AI Compute

Twitter

Facebook

Copy Link

Key Facts

Key PlayersNVIDIA, Microsoft, Google, AWS, CoreWeave

Top News

Intel Starfire brings 18A process to space with radiation-hardened chip for US government

Intel has unveiled Starfire, a space-grade system-on-chip built for the US government that combines eight CPU cores and NPU on its 18A process with an Intel 3 GPU. The radiation-hardened chip operates between -55°C and 125°C, delivering up to 75 TOPS for AI workloads in orbit with samples expected in Q3 2026.

Technology2 sources

Intel Starfire brings 18A process to space with radiation-hardened chip for US government

Technology2 sources

Intel announces $5.7 billion investment in Ireland to expand AI chip manufacturing capacity

Intel is committing €5 billion to expand its Leixlip campus in Ireland, representing roughly 30% of the company's $17 billion capital expenditure planned for 2026. The investment targets data center processors for AI and high-performance computing, upgrading Fab 34—one of Europe's few EUV-capable facilities—while adding several hundred jobs to Intel's 4,900-strong Irish workforce.

Technology8 sources

Intel announces $5.7 billion investment in Ireland to expand AI chip manufacturing capacity

Technology8 sources

AI costs surge as usage-based pricing catches companies off guard, budgets explode

Companies are discovering that AI spending has become a financial management crisis rather than a cost-saving solution. A KPMG survey of 2,145 executives worldwide reveals one-third have limited understanding of their AI costs. The shift from flat-rate to usage-based pricing models has turned predictable software budgets into unpredictable expenses that grow with every model interaction and token consumed.

Business2 sources

AI costs surge as usage-based pricing catches companies off guard, budgets explode

Business2 sources

Monzo founder Tom Blomfield joins Anthropic to tackle AI's compute challenge

Tom Blomfield, who built Monzo and GoCardless into companies worth over $9bn combined, is leaving Y Combinator to join Anthropic's compute team. The move signals how critical compute infrastructure has become as AI companies race to deploy massive chip capacity and train next-generation models.

Business3 sources

Monzo founder Tom Blomfield joins Anthropic to tackle AI's compute challenge

Business3 sources

Samsung's 2nm process lands Tesla AI5 chip at Texas fab, marking foundry breakthrough

Samsung has completed the Tesla AI5 chip design and will manufacture it at its Taylor, Texas facility using its latest 2-nanometer process technology. The development suggests Samsung's 2nm yields have exceeded 60%, a critical threshold that could help the company compete with TSMC and attract more advanced AI chip customers.

Technology3 sources

Samsung's 2nm process lands Tesla AI5 chip at Texas fab, marking foundry breakthrough

Technology3 sources

TSMC expands advanced chip packaging capacity in Taiwan to meet surging AI demand from Nvidia

TSMC is building two additional advanced chip packaging plants at its Chiayi Science Park in southern Taiwan, responding to relentless demand from AI chip designers like Nvidia. The expansion will create over 9,000 jobs and generate $9.35 billion in annual production value. The move transforms a small Taiwanese city into one of the world's largest advanced packaging hubs, addressing the global AI chip supply bottleneck.

Business2 sources

TSMC expands advanced chip packaging capacity in Taiwan to meet surging AI demand from Nvidia

Business2 sources

Apple's M7 Ultra chip could pack 1.5TB of memory to rival Nvidia Blackwell by 2028

Apple is developing the M7 Ultra with up to 1.5TB of unified memory, designed to compete with Nvidia's Blackwell AI accelerators. Expected in 2028, the chip represents a fundamental shift in Apple's silicon strategy, prioritizing AI performance over traditional metrics. But memory shortages and potential pricing could make this the most expensive Mac ever produced.

Technology4 sources

Apple's M7 Ultra chip could pack 1.5TB of memory to rival Nvidia Blackwell by 2028

Technology4 sources

AI demand called 'unlimited' by executives even as chip stocks stumble on market expectations

AI executives insist demand is bottomless, with Lumentum reporting products sold out five years ahead and Pat Gelsinger calling energy the only real constraint. Yet chip stocks keep falling despite strong fundamentals. The disconnect stems from sky-high valuations pricing in flawless execution, while enterprises shift from tokenmaxxing to valuemaxxing, focusing on return on investment over unchecked AI spending.

Business2 sources

AI demand called 'unlimited' by executives even as chip stocks stumble on market expectations

Business2 sources

Memory chip shortage set to worsen through 2027 as AI demand outpaces supply until 2030

The AI boom has shattered the memory market's predictable boom-bust cycle, sending prices soaring when they should be falling. SK Hynix CEO warns 2027 will be the worst year ever for memory supply shortages, with relief not expected until 2028. The big three memory makers—SK Hynix, Samsung, and Micron—are investing hundreds of billions in new fabs, but production won't ramp fast enough to meet AI infrastructure demand that's expected to exceed capacity well into the next decade.

Business5 sources

Memory chip shortage set to worsen through 2027 as AI demand outpaces supply until 2030

Business5 sources

Elon Musk admits he was wrong about Anthropic, praises rival as current AI leader

Elon Musk publicly reversed his stance on Anthropic, calling the AI startup the current industry leader and praising its Mythos and Fable models. The admission marks a dramatic shift from his previous criticism, coming as Anthropic pays SpaceX $1.25 billion monthly for access to the Colossus 1 supercomputer. Musk pledged not to weaponize his infrastructure control against the competitor.

Technology7 sources

Elon Musk admits he was wrong about Anthropic, praises rival as current AI leader

Technology7 sources

Palo Alto CEO Nikesh Arora demands 90% drop in AI token costs for mass enterprise adoption

Palo Alto Networks CEO Nikesh Arora says AI token costs need to plummet by as much as 90% before businesses can deploy the technology at scale. While OpenAI's 54% efficiency improvement is welcomed, Arora argues it's nowhere near enough. His warning reflects a growing frustration among enterprise leaders as total AI bills triple despite per-token prices falling 98%, driven largely by agentic AI usage.

Business4 sources

Palo Alto CEO Nikesh Arora demands 90% drop in AI token costs for mass enterprise adoption

Business4 sources

AMD Zen 6 processors to debut at Advancing AI event on July 22, targeting enterprise AI workloads

AMD officially confirms its next-generation Zen 6 processors will launch at the Advancing AI 2026 conference on July 22-23 in San Francisco. CTO Mark Papermaster revealed the EPYC server CPU lineup built on TSMC's 2nm process will scale up to 256 cores and power Helios AI racks, positioning AMD to compete directly with NVIDIA's dominant AI infrastructure.

Technology2 sources

AMD Zen 6 processors to debut at Advancing AI event on July 22, targeting enterprise AI workloads

Technology2 sources

AI Data Centers Exploit Texas Loopholes as Pollution Reaches Catastrophic Levels

Texas has become ground zero for AI data center expansion, with companies exploiting regulatory loopholes to build massive fossil fuel-powered facilities. At least 15 gas plants tied to data centers are planned, potentially emitting more than 130 million tons of greenhouse gases annually—equivalent to 35 coal-fired power plants. Residents like Omaira Garcia discover construction only after it begins, with exhaust stacks appearing 500 yards from homes.

Policy2 sources

AI Data Centers Exploit Texas Loopholes as Pollution Reaches Catastrophic Levels

Policy2 sources

Apple in talks with PrismML to run massive AI models directly on iPhone without cloud servers

Apple is exploring partnership opportunities with PrismML, a Caltech spinout that has successfully compressed Alibaba's 27-billion-parameter Qwen 3.6 model to run entirely on an iPhone 17 Pro. The breakthrough could allow Apple to shift heavyweight AI processing on-device, enhancing user privacy while dramatically reducing cloud infrastructure costs.

Technology4 sources

Apple in talks with PrismML to run massive AI models directly on iPhone without cloud servers

Technology4 sources

SK hynix and TetraMem unveil in-memory computing chip that processes AI directly inside memory

SK hynix and TetraMem have developed an experimental analog in-memory computing chip that performs AI calculations directly within memory arrays, achieving 21.3 TOPS/W energy efficiency. The memristor-based system-on-chip targets edge AI devices by eliminating data movement bottlenecks, though performance remains limited at 2.54 TOPS peak throughput.

Technology3 sources

SK hynix and TetraMem unveil in-memory computing chip that processes AI directly inside memory

Technology3 sources

Prev12 3 4…73 Next