Google is now supplying its tensor processing units (TPUs) to companies it competes with, including Anthropic and Apple. On Alphabet's third-quarter earnings call, management gave two reasons: outside customers keep the chips competitive, and outside volume lowers the cost of every chip Google uses itself. However, it would not say how much money the business makes.
Marvell is reportedly in advanced talks to acquire Lumentum, a major laser supplier, in a deal that could be announced as early as next week if no other technology giant intervenes. Based on Lumentum's current market capitalization of approximately US$83 billion, the transaction would be Marvell's largest acquisition ever.
Oracle will offer cloud services powered by Cerebras' AI computing systems in its cloud data centers starting in 2027, giving the wafer-scale chipmaker another major cloud customer. Although the companies did not disclose Oracle's planned Cerebras hardware purchases, the deal highlights the shift toward diversified AI compute resources as inference demand accelerates.
Nvidia CEO Jensen Huang's visit to Kazakhstan highlights a widening contest over AI computing capacity, as Europe seeks alternatives to infrastructure concentrated in the US and Asia. Kazakhstan, with 21 million people, is developing data centers and pursuing Nvidia partnerships that could position it as a supplier to Europe.
Surging demand for AI computing is driving advanced packaging toward larger form factors, according to Ting-Hao Lin, R&D vice president at Kinsus Interconnect Technology, speaking at the 2026 TPCA Show. To support this trend, substrate materials must deliver low warpage and exceptional dimensional stability while overcoming challenges in thermal expansion mismatches and structural stress.
The robotics industry has come a long way since robots were first built starting in the mid-20th century. Yet LEACHIM, a robot teacher and one of the world's first humanoids created back in the 1970s, may hold a key clue as companies seek to transition their technological advancements into on-the-ground deployment: develop machines that solve real-world problems.
Concerns over a potential AI bubble are intensifying. From large language models (LLMs) and AI data centers to humanoid robots, investors are increasingly questioning when massive capital investments will translate into corresponding revenue and profits.
Nvidia has won a CPU supply order from AI startup Modal, challenging Intel and AMD in the server CPU market as it expands beyond GPUs into general-purpose computing. Although Modal's annual revenue is below US$1 billion, and neither the order's value nor its deployment scale has been disclosed, the deal underscores a broader shift in AI infrastructure competition toward full-stack system integration.
Invasive brain-computer interfaces (BCIs) are demonstrating long-term commercial potential. Neuralink has enabled paralysed patients to control computers and robotic arms through thought; Synchron's iOS integration lets patients operate iPads; and Science combines implanted chips with smart glasses to help visually impaired patients recognise text again. Potential applications extend to unmanned vehicles and defence.
As AI data centers accelerate construction, supply-chain competition is expanding beyond servers and fully integrated racks to the integrated delivery of computing modules and site engineering. Nvidia unveiled its DSX POD reference design based on the Vera Rubin platform at GTC Berlin, with partners installing, integrating, and validating equipment inside containerized modules before transporting them to customer sites for final assembly and operation.
Liquid cooling in AI data centers is moving toward more granular control architectures. Patrick Zeng, general manager of data center thermal management at Texas Instruments (TI), told DIGITIMES that as next-generation GPU computing power and rack power continue to rise, thermal management will shift from centralized control by a CDU to independent control for each server tray, a shift that will increase the use of sensors, drivers, MCUs, and other components within each rack.
Tesla unveiled its new Roadster, equipped with rocket-style thrusters, on October 15 and announced plans to begin production in 2028, more than a decade after the second-generation prototype was first showcased.
Tesla's move into early production of its Optimus humanoid robot has exposed an unusual challenge in scaling embodied artificial intelligence (AI): the company may need to manufacture thousands of robots before it can generate enough training experience to make them commercially useful.
DIGITIMES observes that as AI servers and high-speed networking equipment continuously pursue higher data transmission rates, printed circuit boards (PCBs) are being upgraded through higher layer counts and lower-loss materials. In this transition, via structures have become critical bottlenecks affecting signal integrity. When high-frequency signals route between layers in a multilayer board, residual barrel copper can impair transmission, generating robust demand for high-precision backdrilling equipment.
China is proposing to extend AI technology export controls to chip design, memory technologies, and large-scale computing clusters, introducing specific technical thresholds for each.
Chinese AI companies expanding overseas could face more clearly defined technology export restrictions under proposed regulations covering foundation models, AI agents, chip design, and large-scale computing clusters.
