Meta just unveiled the next generation of its custom AI inference accelerator, MTIA. Built on TSMC's 5nm process, the new silicon more than doubles compute and memory bandwidth, delivering 3x the performance of MTIA v1 across key models. Why it matters: Meta is scaling its in-house chip lineup to slash inference costs and ease reliance on commercial GPUs.
Prime Minister Justin Trudeau just announced a $2.4 billion funding package to cement Canada's position in global AI. The bulk of the funds—$2 billion—will go toward expanding computing infrastructure and launching a sovereign compute strategy, alongside $50 million to create a new Canadian AI Safety Institute. Why it matters: As the race for AI compute tightens worldwide, Canada is doubling down on public hardware investment to keep local talent and startups from heading abroad.
Microsoft and OpenAI are reportedly plotting a massive data-center project featuring an AI supercomputer dubbed "Stargate," which could cost up to $100 billion. The system would pack millions of chips, demand gigawatts of power, and potentially launch as early as 2028. It highlights their massive bet on compute, though Microsoft's ultimate green light reportedly hinges on OpenAI actually delivering major model breakthroughs.
Microsoft is preparing to run parts of its Copilot AI locally on PCs to cut latency and boost privacy, according to Intel. The catch? Next-gen AI PCs will reportedly require NPUs pushing over 40 TOPS, a bar no consumer chip on the market currently clears.
Amazon just completed its $4B commitment to Anthropic by injecting another $2.75B into the Claude maker. The deal requires Anthropic to rely on AWS and Amazon's custom chips, tightening the cloud giant's grip in the generative AI race against Microsoft and OpenAI.
Nvidia just swept all seven categories in the latest MLPerf inference benchmarks, showcasing its new H200 chip and TensorRT-LLM software optimizations. Inference now drives roughly 40% of Nvidia's data center revenue. Intel submitted results for Gaudi2 and 5th Gen Xeon, but AMD once again chose not to submit its MI300.
Meta is going to extreme lengths to win the AI talent war, with Mark Zuckerberg personally emailing Google DeepMind researchers to lure them away. The company is even skipping interviews to hand out direct offers and scrapping its old rule against counteroffers. Why it matters: High-level AI talent has quickly become the most fiercely contested commodity in tech.
NVIDIA just unveiled its next-generation Blackwell computing platform at GTC, promising a massive leap forward for trillion-parameter AI models. The new architecture delivers up to 5x better FP4 inference performance than Hopper, alongside the new GB200 Grace Blackwell Superchip and NIM software microservices. Why it matters: Accelerated computing is officially replacing general-purpose silicon, and tech giants are already lining up to build the next wave of generative AI.
Nvidia just launched Nvidia Quantum Cloud, a new microservice enabling researchers to simulate quantum processing units directly on GPU hardware. Built on its CUDA-Q platform, it helps teams test complex algorithms in chemistry, biology, and finance before fault-tolerant quantum hardware matures. The company also unveiled cuPQC to accelerate post-quantum encryption.
Nvidia just introduced Project GR00T, a general-purpose foundation model platform tailored for humanoid robots, alongside a dedicated Jetson Thor computer powered by the Blackwell architecture. Almost every major robotics player—from Boston Dynamics to Figure AI—is already on board, positioning the chip giant as the core brain behind embodied AI.
OpenAI CEO Sam Altman just sat down with Lex Fridman to unpack the November boardroom drama that almost destroyed the company. Altman called it the most painful experience of his career, but noted it built resilience for the massive power struggles ahead. He also predicted compute will become the world's most precious commodity as systems reach major milestones this decade.
Nvidia just rolled out its Earth-2 climate digital twin cloud platform at GTC 2024. Powered by generative AI models like CorrDiff and Omniverse APIs, it simulates high-resolution weather at a two-kilometer scale up to 1,000 times faster and 3,000 times more energy-efficiently than traditional numerical models. Early adopters include Taiwan's weather bureau and The Weather Company.
Inflection just rolled out Inflection-2.5, claiming it rivals GPT-4 while using only 40% of the training compute. The upgrade powers its personal assistant Pi, joining Claude 3 Opus and Gemini 1.5 in a sudden wave of contenders breaking OpenAI's performance barrier.
AI chipmaker Groq just acquired analytics startup Definitive Intelligence for an undisclosed sum to scale its on-demand platform, GroqCloud. Definitive Intelligence previously helped build the cloud infrastructure, which lets developers access Groq's ultrafast LPUs on demand. Its CEO Sunny Madra will now take charge of GroqCloud, alongside a new Groq Systems division targeting government and enterprise deployments.
Microsoft just detailed DirectSR, a new Windows API built with Nvidia, AMD, and Intel to streamline AI super resolution across PC games. Instead of writing custom integrations for DLSS, FSR, and XeSS separately, developers can now support all three through a single code path. It rolls out soon via an Agility SDK preview.
Microsoft just announced a multi-year partnership with French startup Mistral AI, making its new flagship model Mistral Large available first on Azure AI. Mistral gets Azure supercomputing power for model training, while Azure customers can run the model via MaaS alongside OpenAI's lineup. Why it matters: Microsoft is aggressively diversifying its cloud AI portfolio beyond OpenAI.
Google just deepened its partnership with Reddit, gaining direct access to Reddit's Data API to show fresher forum content in search and train its AI models. In return, Reddit will use Google Cloud's Vertex AI to overhaul its platform search. Why it matters: authentic human discussions have become the rarest commodity in the AI race, and Google just plugged straight into the web's largest conversation board.
Groq just went viral with an AI chip architecture that makes standard cloud GPUs look sluggish. Powered by custom Language Processing Units (LPUs), independent benchmarks clocked Groq hitting 247 tokens per second—over 13 times faster than Microsoft’s cloud baseline. Why it matters: this raw speed bypasses memory bottlenecks, finally making lag-free, real-time voice conversations with AI viable.
NVIDIA just released Chat with RTX, a free tech demo that lets you run customized chatbots completely offline on Windows. By connecting local files and YouTube links to open-source models like Mistral and Llama 2 via TensorRT-LLM, you can query notes and documents without sending data to the cloud. You will need a GeForce RTX 30 Series GPU or higher with at least 8GB of VRAM.
NVIDIA CEO Jensen Huang just told world leaders at Dubai's World Governments Summit that every nation needs to own its own "sovereign AI" to protect local data, culture, and history. Huang urged leaders to build out their own compute infrastructure, arguing that natural language has already turned everyone into a programmer.
