUDA ecosystem.