Google plans new chip to run Gemini models more efficiently, the Information reports ↗
Google is reportedly developing a server chip that embeds parts of Gemini directly into the hardware. The project, informally called “Frozen v2”, is designed to serve AI responses while using far less computing power.
The chip could deliver six to ten times more tokens per unit of electricity than Google’s latest custom AI hardware. That sounds rather wild... though the design is still being finalised.
The project would sit alongside Google’s existing TPUs rather than replace them, as the company grapples with an AI capacity crunch that has reportedly forced Google Cloud to turn away some customer deals.
China's Moonshot pauses Kimi subscriptions amid hot demand, IPO push ↗
Moonshot AI temporarily stopped accepting new Kimi subscriptions after demand for its K3 model overwhelmed the computing capacity available.
K3 carries 2.8 trillion parameters, and its popularity has exposed the awkward reality behind powerful models - clever software still demands an enormous heap of chips and electricity.
The Chinese startup is also seeking fresh capital and exploring a Hong Kong listing. It has reportedly discussed the plan with financial advisers after raising billions from investors.
US judge approves Anthropic's $1.5 billion settlement of copyright lawsuit ↗
A federal judge approved Anthropic’s $1.5 billion settlement with authors who accused the company of using pirated books while developing Claude.
It is the largest known payout in a US copyright case and the first major AI training lawsuit to reach a settlement. More than 91% of eligible authors and publishers have claimed a share.
The court had previously found that training AI on books could qualify as fair use, while separately criticising Anthropic’s storage of millions of pirated titles. A legal line drawn in pencil, to put it mildly.
Hut 8 signs $9.8 billion AI data center lease, fully commercializes Texas campus ↗
Former crypto miner Hut 8 signed a second 15-year AI data-centre lease worth $9.8 billion with an unnamed investment-grade customer.
The agreement covers 352 megawatts and doubles the customer’s contracted footprint at Hut 8’s Texas campus. The site’s base contract value now stands at $19.6 billion - electricity has become the new oil barrel, more or less.
Hut 8 redesigned its first data hall around Nvidia’s architecture, increasing capacity by 57% without adding more land or utility access.
Tempus to buy cancer test maker Personalis for $1.5 billion ↗
AI-powered precision-medicine company Tempus agreed to acquire Personalis in a deal valued at roughly $1.5 billion.
The purchase gives Tempus control of NeXT Personal, a blood test that creates a genetic signature for each patient’s cancer and searches for minute traces left behind after treatment.
Those traces can be too small for conventional scans to detect. The combination strengthens Tempus’s oncology business, though investors reacted cautiously to the deal’s stock-heavy structure.
Could AI chip boom make ASML Europe's first trillion-dollar firm? ↗
AI chip demand has pushed ASML close to a $700 billion valuation, raising the once faintly bonkers possibility that it could become Europe’s first trillion-dollar company.
ASML is the sole supplier of extreme-ultraviolet lithography machines needed to manufacture the world’s most advanced chips. Its flagship systems are nearly sold out, which is quite a queue.
The biggest uncertainty is whether cloud giants will sustain their immense data-centre spending. Still, several analysts now see a trillion-dollar valuation as plausible... or so it appears.
FAQ
Why is Google developing a new chip for Gemini models?
Google is reportedly developing the “Frozen v2” server chip to serve Gemini responses while using less computing power. The design could deliver six to ten times more tokens per unit of electricity than Google’s latest custom AI hardware. Rather than replacing Google’s TPUs, it would complement them and help ease the mounting strain on AI computing capacity.
Why did Moonshot AI pause new Kimi subscriptions?
Moonshot AI temporarily stopped accepting new Kimi subscriptions after demand for its K3 model outpaced the computing capacity available. The pause underscores a persistent challenge in generative AI: even sophisticated software depends on vast quantities of chips, electricity and data-centre infrastructure. Moonshot is also reportedly raising capital and considering a possible Hong Kong listing.
What does Anthropic’s copyright settlement mean for AI training?
Anthropic’s $1.5 billion settlement resolves claims that pirated books were used during the development of Claude. The court had indicated that training an AI model on books could qualify as fair use, while treating the storage of pirated copies as a separate matter. The case therefore draws a clear legal distinction between using content for model training and obtaining or retaining that content unlawfully.
Why are AI data-centre leases becoming so valuable?
AI models require immense computing power, electricity and specialised infrastructure, making large data-centre campuses increasingly valuable. Hut 8’s new 15-year lease covers 352 megawatts and carries a reported value of $9.8 billion. The company also redesigned a data hall around Nvidia’s architecture, expanding capacity without requiring additional land or utility access.
How could the Personalis acquisition help Tempus?
Acquiring Personalis would give Tempus control of NeXT Personal, a blood test designed to detect minute traces of cancer after treatment. The test creates a genetic signature for an individual patient’s cancer, then searches for remaining tumour material that conventional scans may miss. The acquisition is intended to strengthen Tempus’s precision-medicine and oncology operations.
Could the AI chip boom make ASML a trillion-dollar company?
ASML could benefit because it is the sole supplier of extreme-ultraviolet lithography machines used to manufacture the most advanced chips. Strong demand for AI chips has pushed its valuation close to $700 billion, while its flagship systems are nearly sold out. Reaching a trillion-dollar valuation would depend heavily on whether major cloud companies continue making vast investments in data centres and semiconductors.