Nvidia and Google launch AI energy alliance; SpaceX sets first orbital Starship flight

Nvidia, Google and Emerald AI form alliance for grid-flexible AI data centers

Emerald AI, Google and Nvidia announced the AI Energy Management Alliance (AEMA) on September 16. The coalition wants data centers to lower or shift power use when the grid is under stress in exchange for faster interconnection and lower system costs. Founding members include the three companies plus launch partners such as Anthropic, National Grid, AES, Constellation and NRG. AEMA cites a Brattle Group estimate that each gigawatt of flexible AI load could avoid about $733 million in power-system costs, and claims moderate flexibility could unlock up to 100 gigawatts of existing U.S. grid capacity. Nvidia and Emerald AI have completed flexibility demonstrations at several commercial sites and plan to bring a nearly 100-megawatt power-flexible AI factory online in Virginia later this year. Unite.AI / NVIDIA announcement

SpaceX targets September 22 for first orbital Starship flight

SpaceX announced on September 15 that it is aiming for September 22 to launch Starship Flight 14, the vehicle's first attempt to reach Earth orbit. A 75-minute launch window opens at 7:15 a.m. Central Time from the company's Starbase site in South Texas. The upper stage is expected to fly at roughly 275 kilometers altitude for about 10 hours and complete six orbits before splashing down in the Pacific Ocean west of Chile. The mission will also deploy 26 Starlink V3 satellites; SpaceX says each satellite will add 1 Tbps of capacity to the constellation. Three of the satellites carry cameras to inspect the heat shield during flight. The Super Heavy booster will attempt launch, stage separation, boostback and landing burns before an offshore splashdown in the Gulf of Mexico; SpaceX is not attempting a tower catch on this flight. Space.com

OpenAI in talks for pre-IPO funding at about $1.2 trillion valuation

OpenAI is in early discussions with large investors about a new private funding round that could value the company at roughly $1.2 trillion, according to a September 15 Financial Times report cited by Reuters. The talks were initiated by investors, not OpenAI, and the figure could change in coming months. OpenAI's last confirmed round, in March 2026, raised $122 billion at an $852 billion post-money valuation. Separately, CEO Sam Altman said on September 12 that OpenAI would not go public in 2026, citing AI safety concerns. Rival Anthropic is expected to begin marketing its IPO in mid-October. Reuters via The Star

Apple reportedly plans return to enterprise servers with M8 Ultra chips

Apple is developing an enterprise AI server that could mark its return to dedicated server hardware after discontinuing Xserve in 2011, The Information reported on September 16, citing people familiar with the project. The system is aimed at AI inference and would come in configurations with two or four planned M8 Ultra chips. Apple is said to be considering Nvidia's NVLink Fusion interconnect to link the processors. The product is not expected before 2029 and could still be canceled or launched without Nvidia technology. Apple and Nvidia did not comment. CNA / Reuters

MediaTek unveils Dimensity 9600 Pro, its first 2nm smartphone chip

MediaTek introduced the Dimensity 9600 Pro on September 15, calling it the company's first flagship smartphone processor built on TSMC's 2-nanometre process. The chip uses a 2+3+3 all-big-core CPU design with up to 4.55 GHz peak clock, 34.5 MB of cache and support for LPDDR6 memory and UFS 5.0 storage. MediaTek says CPU single-core and multi-core performance rise 17% and 15% respectively over the prior generation, while multi-core peak power drops 61%. A dedicated NPU enables on-device models up to 30 billion parameters. Vivo's X500 Pro series and Oppo's Find X10 Pro Max are among the first devices expected to use the chip. MediaTek

Intel CEO warns memory shortage will worsen, prices up five to seven times

Speaking at the AI Infrastructure Forum in Santa Clara, California, on September 15, Intel CEO Lip-Bu Tan said the global memory shortage will continue to worsen. He said memory capacity is very limited, many projects are being delayed because they cannot secure enough memory, and prices have risen five to seven times. Tan added that building mid-range and low-cost phones and laptops has become especially difficult. He also flagged power supply and cooling technology as major challenges and said Intel is investing in liquid and microfluidic cooling. Seoul Economic Daily

Google ships Gemini 3.8 Live speech-to-speech models

Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, describing both as native speech-to-speech models for real-time voice agents. The standard version is positioned for scale and cost efficiency; the Extended Thinking variant adds background multi-step reasoning and asynchronous tool calls while continuing the conversation. Google says Extended Thinking ranks first on Artificial Analysis' Speech-to-Speech Quality Index with a score of 82.6, achieves 68.6% on the τ-Voice agent benchmark and scores 97.7% on Big Bench Audio. The models support 97-plus languages and are available through the Gemini API and Google AI Studio, with enterprise access in private preview. Google Blog

South Korea launches $1 billion physical AI manufacturing program

South Korea's Ministry of Science and ICT formally launched the Gyeongnam-Jeonbuk AI Transformation R&D program on September 16. The government plans to invest 1.4131 trillion won, about $1.01 billion, from 2026 through 2030. Gyeongsangnam-do will focus on human-AI collaborative precision control using large action models trained on factory data, while Jeollabuk-do will build a software platform for coordinating robots and equipment from different makers. Thirteen manufacturers have agreed to share tacit production knowledge. The ministry aims to spread validated technologies nationwide and eventually export them as "K-manufacturing plants." Seoul Economic Daily