AMD Unveils ZAYA1 Alongside Amazon's Kiro Development

The artificial intelligence sector is buzzing with new developments, from significant hardware advancements to complex policy debates and evolving corporate strategies. Zyphra, in a year-long partnership with AMD and IBM, recently unveiled ZAYA1, a major Mixture-of-Experts foundation model. This model, announced on November 25, 2025, is notable for being trained entirely on AMD's MI300X GPUs, AMD Pensando networking, and the ROCm open software stack. Benchmarks from November 14, 2025, show ZAYA1-Base, with 8.3 billion parameters, performing comparably to or better than leading open models like Llama-3-8B and Qwen3-4B in reasoning, math, and coding, positioning AMD as a strong alternative to NVIDIA in large-scale AI development. The MI300X's 192GB of high-bandwidth memory significantly simplified training and accelerated model saving by over ten times.Amidst these technological strides, the regulatory landscape for AI is becoming a battleground. California recently passed its own AI rules, but these face opposition from federal efforts. President Donald Trump's team proposed an executive order to block state AI regulations, and Republicans in Congress are considering similar measures within the National Defense Authorization Act. Advocacy group Build American AI launched a $10 million campaign, using digital, social, and TV ads, to push for a single national AI framework, arguing it's crucial for US leadership against China. New York's growing AI research and $28.5 billion in 2024 venture capital, much of it AI-driven, could be jeopardized by the state's proposed RAISE Act, which critics say has unclear rules that might deter developers and limit access to open-source models.Major tech companies are also refining their AI strategies. Amazon, for instance, issued an internal memo on November 24, directing its engineers to use its proprietary AI code generation tool, Kiro, instead of third-party options like OpenAI's Codex or Anthropic's Claude Code. Amazon, despite its investments in Anthropic and partnerships with OpenAI, aims to strengthen its internal AI development with Kiro, which launched in July and is now globally available. Walmart is exploring new revenue streams through its AI shopping agent, Sparky, testing "Sponsored Prompt" ads within the mobile app feature added in June.However, the rapid expansion of AI also brings scrutiny and skepticism. Private investment in generative AI reached $33.9 billion in 2024, yet companies are facing lawsuits for allegedly overstating AI capabilities or concealing risks, as measuring AI performance proves challenging. Cases include Innodata, accused of claiming AI expertise while cutting R&D, and Lyft, facing claims about false advantages from its AI model. Reddit is also alleged to have not disclosed that Google's AI Overviews significantly reduced its website traffic. Furthermore, leading AI scientists, including Evelina Fedorenko, Steven T. Piantadosi, and Edward A.F. Gibson, argue that current large language models are not true intelligence, emphasizing that human thinking is largely separate from language, contrary to predictions of superintelligence by 2026 from figures like Mark Zuckerberg and Sam Altman.The AI boom's economic and societal impacts are becoming evident. The S&P Cotality Case-Shiller Indices indicate national home prices have stopped growing, a stagnation attributed to high housing costs relative to wages and a slowing job market, even as AI spending drives overall economic growth. Zillow senior economist Orphe Divounguy forecasts only a modest 1.2 percent rise in home values over the next year, contingent on stable employment. In the entertainment industry, filmmaker Shekhar Kapur noted on November 25, 2025, at the 56th IFFI, that actors like Tom Cruise, Hrithik Roshan, and Anil Kapoor have copyrighted themselves in response to AI's rise, highlighting concerns about AI's role in creative fields. PwC partner Diego Jarne Munoz advises companies to move beyond small AI tests, build repeatable systems, and establish a strong data foundation to truly scale AI and achieve value.

Key Takeaways

• Zyphra, AMD, and IBM partnered to develop ZAYA1, a Mixture-of-Experts AI model trained entirely on AMD MI300X GPUs, AMD Pensando networking, and ROCm software.
• ZAYA1-Base, with 8.3 billion parameters, performs comparably to or better than Llama-3-8B and Qwen3-4B, offering a strong alternative to NVIDIA for large-scale AI training.
• AMD's MI300X GPUs provide 192GB of high-bandwidth memory, which facilitated efficient training and ten times faster model saving for Zyphra.
• A $10 million campaign by Build American AI advocates for a single national AI framework, opposing state-level regulations like those passed in California and proposed in New York (RAISE Act).
• Amazon mandated its engineers to use its internal AI code generation tool, Kiro, over third-party options such as OpenAI's Codex and Anthropic's Claude Code, despite its investments and partnerships.
• Walmart is testing "Sponsored Prompt" ads within its Sparky AI shopping agent, integrated into its mobile app in June, to explore new advertising revenue streams.
• Private investment in generative AI reached $33.9 billion in 2024, but companies face lawsuits for overstating AI capabilities or hiding risks, exemplified by cases involving Innodata, Lyft, and Reddit.
• Leading AI scientists argue that current large language models are not true intelligence, as human thinking is largely separate from language, challenging predictions of superintelligence.
• The AI boom is impacting the US economy, with national home prices stagnating due to high costs and a slowing job market, despite AI spending driving overall economic growth.
• Actors like Tom Cruise, Hrithik Roshan, and Anil Kapoor have copyrighted themselves in response to the rise of AI in the entertainment industry.

Zyphra AMD IBM launch ZAYA1 AI model

Zyphra, AMD, and IBM partnered for a year to test AMD's GPUs for large-scale AI training. This effort resulted in ZAYA1, the first major Mixture-of-Experts foundation model built entirely on AMD GPUs and networking. ZAYA1, trained on AMD's MI300X GPUs, performs as well as or better than other open models like Qwen3-4B and Llama-3-8B in areas like reasoning, math, and code. This offers businesses a strong alternative to NVIDIA for AI development. The MI300X GPUs provide 192GB of high-bandwidth memory, simplifying complex training projects.

Zyphra achieves AI breakthrough with AMD technology

On November 25, 2025, Zyphra announced a major AI training breakthrough with its ZAYA1 model. This large-scale Mixture-of-Experts foundation model is the first to be trained entirely on AMD's GPU and networking platform. It uses AMD Instinct MI300X GPUs, AMD Pensando networking, and the ROCm open software stack. ZAYA1-Base shows strong performance in reasoning, math, and coding, matching or beating models like Llama-3-8B. The MI300X's 192 GB memory helped Zyphra train efficiently and save models over ten times faster.

AMD powers Zyphra's ZAYA1 AI model training

Zyphra created ZAYA1, the first large-scale Mixture-of-Experts foundation model trained completely on AMD Instinct MI300X GPUs, AMD Pensando networking, and the AMD ROCm software. The MI300X's 192 GB of memory made training easier and saved models ten times faster. Benchmarks from November 14, 2025, show ZAYA1-base, with 8.3 billion total parameters, performs as well as or better than leading models like Llama-3-8B. This achievement highlights AMD's ability to support complex AI workloads, supported by a system co-engineered with IBM.

California AI laws face federal challenge

California lawmakers passed new rules for artificial intelligence this year. However, President Donald Trump and Republicans in Congress are trying to stop states from making their own AI laws. Trump's team proposed an executive order that would block state AI rules, and Republicans in Congress are considering a similar measure in the National Defense Authorization Act. California Attorney General Rob Bonta opposes this, saying it would hurt states' ability to manage new technology and let AI companies avoid important protections. California is home to many top AI companies.

AI group launches 10 million dollar federal policy push

Build American AI, an advocacy group, started a 10 million dollar campaign to urge Congress to create a single national AI framework. The group, part of Leading the Future super PAC, uses ads on digital, social, and TV platforms. They argue that a federal plan is vital for the US to lead in AI against China, protecting jobs and American values. This campaign supports efforts in Congress to stop states from making their own AI laws. President Trump also backs this idea, having considered an executive order to challenge state AI regulations.

Amazon promotes Kiro AI coding tool internally

On November 24, Amazon told its engineers to use its own AI code generation tool, Kiro, instead of third-party options. An internal memo stated the company will not support other AI development tools, which includes popular ones like OpenAI's Codex and Anthropic's Claude Code. Amazon launched Kiro in July and recently made it available worldwide with new features. The company aims to strengthen its own AI development, despite having invested in Anthropic and partnering with OpenAI.

New York AI growth threatened by RAISE Act

Cameron D. Miller writes that New York's growing AI research and investment could face problems if the RAISE Act becomes law. This bill, passed in Albany, might push AI developers and investors to other states. It could also harm partnerships between universities and companies, which are vital for breakthroughs like new cancer treatments. New York City saw 28.5 billion dollars in venture capital in 2024, with AI driving much of this growth. The RAISE Act's unclear rules could make researchers hesitant to develop new AI tools and might limit access to open-source models.

Shekhar Kapur discusses AI and actor copyrights at IFFI

On November 25, 2025, filmmaker Shekhar Kapur spoke at the 56th IFFI in Goa about artificial intelligence in entertainment. He mentioned that many actors, including Tom Cruise, Hrithik Roshan, and Anil Kapoor, have copyrighted themselves due to AI's rise. Kapur believes AI cannot truly be human and that human predictability is the only way AI could take over. He also emphasized that storytelling remains constant, regardless of changing attention spans.

Language models are not true intelligence scientists say

Top AI leaders like Mark Zuckerberg and Sam Altman predict superintelligence and Artificial General Intelligence by 2026. However, this article argues that current AI systems, mainly large language models, are not true intelligence. These models primarily process linguistic data and predict text, but neuroscience shows human thinking is largely separate from language. Experts like Evelina Fedorenko, Steven T. Piantadosi, and Edward A.F. Gibson published in Nature that language is mainly a communication tool, not thought itself. Therefore, simply scaling language models will not create human-level intelligence.

Companies face lawsuits for overhyping AI abilities

Private investment in generative AI reached 33.9 billion dollars in 2024, but companies now face lawsuits for overstating AI capabilities or hiding risks. Measuring AI performance is difficult, leading to unfulfilled promises. For example, in the Innodata case, the company is accused of claiming AI expertise while cutting research and development. Similarly, Lyft faced claims about its AI model's "significant advantage" being false. The Reddit case alleges the company did not reveal that Google's AI Overviews was drastically cutting its website traffic.

AI boom impacts US home sales and prices

The artificial intelligence boom is changing the US economy, and its effects are now reaching home sales. Recent S&P Cotality Case-Shiller Indices show national home prices have stopped growing. This stagnation comes from high housing costs compared to wages and a slowing job market, despite overall economic growth driven by AI spending. Zillow senior economist Orphe Divounguy predicts only a small 1.2 percent rise in home values over the next year, but warns this depends on stable employment rates.

Walmart tests ads for its Sparky AI shopping agent

Walmart is looking into new ways to make money from ads, especially with its new AI shopping agent called Sparky. The company has been testing an ad type called "Sponsored Prompt" within Sparky, which was added to Walmart's mobile app in June. This move shows Walmart's interest in finding revenue opportunities as more people use chatbots for online shopping. Sparky is a key part of Walmart's bigger plan to use AI across its business.

PwC expert shares how to scale AI with data

Diego Jarne Munoz, a partner at PwC, explains how companies can get real value from artificial intelligence. He says organizations should move past small tests and build repeatable systems that save time and money. Many companies start using AI without properly organizing their data, which causes problems later. Munoz believes that how much a business uses AI is the best sign of success, not just having a good model. He advises building a strong data foundation alongside early AI projects.

Sources

• ZAYA1: AI model using AMD GPUs for training hits milestone
• AI training breakthrough achieved using AMD technology
• AMD Powers Zyphra’s New AI Model Training
• Can CA protect its AI regulations from Trump’s executive order?
• AI advocacy group launches $10M campaign pushing for federal AI framework
• Amazon pushes in-house AI coding tool Kiro over competitors', memo shows
• Op-Ed | AI Research and Investment Face Regulatory Headwinds if RAISE Act Becomes Law
• Firstpost at IFFI 2025 | Filmmaker Shekhar Kapur on Artificial Intelligence: 'Tom Cruise, Hrithik Roshan, Anil Kapoor have copyrighted themselves and...'
• Is language the same as intelligence? The AI industry desperately needs it to be
• Overpromising Gen AI’s Abilities Creates a Pitfall for Companies
• AI’s Ripple Effects Extend to Home Sales
• Exclusive | Walmart Is Exploring Bringing Ads to Sparky, its New AI Shopping Agent
• Scaling AI With Trusted Data Foundations