Shoppers are adding to cart for the holidays

Peak streaming time continues after Black Friday on Roku, with the weekend after Thanksgiving and the weeks leading up to Christmas seeing record hours of viewing. Roku Ads Manager makes it simple to launch last-minute campaigns targeting viewers who are ready to shop during the holidays. Use first-party audience insights, segment by demographics, and advertise next to the premium ad-supported content your customers are streaming this holiday season.

Read the guide to get your CTV campaign live in time for the holiday rush.

Learn more.

Today:

• Google Challenges Nvidia’s AI Chip Dominance

• China’s Ghana TPU claims to outpace Nvidia’s A100

• Google’s “Willow” Chip Hits Quantum Advantage

• Sora & Nano Banana Pro throttle free use

• Court orders OpenAI to reveal book-dataset chats

Google Challenges Nvidia’s AI Chip Dominance

Google is now selling its TPUv7 chips, powerful processors for training and running AI. Cloud firm Anthropic plans to buy over one million of them. TPUs cost less to run than Nvidia GPUs, so big buyers threaten Nvidia’s hold over expensive AI hardware markets today.

KEY POINTS

• Google opens TPU sales: After years of using TPUs only inside Google, the company is offering full systems to outside customers such as Anthropic, Meta, SSI, xAI, and possibly OpenAI.

• Cost wins over Nvidia: Analysis shows TPUs can cut training and serving costs by 30-60 percent versus Nvidia’s newest GPUs, largely because of lower chip prices and efficient system design.

• Ripple effects on industry: Huge TPU orders force Nvidia to discount or invest in clients, spur new data-center financing models, and give “neo-cloud” providers fresh business chances.

Why it matters

Cheaper, faster chips mean AI firms can build smarter tools for less money. If TPUs spread, competition rises, prices fall, and new ideas reach users sooner. Nvidia may lose its lead, pushing all hardware makers to improve—and that helps everyone who relies on AI.

China’s Ghana TPU claims to outpace Nvidia’s A100

A Chinese startup founded by ex-Google engineers claims its custom AI chip, “Ghana,” outperforms Nvidia’s A100 by 1.5x and uses no foreign tech. Though still behind latest chips, it signals China’s growing push for semiconductor independence amid trade barriers and hardware shortages.

KEY POINTS

• Chip Performance: The “Ghana” TPU claims 1.5x speed over Nvidia’s A100 and 42% better energy efficiency.

• Self-Reliance: Developed with fully domestic technology, avoiding any foreign intellectual property or tools.

• Geopolitical Context: Emerges as China ramps efforts to reduce reliance on U.S. chips and respond to export restrictions.

Why it matters

This chip shows how China is trying to catch up in the AI hardware race. If it works well, it could help China rely less on U.S. technology. That would shift who controls access to the most powerful chips—key tools for AI progress.

Google’s “Willow” Chip Hits Quantum Advantage

Sundar Pichai says quantum computing is where AI was five years ago—on the verge of breakthrough. Google’s “Willow” chip and new “Quantum Echoes” algorithm mark major progress, with real-world applications in science, medicine, and energy expected within five years.

KEY POINTS

• Quantum Is the Next AI: Pichai compares today’s quantum state to AI's inflection point in the late 2010s, forecasting major advances by 2030.

• Real-World Breakthroughs: Google's quantum push could lead to innovations in materials, drug discovery, and clean energy by simulating natural processes better than ever.

• Quantum Echoes & Willow Chip: Google’s Willow chip achieved verifiable quantum advantage using the new Quantum Echoes algorithm, pushing the tech from lab to reality.

Why it matters

Quantum computing could become the next big leap in technology, reshaping science and industry. Google’s progress shows that we’re moving closer to practical quantum machines. Like AI before it, this shift could unlock new inventions, solve complex problems, and change how we live.

🧠RESEARCH

NVIDIA Nemotron Parse 1.1

Nemotron-Parse-1.1 is NVIDIA’s small AI model for reading documents and images. It improves text recognition (OCR), markdown layout, tables, and diagrams, and handles longer, dense pages. The 885M-parameter two-part model matches larger systems, is released openly with tools and data, plus a faster “TC” variant with slight quality loss overall.

Video Generation Models Are Good Latent Reward Models

Researchers show video generators can also act as reward models that judge which videos people prefer. Their PRFL method trains directly in the model’s hidden “latent” space instead of decoded pixels, improving motion and structure, cutting memory use and training time, and better matching human preferences in tests across benchmarks.

The Image as Its Own Reward: Reinforcement Learning with Adversarial Reward for Image Generation

The paper proposes Adv-GRPO, a training method where images themselves become the “reward” signal. Instead of trusting a single scoring model that can be tricked, it compares outputs to reference images using strong vision models, giving rich feedback that improves realism, style control, and task performance over prior diffusion methods.

🛠️TOP TOOLS

Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.

AKOOL – AI Face Swapping for Photos, Video & Live Camera - browser‑based AI tool that swaps one face onto another in photos, pre‑recorded videos, and a live webcam mode

Akuma AI – AI Anime Art & Character Role‑Play - web platform focused on anime‑style creation

Albus AI – Build AI‑Powered Knowledge Bases & No‑Code Search - AI workspace for turning your files and links into a searchable knowledge base

📲SOCIAL MEDIA

🗞️MORE NEWS

• Google and OpenAI are limiting free use of their popular AI tools, Nano Banana Pro and Sora, due to overwhelming demand. Free users now get fewer generations, while paid users keep full access. Paid upgrades encouraged.

• OpenAI must share internal chats about deleting book datasets, losing a key legal fight in a copyright case. The ruling could lead to bigger damages if OpenAI is found to have knowingly broken the law.

• OpenAI’s key partners have taken on $100 billion in debt to back its ambitious growth plans. This massive funding highlights the high costs of building advanced AI and staying competitive in the global tech race.

• OpenAI's GPT-5 helped researcher Sebastien Bubeck solve a complex math problem in hours instead of a month. It designed, simulated, and proved a solution—hailed as the most impressive output he’s ever seen.

• Kimi launched a 48-hour free trial for its new slide generator using Nano Banana Pro. It turns documents into editable slides, though results can be uneven. It includes K2 search, but lacks design templates.

• Over 21% of peer reviews at a top AI conference were written entirely by AI, sparking concerns about fairness and trust. Researchers flagged vague, inaccurate feedback. Organizers now plan to enforce AI disclosure rules.