🔍 NVIDIA’s AI Chip Strategy 2025: Blackwell, B40, and the Big Bet on China

 

As the AI arms race intensifies, NVIDIA continues to set the pace—not just with raw performance, but with strategic adaptability. From dominating benchmark charts to cleverly navigating U.S.-China tech tensions, NVIDIA’s recent moves in chip development reflect calculated precision and bold vision. Here’s a deep dive into what’s new and why it matters.

---

🎯 Reentering China: The B40 Strategy

Following the U.S. export ban on its high-performance H20 chips, NVIDIA didn’t flinch. Instead, it pivoted—fast.

The newly developed B40 chip, based on NVIDIA’s advanced Blackwell architecture, is crafted specifically for the Chinese market. This chip delivers solid AI capabilities while adhering to U.S. export restrictions. The B40 is expected to enter mass production soon, signaling NVIDIA’s intent to reclaim lost ground in China—without stepping out of regulatory bounds.

Why does this matter? China remains one of the largest potential markets for AI infrastructure. With the B40, NVIDIA is not only complying with export regulations but also offering a product compelling enough to attract major Chinese firms back into its hardware ecosystem.

---

🔗 Meet the B30: A Legal Loophole With Massive Potential

But the B40 isn’t the only ace up NVIDIA’s sleeve. The upcoming B30 chip is designed for modular scalability—meaning while a single unit doesn’t breach export thresholds, multiple chips can be connected to create powerful AI clusters.

This architectural choice is no accident. It’s a brilliant workaround: stay within the legal ceiling, but deliver performance that scales for high-end AI training.

Chinese tech giants like Tencent, Alibaba, and ByteDance are reportedly already in talks to deploy these chips, eager to keep pace in the LLM (Large Language Model) boom.

---

🧠 Blackwell Architecture: Breaking the MLPerf Barriers

Outside of its China play, NVIDIA is crushing the competition on the performance front.

In the latest MLPerf Training benchmarks—the gold standard in AI chip testing—NVIDIA’s Blackwell-based systems dominated across the board. Most notably, they delivered top-tier performance in large language model training, outperforming rivals on models like Meta's LLaMA 3.1 405B.

The results aren’t just marketing fluff. They prove NVIDIA is ready for the next wave of generative AI—whether it’s powering LLMs, multimodal AI systems, or enterprise-scale inference pipelines.

---

🧭 Why This Strategy Matters

NVIDIA is playing 5D chess while the rest of the field is still figuring out the rules. Here’s why their AI chip strategy matters:

• Resilience Under Pressure: Instead of losing momentum due to export restrictions, NVIDIA diversified its portfolio and kept moving forward.

• Scalable Design Thinking: Chips like the B30 are designed with compliance and expansion in mind, giving buyers flexibility in deployment.

• Benchmark Supremacy: Winning on MLPerf shows NVIDIA isn't just compliant or clever—it’s still the undisputed king of AI performance.

---

🔚 Closing Thoughts: Not Just Chips—Strategic Silicon

With the B-series chips, NVIDIA is demonstrating what next-gen leadership in AI hardware really looks like. It’s not just about who builds the fastest chip—it’s about who builds the smartest roadmap.

Whether you're an investor, a technologist, or just an AI enthusiast, keep your eyes on NVIDIA. The chip war is far from over—but as of now, the scoreboard is tilted in their favor.

Artificial intelligence can spot COVID-19 in lung ultrasound

 When a person is infected with COVID-19, the virus can cause specific changes in the lungs, which can be observed through imaging techniques like ultrasound. These changes often manifest as characteristic patterns, such as thickened pleural lines, irregular or confluent B-lines, and subpleural consolidations.

AI algorithms are trained on large datasets of lung ultrasound images, including those from COVID-19 patients and those without the virus. Through this training, the algorithms learn to recognize the distinct patterns associated with COVID-19 infection. Once trained, these algorithms can quickly analyze new ultrasound images and flag any abnormalities indicative of COVID-19.

This AI-driven approach offers several benefits. It can aid healthcare professionals in rapidly identifying potential cases of COVID-19, especially in settings where access to PCR testing or CT scans may be limited. Additionally, it can help prioritize resources by directing attention to cases that are more likely to be positive for COVID-19.

AI has been increasingly utilized in healthcare, including for the detection and diagnosis of COVID-19. Lung ultrasound has emerged as a promising tool for identifying characteristic patterns associated with the virus. AI algorithms trained on vast datasets of lung ultrasound images can analyze these patterns with high accuracy, aiding healthcare professionals in detecting COVID-19-related lung abnormalities swiftly and accurately. This application of AI showcases the potential for technology to assist in diagnosing and managing infectious diseases more effectively.

Certainly! Artificial intelligence algorithms are being developed to analyze lung ultrasound images and identify patterns associated with COVID-19. This technology can help healthcare professionals swiftly and accurately detect COVID-19-related lung abnormalities, improving diagnosis and management of the disease. The advantage of using lung ultrasound for COVID-19 detection lies in its accessibility, portability, and safety compared to other imaging modalities such as CT scans. Ultrasound is non-invasive, does not involve radiation exposure, and can be performed at the bedside, making it particularly suitable for use in various healthcare settings, including emergency departments, intensive care units, etc.

Compiled by Bhumika Sharma

Optimizing model training: Strategies and challenges in artificial intelligence

 When you do model training, you send data through the network multiple times. Think of it like wanting to become the best basketball player. You aim to improve your shooting, passing, and positioning to minimize errors. Similarly, machines use repeated exposure to data to recognize patterns.

This article will focus on a fundamental concept called backward propagation. After reading, you’ll understand:

1. What backward propagation is and why it’s important.

2. Gradient Descent and its type.

3. Backward propagation in Machine Learning.

Let’s delve into backpropagation and its significance.

What is backpropagation and why does it matter in neural networks?

In Machine Learning, machines take actions, analyze mistakes, and try to improve. We give the machine an input and ask for a forward pass, turning input into output. However, the output may differ from our expectations.

Neural Networks are supervised learning systems, meaning they know the correct output for any given input. Machines calculate the error between the ideal and actual output from the forward pass. While a forward pass highlights prediction mistakes, it lacks intelligence if machines don’t correct these errors. For learning in depth about machine learning & Neural networks you can join multiple data science courses available online. Algorithms’ insight into ML and neural networks and their practical application is important to understand. 

After the forward pass, machines send back errors as a cost value. Analyzing these errors involves updating parameters used in the forward pass to transform input into output. This process, sending cost values backward toward the input, is called “backward propagation.” It’s crucial because it helps calculate gradients used by optimization algorithms to learn parameters.

What is the time complexity of a backpropagation algorithm?

The time complexity of a backpropagation algorithm, which refers to how long it takes to perform each step in the process, depends on the structure of the neural network. In the early days of deep learning, simple networks had low time complexity. However, today’s more complex networks, with many parameters, have much higher time complexity. The primary factor influencing time complexity is the size of the neural network, but other factors like the size of the training data and the amount of data used also play a role.

Essentially, the number of neurons and parameters directly impacts how backpropagation operates. The time complexity of the forward pass (the movement of input data through the layers) increases as the number of neurons involved grows. Similarly, in the backward pass (when parameters are updated to repair errors), additional parameters result in increased temporal complexity. 

Gradient descent

Gradient Descent is like training to be a great cricket player who excels at hitting a straight drive. During the model training, you repeatedly face balls of the same length to master that specific stroke and reduce the room for errors. Likewise, gradient descent is an algorithm that aims to minimize the error-proneness, or cost function, to produce the most accurate result possible. Artificial Intelligence uses this Gradient Descent data to train a model. Software model training in depth is covered in many online full stack developer courses. Learning from Online material will give a good hands-on experience in Model Training in ML and software architecture.

But, Before starting training, you need the right equipment. Just as a cricketer needs a ball, you need to know the function you want to minimize (the cost function), its derivatives, and the current inputs, weight, and bias. The goal is to get the most accurate output, and in return, you get the values of the weight and bias with the smallest margin of error.

Gradient Descent is a fundamental algorithm in many machine-learning models. Its purpose is to find the minimum of the cost function, representing the lowest point or deepest valley. The cost function helps identify errors in the predictions of a machine learning model.

Using calculus, you can find the slope of a function, which is the derivative of the function concerning a value. Knowing the slope for each weight guides you toward the lowest point in the valley. The Learning Rate, a hyper-parameter, determines how much you adjust each weight during the iteration process. It involves trial and error, often improved by providing the neural network with more datasets. A well-functioning Gradient Descent algorithm should decrease the cost function with each iteration, and when it can’t decrease further, it is considered converged.

There are different types of gradient descents.

Batch gradient descent

It calculates the error but updates the model only after evaluating the entire dataset. It is computationally efficient but may not always achieve the most accurate results. 

Stochastic gradient descent

It updates the model after every training example, showing detailed improvement until convergence.

Mini-batch gradient descent

It is a deep learning technique that combines batch and stochastic gradient descent. The dataset is separated into small groups and analyzed separately.

Backpropagation algorithm in machine learning?

Backpropagation is a type of learning in machine learning. It falls under supervised learning, where we already know the correct output for each input. This helps calculate the loss function gradient, showing how the expected output differs from the actual output. In supervised learning, we use a training data set with clearly labeled data and specified desired outputs.

The pseudocode in the backpropagation algorithm?

The backpropagation algorithm pseudocode serves as a basic blueprint for developers and researchers to guide the backpropagation process. It provides high-level instructions, including code snippets for essential tasks. While the overview covers the basics, the actual implementation is usually more intricate. The pseudocode outlines sequential steps, including core components of the backpropagation process. It can be written in common programming languages like Python.

Conclusion

Backpropagation, also known as backward propagation, is an important phase in neural networks’ training. It calculates gradients of the cost function concerning learnable parameters. It’s a significant topic in Artificial Neural Networks (ANN). Thanks for reading so far, I hope you found the article informative.

Compiled by Bhumika Sharma

Students use data science to improve food access

 Food access is a serious problem that affects all kinds of communities, including the Binghamton region. Using data science, Binghamton University students are doing what they can to help.

The digital and data studies program at Binghamton University has teamed up with a local organization to address food insecurity. The mutually beneficial connection has helped make great strides in improving food access to the local area.

It all started when the Broome County Food Council surveyed community members on food-related issues. The organization collected a lot of data—1,300 people responded—but needed to analyze it.

That’s where Faculty Engagement Associate Barrett Brenton stepped in. He connected the council with what he saw as an ideal partner: a class in the digital and data studies program.

The program, which is less than two years old, serves as a minor for students. According to coordinator Melissa Haller, the goal of the program is to teach students valuable tech skills. The true emphasis, however, is to think about how students can use those skills to benefit their communities and do social good with data. She found this project as a perfect opportunity to do so.

“It gives them the opportunity to work with a real-world organization — the Broome County Food Council — and to really see the kind of impact the work that they do can have on the local community,” said Haller, who teaches the class.

As for the students, they are also reaping the benefits of this experience.

“At first, we saw Binghamton as our place of study and where we come to work,” said Julia Gnad, a math major and student in the class. “But, I think this project made it more of a community for us. I look back on this and it’s just made this area more to me. I take a lot of pride in helping this community that I’ve called home for four years. I think we were all really grateful for how this project helped us to realize what a career in data analytics looks like and what it could be for us.”

Gnad is one of three students in the class, along with Kajsa Kenney and Brianna Sexton. The class ended after the fall semester, but the trio was so passionate about helping the council that they continued to work with it throughout the spring semester.

According to the survey, two of the biggest obstacles to food access in the area are income and transportation issues. Food Council Coordinator Theresa Krause explained that getting a handle on this issue was especially complex, due to Broome County being “very high rural, very high urban and very unique.”

The end goal is to develop a food access plan for the entire community. The strong analysis of this survey has started to answer some key questions in order to do that, Krause said.

For Haller, the overwhelming success of this project has inspired her to teach a course titled “Community Practice” in the fall. In this course, analysis with the Council will continue, and the class will also help other organizations in the community.

Haller and Brenton both attribute the success of this project to students and believe that it is a good representation of what the digital and data studies program, and other programs like it, can do.

“The analysis done by students in this capstone project is really the kind of exemplar that we want to show how the University can utilize expertise,” Brenton said. “Again, not coming in as experts, but utilizing the expertise we have at the University campus to then engage with the broader community.”

The council will keep working on getting food to the people and people to the food using the information provided by the students, Krause said. The data analysis will be incorporated at numerous future meetings. In September, the council will prevent the food access plan that is currently being developed to the community.

“I see that they brought their gifts and their talents to the table. That spoke to the organization of the council, and this created an excitement and it created a movement going forward – a trust. This is exactly what the council is doing in the community. It was very inspiring. And as we move forward with the food access plan for Broome and developing the strategic initiatives, it just continues to grow and gain roots of foundation,” Krause said.

Haller said she has been amazed by both the students’ work and the strong partnership with the council. She greatly appreciates that the true goal of the new program — to make positive changes with data — is being accomplished through this project.

“I think when you’re learning how to analyze data, it’s very easy to see the work that you do as just numbers,” she said. “And I think one of the incredible benefits of getting to work with the Food Council is that the students have been able to see firsthand what those numbers mean.”

Compiled by Bhumika Sharma

Top 12 AI podcasts to listen to.

 Artificial intelligence is a hot topic in every business and household today.

While AI was first developed in the 1950s, it has taken on a new life in recent years -- in large part due to the launch of OpenAI's ChatGPT in 2022.

Many companies have already adopted AI in some way -- including marketing, customer service and data analysis -- with more companies joining the fray each day. But with AI technology changing so fast, it can be hard to stay up to date on the latest developments and news.

Podcasts are an effective way to stay current on news in the AI world. There are AI podcasts to meet the needs of listeners of all levels. Some break down the latest AI news into easy-to-digest bites. Others give listeners access to the top minds in AI and related industries. TechTarget's Targeting AI is just one of many podcasts available to listeners. It features evergreen content on the world of AI as well as current AI developments.

Here are 12 of the top AI podcasts available on Apple, Spotify and other platforms.

Each of these podcasts was chosen from searches on Google, Spotify and Apple podcasts. All have above a 4-star rating and some are award-winning. All are hosted by long-time tech journalists, industry experts or researchers.

AI Breakdown

From The Breakdown Network, AI Breakdown is a daily podcast focused on AI news analysis. Host Nathaniel Whittemore guides listeners through the latest news in AI and examines what these changes mean for advancements in human creativity, disruptions to work and industries, and the ever-changing relationships between humans and computers.

Where to listen: Spotify, Apple and YouTube.

Average episode length: 25 minutes.

AI in Business

The AI in Business podcast is presented by Emerj, a publishing and marketing research company focusing on enterprise AI ROI. Hosted by Daniel Faggella, this podcast is geared toward nontechnical business leaders who want to integrate AI into their business practices to accelerate growth and deliver ROI. Through interviews with leaders from companies such as Facebook, Mastercard and IBM, Faggella uncovers use cases, best practices and the keys to success in implementing AI in business.

Where to listen: Apple, Spotify and Soundcloud.

Average episode length: 27 minutes.

AI Today Podcast

Hosts Kathleen Walch and Ronald Schmelzer, both founders and managing partners of AI at Cognilytica, discuss the latest AI news in the AI Today Podcast. Discussions include cutting-edge AI technology and interviews with expert guests in easily digestible content that can be applied to real-world issues in the AI and tech industries.

Where to listen: Apple and Spotify.

Average episode length: 20 minutes.

Data Skeptic

Data Skeptic is an interview-based podcast hosted by Kyle Polich that discusses topics including AI, machine learning, data science and statistics. It features themed seasons and offers a bingeable season of AI-related content -- including discussions on large language models, brain-inspired AI and safety concerns with AGI.

Where to listen: Apple and Spotify.

Average episode length: 40 minutes.

Eye on AI

The Eye on AI podcast focuses on an AI expert in each episode. Host Craig S. Smith -- an award-winning correspondent for The New York Times -- interviews AI researchers, tech business leaders and other experts who are leading trends in the AI and machine learning world. The podcast covers many topics including the use of AI in advanced robotics, harnessing AI for synthetic biology and the potential risks of AI use.

Where to listen: Apple and Spotify.

Average episode length: 50 minutes.

Hard Fork

Hard Fork is a technology podcast from The New York Times, hosted by journalists Kevin Roose and Casey Newton. The podcast covers the latest stories in tech and frequently features news and discussions related to AI. Hard Fork won the 2024 iHeart Podcast Award for best in tech.

Where to listen: Apple, Spotify and YouTube.

Average episode length: 70 minutes.

Lex Fridman Podcast

Lex Fridman is an AI researcher at MIT with current research in robot-human interaction and machine learning. He hosts his self-named podcast, which discusses AI, history and current world events. It features in-depth interviews with expert guests, including Sam Altman, Yann LeCun and Elon Musk.

Where to listen: Apple, Spotify and YouTube.

Average episode length: 2+ hours.

Me, Myself and AI

From MIT Sloan Management Review and Boston Consulting Group, Me, Myself and AI is a podcast that asks the question: Why aren't more companies finding success with AI? Hosts Sam Ransbotham and Shervin Khodabandeh attempt to answer this question and more through interviews with leaders from organizations that have found success through AI adoption such as NASA, Volvo and Duolingo.

Where to listen: Apple and Spotify.

Average episode length: 28 minutes.

Practical AI

Practical AI is a weekly podcast, presented by Changelog, that takes a refreshing look at the world of AI and machine learning. Chris Benson and co-host Daniel Whitenack break down the latest trends in AI through discussions with technology professionals, students and industry experts. Taking a practical approach, Benson and Whitenack tackle diverse topics including machine learning, neural networks, generative adversarial networks and large language models in an accessible way that lets listeners apply this information to real-world situations.

Where to listen: Apple and Spotify.

Average episode length: 45 minutes.

The AI Podcast

Host Noah Kravitz interviews guests who are making a difference in their industry through AI. Guests have included a doctor developing AI-powered technology to detect potential heart disease and a startup CEO using AI-driven dubbing to break down language and cultural barriers. Each episode of Nvidia's AI Podcast showcases a unique story about AI and its effect on the world.

Where to listen: Apple, Spotify and Soundcloud.

Average episode length: 30 minutes.

The Artificial Intelligence Show

The Artificial Intelligence Show -- formerly known as The Marketing AI Show -- aims to make AI accessible to the business sector. Host Paul Roetzer is the founder and CEO of the Marketing AI Institute and creator of the Marketing AI Conference. He and co-host Mike Kaput break down the latest AI news to give business owners and professionals actionable insight to accelerate business and career growth.

Where to listen: Apple, Spotify and YouTube.

Average episode length: 65 minutes.

The TWIML AI Podcast

Hosted by AI industry analyst and thought leader, Sam Charrington The TWIML AI Podcast (formerly This Week in Machine Learning & AI) covers topics including machine learning, AI, deep learning, neural networks and natural language processing. Each episode gives tech-savvy business and IT professionals access to the minds of experts and leaders from the machine learning and AI industry. With more than 7 million downloads, The TWIML AI Podcast is a leading voice in the ML/AI industry.

Where to listen: Apple and Spotify.

Average episode length: 45 minutes.

Compiled by Bhumika Sharma

Data Scientist vs Data Analyst

 

Data analysts and data scientists both work with data, but they have different roles and skill sets:

• Data analysts

  Use descriptive and prescriptive analytics to interpret existing data and help businesses answer questions. They use tools like Excel, SQL, and business intelligence programs to collect, process, and analyze large data sets, develop databases, and present insights to stakeholders. Data analysts typically need to understand basic statistics, descriptive statistics, and foundational math, and know SQL and some Python.
• Data scientists

  Use predictive analytics and machine learning to create frameworks and algorithms to capture, store, manipulate, and analyze data to generate value for organizations. They develop new ways to ask and answer business questions, and create sophisticated models to predict future trends. Data scientists typically need to know advanced statistics, linear algebra, calculus, SQL, scripting languages like Python and R, and tools like Jupyter notebook, Google collab notebooks, and Our Studio. 

Which is better data analyst or data scientist?

Neither role is universally "better" as it depends on individual interests and skills. Data analysts typically interpret existing data to help businesses make informed decisions using tools like Excel and SQL. Data scientists, however, often create sophisticated models to predict future trends and require skills in programming languages like Python and machine learning techniques. The best role for you depends on your career goals and technical inclination.

Similarities Between Data Analysts and Data Scientists

Aspect	Data Analysts	Data Scientists
Core Focus	Data	Data
Primary Objective	Analyze data to find actionable insights	Analyze and model data to predict and optimize outcomes
Key Skills	Statistical analysis Data visualization	Advanced statistical analysis Data visualization
Tools Used	SQL Excel Basic analytics tools (e.g., R)	SQL Python/R (used for advanced analytics) Advanced analytics tools
Work Environment	- Collaborative, often part of a data team	- Collaborative, often part of a data or cross-functional team
Decision Making	- Supports business decisions through insights	- Drives business decisions through predictive analytics and insights
Business Impact	- Helps businesses understand and utilize data	- Helps businesses forecast, optimize, and innovate using data
Continuous Learning	- Requires staying updated with current analytics trends and tools	- Requires keeping up with advancements in machine learning, AI, and big data technologies
Communication	- Must effectively communicate findings to stakeholders	- Must explain complex models and predictions to non-technical stakeholders

Conclusion

Choosing between a career in data science and data analysis ultimately depends on your interests and strengths. If you are inclined towards more technical, algorithmic challenges and enjoy delving deep into machine learning and predictive modeling, data science might be the right path. It requires strong programming skills and a robust understanding of advanced statistics. On the other hand, data analysis could be a better fit if you prefer exploring clear insights from data and presenting them in an impactful way, with a lesser focus on heavy coding and complex algorithms. This path involves mastering data manipulation, visualization tools, and statistical analysis but doesn't usually require as deep a dive into programming as data science. Assessing your aptitude for mathematics, enthusiasm for technological innovation, and career goals will help guide your decision.

Complied by Bhumika Sharma