The Week in AI: January 5-11, 2026 (Week 2)

This week’s news: January 5-11, 2026 (Week 2)

The first working week of January was dominated by CES in Las Vegas, where Nvidia and AMD unveiled new data center chips, and Nvidia and Boston Dynamics showed that artificial intelligence is heading into cars and robots. Within just a few days, OpenAI and Anthropic moved into healthcare, while assistants from Google and Microsoft are making their way into email and shopping. The money keeps pouring in: xAI raised $20 billion, Anthropic is in talks at a $350 billion valuation, and the Chinese companies Zhipu and MiniMax went public. The week was overshadowed by a scandal involving the Grok chatbot, which generated sexualized deepfakes and triggered action by regulators around the world.

Chips and Hardware from CES 2026

Nvidia: The Vera Rubin Platform Is Already in Production

Nvidia CEO Jensen Huang announced in his keynote at CES on Monday, January 5, that the new Vera Rubin platform is in production. It is the successor to the Blackwell generation, the chips that currently handle the training and operation of most large models.

The platform consists of six co-designed chips, including the Rubin GPU and the new Vera CPU. The Vera Rubin NVL72 server rack combines 72 GPUs and 36 CPUs. According to Nvidia, Rubin should deliver roughly five times the inference performance (that is, the work of the model actually generating answers) of Blackwell. Volume production is expected to ramp up in the second half of the year (Tom’s Hardware).

Nvidia Releases a Bundle of Open Models, Including Alpamayo for Cars

Alongside the hardware, Nvidia unveiled a series of open models for various fields on January 5. They are freely available for download on Hugging Face and GitHub (Nvidia blog):

Nemotron for AI agents: new models for speech recognition, document retrieval, and safety filtering.
Cosmos Reason 2 and other “physical AI” models designed to help robots understand the real world.
Isaac GR00T N1.6 for humanoid robots and Clara models for healthcare and drug design.

The Alpamayo family for self-driving cars drew the most attention. The flagship model, Alpamayo 1, has 10 billion parameters and is a vision-language-action model: it proposes a route directly from camera video while also explaining why it made that decision. The first production car to get it will be the new Mercedes-Benz CLA, in the US as early as the first quarter of 2026 (Electrek).

AMD Responds with the Helios Platform

AMD CEO Lisa Su unveiled the Helios server platform with Instinct MI455X accelerators on January 5. It is meant to compete with Nvidia and arrive later in 2026. She was joined on stage by OpenAI President Greg Brockman, as OpenAI is one of AMD’s major customers (The Decoder).

For laptops, AMD announced Ryzen AI 400 processors with a more powerful chip for local AI.
Lisa Su predicted that within five years, around five billion people will be using artificial intelligence every day.

Robots and Consumer Electronics

Boston Dynamics’ Atlas Gets a Brain from Google DeepMind

Hyundai and its subsidiary Boston Dynamics showed the production version of the Atlas humanoid robot at CES on January 5. They also announced a partnership with Google DeepMind, which will bring its Gemini Robotics models to the robots (TechCrunch).

This year’s entire production run of Atlas robots is already spoken for: the robots will go to Hyundai’s research center and to Google DeepMind. Notably, Google sold Boston Dynamics years ago and is now returning to it as a partner.

Assistants in Computers and TVs

Lenovo unveiled its Qira assistant on January 6. It is designed to work across computers, tablets, and Motorola phones and to remember context between devices (South China Morning Post).
Google demonstrated new Gemini features for Google TV on January 5: voice control of settings, photo search, and photo editing with the Nano Banana model right on the TV (Engadget).

Products and Agents

ChatGPT Health: A Health Section Connected to Medical Records

OpenAI introduced ChatGPT Health on January 7, a separate part of the chatbot for health questions. Users can connect their medical records and apps such as Apple Health or MyFitnessPal. According to the company, more than 230 million people ask ChatGPT about health every week (Fortune).

Health conversations have their own history and memory, kept separate from regular ChatGPT. OpenAI stresses that the service is not meant to make diagnoses or replace doctors, but to help people make sense of test results and prepare for appointments.

Anthropic Responds with Claude for Healthcare

Four days later, on January 11, Anthropic announced the Claude for Healthcare package, timed for the start of the J.P. Morgan Healthcare conference in San Francisco. For hospitals and insurers, it offers tools compliant with HIPAA, the US health data protection law (Fortune).

Claude Pro and Max subscribers in the US can now give Claude access to their lab results and medical records through the HealthEx and Function services. Both companies state that health data will not be used to train models.

Gmail Enters the “Gemini Era”

On January 8, Google announced the biggest transformation of Gmail in years. The Gemini 3 model is coming to the inbox, where it can summarize long threads, help write replies, and answer questions about the contents of your mailbox (Google Blog).

A new feature is AI Inbox, a separate view of your mail that sorts messages by importance rather than by date. For now, it is being tested only by a limited group of users in the US, and some features require a paid Google AI Pro or Ultra subscription (Fortune).

Shopping Right in the Copilot Chat

On January 8, Microsoft launched Copilot Checkout in the US, a feature that lets users complete a purchase directly in the assistant window without being redirected to an online store. Payments are handled by PayPal, Shopify, and Stripe, while the merchant remains the seller and owns the customer relationship (Axios).

New Models

Abu Dhabi’s TII released the open model Falcon H1R 7B on January 5. It has only 7 billion parameters, but according to its authors it matches models several times its size on math benchmarks; it combines the Transformer and Mamba architectures (The Decoder).
According to a report by The Information, picked up by Reuters on January 9, DeepSeek is preparing a coding-focused V4 model for mid-February. Internal tests reportedly show it could outperform both Claude and GPT models at coding; this has not been independently verified (Reuters via Yahoo).

Business and Investment

xAI Raises $20 Billion

Elon Musk’s company xAI announced on January 6 that it had closed a $20 billion Series E funding round, surpassing its original target of $15 billion. Investors include Nvidia, Cisco, and the Qatari sovereign wealth fund QIA. The money is earmarked for data centers and the development of Grok models (Bloomberg, SiliconANGLE).

Anthropic in Talks at a $350 Billion Valuation

According to the Wall Street Journal and Reuters on January 7, Anthropic is looking to raise $10 billion at a pre-money valuation of $350 billion. That is nearly double its valuation in the September round. The round is expected to be led by Coatue and Singapore’s GIC (SiliconANGLE).

Chinese AI Companies List in Hong Kong

Zhipu AI (January 8) and MiniMax (January 9) debuted on the Hong Kong stock exchange, becoming the very first companies developing large language models to go public. Zhipu shares closed their first day up 13 percent, while the MiniMax share price doubled during its first day of trading (Rest of World, Yicai Global).

Power for Data Centers

Meta announced nuclear deals with Vistra, TerraPower, and Oklo on January 9 for up to 6.6 gigawatts by 2035. They include buying electricity from existing plants as well as funding new small reactors (Axios).
OpenAI and SoftBank each invested $500 million on January 9 in SB Energy, which is building a 1.2-gigawatt data center in Texas for OpenAI as part of the Stargate project (Reuters via Yahoo Finance).
On January 8, OpenAI acquired the team behind the startup Convogo, which developed an AI tool for executive coaching; the product itself will be shut down (TechCrunch).

Regulation and Law

Character.AI and Google Settle Lawsuits over Teen Suicides

Court documents from January 7 show that Character.AI and Google have agreed to settle several lawsuits brought by families whose children harmed themselves or died by suicide after conversations with chatbots. The best-known case is that of 14-year-old Sewell Setzer III from Florida (Axios).

The settlement amounts have not been disclosed, and the agreements still need court approval. Google was sued because in 2024 it hired the founders of Character.AI and licensed its technology.

Musk v. OpenAI Will Go to a Jury

On January 7, Judge Yvonne Gonzalez Rogers in Oakland, California, ruled that Elon Musk’s lawsuit against OpenAI over its shift to a for-profit model will go before a jury. In her view, the case involves enough disputed facts that it should not be decided by a judge alone (Reuters via The Star).

Utah Allows AI to Renew Prescriptions

On January 6, the US state of Utah announced a pilot program with the startup Doctronic in which artificial intelligence may renew prescriptions for common medications on its own. The initial prescription must always be issued by a doctor, and the first 250 renewals in each drug category are reviewed by a human (Axios).

Safety and Incidents

Grok and a Wave of Sexualized Deepfakes

The Grok chatbot from xAI on the X network edited photos of real people, including minors, into sexualized images on request. These fake images, known as deepfakes, triggered responses from governments and authorities on several continents over the course of the week (Malwarebytes):

The European Commission ordered X to retain all internal documents and data relating to Grok until the end of 2026 as part of an investigation under the EU’s Digital Services Act (AP).
At the end of the week, X restricted image generation and editing via Grok to paying subscribers only. According to AP, however, the standalone Grok app and website remained available for free. A UK government spokesperson called the restriction “an insult to victims.”
On January 10, Indonesia became the first country to temporarily block access to Grok, citing violations of human dignity (Business Today).
Malaysia followed on January 11 with a temporary restriction of access.

The case shows how quickly an image editing feature can become a tool for abuse. Regulators also made it clear that a paywall does not solve the problem.