ai-chips

TechCrunch reports on the growing trend of major AI companies like OpenAI and SpaceX developing custom chips to reduce reliance on Nvidia, featuring a podcast discussion with hosts Kirsten Korosec, Anthony Ha, and Sean O'Kane.

The article argues that current high LLM pricing is unsustainable due to diminishing performance gains, the rise of open-weight models, specialized AI chips reducing inference costs, and zero switching costs, predicting significant price drops as competition intensifies.

Apple plans to skip the high-end M6 Mac chips and instead launch an AI-focused M7 line, including M7 Pro, M7 Max, and M7 Ultra.

Nvidia's AI chips are selling at record high prices in China due to US export restrictions, while the company also announced a new liquid-cooling system to reduce data center water usage.

The Chip Security Act, which would mandate location-tracking mechanisms for advanced AI chips to prevent diversion to China, has gained support from six companies specializing in tracking sensitive shipments, while opponents argue it could hamper U.S. chip sales.

Former White House AI advisor Dean Ball argues that China's efforts to achieve AI chip independence are largely performative and not substantive.

ASML is shipping its new $400 million high-NA EUV lithography machine, which can pattern features as small as 8 nanometers, crucial for advancing Moore's Law and meeting AI industry demand for denser, more powerful chips.

JPMorgan releases ASIC industry report, predicts AI custom chips entering golden cycle, Broadcom and Marvell are biggest beneficiaries, and expects AI ASIC shipments to surpass GPU for the first time by 2027.

Google is adopting Nvidia's strategy to build a competitive AI chip business, renting TPU computing power to Anthropic and boosting inference performance to rival Nvidia's dominance.

Amazon Web Services is in early talks to sell its Trainium AI chips to other companies, directly challenging Nvidia's dominance. CEO Andy Jassy noted that if they sold chips externally, it could represent a $50 billion run rate business.

The article recounts Baidu Research US's investment in Cerebras, a wafer-scale chip company, a decade ago. It analyzes the shift in the AI chip market from training to inference and the importance of non-consensus investments.

Tensordyne announced a breakthrough inference system using logarithmic math in hardware, claiming 17x more tokens per watt and 13x higher throughput than NVIDIA Blackwell, achieved by replacing complex multiplication with simple addition in log space.

Google is in talks with Samsung to manufacture a component of its next-generation AI chip, codenamed Icefish, using 2-nanometer technology, while the main part will be made by TSMC. The chip aims to offer an alternative to Nvidia's GPUs and is expected to enter mass production as soon as 2028.

Google is backing Anthropic's $35 billion chip lease at five data centers, revealing complex financial alliances in the AI industry.

Groq is raising $650M despite its technology licensing to Nvidia because the corporate entity retained its datacenter operations and inference API, focusing on fast small-model inference.

The US Commerce Department closed a loophole that let Chinese AI companies purchase advanced Nvidia and AMD chips through overseas subsidiaries, extending export-licence rules to cover entities headquartered in China regardless of physical location. The guidance targets future shipments and does not affect existing hardware.

Microsoft and NVIDIA are collaborating to integrate Grace Blackwell chips into laptops, aiming to challenge Apple's six-year dominance with Apple Silicon.

BIS issued guidance requiring licenses for advanced AI chip exports to Chinese-headquartered firms located outside China, highlighting previous enforcement gaps for overseas subsidiaries like Tencent Malaysia buying Nvidia Blackwell chips.

AI chip startup Groq is reportedly raising $650M from existing investors to grow its inference cloud business, following a $20B technology licensing deal with Nvidia.

TSMC's senior VP says energy efficiency is now the primary constraint in AI chip design, surpassing raw computing power. The shift is driving changes in transistor density, advanced packaging, and chip stacking to reduce power consumption.