Companies are fundamentally redesigning their operations by placing artificial intelligence at the core of their strategies, treating AI as a collaborative partner rather than just a tool. This approach aims to accelerate growth and continuously generate value, with leaders focusing on metrics like AI usage and trust. Simultaneously, investors are shifting their focus to AI systems that offer robust control and resilience, addressing growing concerns about procurement risk and operational oversight. Governments and businesses now demand AI solutions that are controllable, strong, and allow for human supervision, emphasizing the importance of a solid AI foundation that supports privacy and open standards.
Hardware innovations are significantly enhancing AI performance and reducing costs. NVIDIA's Blackwell platform, for instance, slashes AI inference costs by 4 to 10 times, a benefit reported by companies like Baseten, DeepInfra, Fireworks AI, and Together AI. Sully.ai specifically saw a 90 percent reduction in healthcare AI costs, while Latitude cut gaming AI expenses by 4 times. These savings are largely due to combining Blackwell hardware with optimized software and open-source AI models, alongside the adoption of low-precision formats like NVFP4. Furthermore, new AI hardware is optimizing large language model inference by splitting tasks into prefill and decode stages, with NVIDIA utilizing LPDDR5X memory in its Vera CPUs to create efficient, power-saving memory pools for these distinct processes, leading to more efficient and adaptable data centers.
The integration of AI extends into education and complex problem-solving. Emerson Los Angeles now offers an
Key Takeaways
- AI-first companies are integrating AI as a central partner to improve efficiency and value creation, with leaders tracking AI usage and trust.
- Investors prioritize AI systems that offer strong control, resilience, human oversight, privacy, and open standards.
- NVIDIA's Blackwell platform reduces AI inference costs by 4 to 10 times, combining hardware, optimized software, and open-source models.
- Sully.ai achieved a 90 percent reduction in healthcare AI costs using Nvidia's Blackwell platform.
- Companies like Baseten, DeepInfra, Fireworks AI, and Together AI report significant cost savings with Blackwell.
- New AI hardware, including NVIDIA's Vera CPUs with LPDDR5X memory, splits AI inference tasks (prefill and decode) for faster, more efficient performance.
- Emerson Los Angeles offers an "AI & Media Production" course, teaching students to use AI as a creative partner in filmmaking for tasks like image generation and voice synthesis.
- AI is assisting mathematicians in solving complex problems, such as the Erdős discrepancy problem, demonstrating effective human-AI collaboration.
- The adoption of low-precision formats like NVFP4 is a key factor in achieving major AI cost reductions.
- The World Economic Forum is actively studying how AI is transforming businesses and work methodologies.
AI-First Companies Redesign Work for Human-AI Teams
AI-first companies are changing how they work by putting artificial intelligence at the center. They treat AI as a partner, not just a tool, to improve how things get done. This new way helps businesses grow faster and create value all the time. Leaders focus on measuring things like how much people use AI and how much they trust it. The World Economic Forum is studying these changes to understand how AI transforms businesses.
Investors Seek Control and Resilience in AI Systems
Investors are changing how they put money into AI, focusing on control and resilience. As AI becomes central to businesses, concerns like procurement risk and operational control are growing. Governments and companies now require AI systems to be controllable, strong, and allow human oversight before they will use them. This means that how AI systems are built is becoming very important for competitive advantage. Investing in "good" AI means building a strong AI foundation that supports privacy, open standards, and human control.
New AI Hardware Splits Tasks for Faster Performance
AI inference for large language models now splits into two parts: prefill and decode. The prefill stage handles input, while the decode stage creates output. Using one GPU for both is not efficient, so hardware is being separated to optimize each task. NVIDIA is using LPDDR5X memory in its Vera CPUs to create large, power-saving memory pools for this. This change helps data centers lower costs and improve speed by using the right memory for each part of the AI process. This new method makes AI systems more efficient and adaptable for future needs.
Nvidia Blackwell Slashes AI Costs Up to Ten Times
AI inference costs have dropped significantly, by 4 to 10 times, using Nvidia's Blackwell platform. This improvement comes from combining Blackwell hardware with optimized software and open-source AI models. Companies like Baseten, DeepInfra, Fireworks AI, and Together AI reported these savings across various industries. For example, Sully.ai cut healthcare AI costs by 90 percent, and Latitude reduced gaming AI costs by 4 times. Adopting low-precision formats like NVFP4 also played a big role in achieving these major cost reductions.
Emerson Course Teaches Students AI for Filmmaking
Emerson Los Angeles now offers a new course called "AI & Media Production" that teaches students how to use artificial intelligence in filmmaking. Taught by Emmy-nominated director Stuart Archer, the class helps students see AI as a creative partner, not just a tool. Students learn to generate images and motion, create storyboards, and even use voice synthesis. The course helps them understand AI's potential and ethical concerns in the creative industry. Students like Aadi Sinha and Yifei Wang found the class valuable for making realistic visuals and improving their work.
AI Helps Solve Complex Math Problems
Artificial intelligence is starting to help mathematicians solve long-standing problems, especially those known as Erd
Sources
- How AI-first operating models unlock scalable, responsible value
- Beyond the hype: How investors are rethinking alpha, risk, and demand in the AI stack
- Disaggregated AI Inference: Optimizing Hardware for Prefill and Decode Phases - News and Statistics
- AI inference costs dropped up to 10x on Nvidia's Blackwell — but hardware is only half the equation
- New Course Lets Students Experiment with AI on Screen
- AI uncovers solutions to Erdős problems, moving closer to transforming math
Comments
Please log in to post a comment.