Good morning. March Madness brackets are finally open, so I thought it’d be fun to see if a GenAI model could make The Perfect Bracket, thus winning me the ESPN sweepstakes.

Grok 3 and ChatGPT both outputted plausible-looking brackets, except for the fact that many of the matchups weren’t real and several of the proposed teams simply do not exist. So.

Otherwise, I’m going to win this year, I can feel it.

Although, I say that every year.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

• 🌎 AI for Good: Antarctica’s ice sheet

• 💻 Baidu launches new models to keep up with DeepSeek, OpenAI

• ⚡️ Infrastructure startup Crusoe has a plan to keep the data center lights on

• 👁️‍🗨️ Nvidia has a lot to prove at GTC

AI for Good: Antarctica’s ice sheet

As the planet has grown warmer, the ice sheets that make up Antarctica have been steadily shrinking, with NASA estimating that Antarctica has lost roughly 136 billion metric tons of ice per year since 2002. 

While researchers know Antarctica’s ice sheets are shrinking, current models are unable to accurately predict when and how the sheets will move and melt. 

What happened: A team of Stanford researchers combined math, physics and deep learning algorithms to better predict the future evolution of the Antarctic ice sheets. 

• The team built a machine learning model to analyze ice thickness and large-scale movements, data sourced from remote sensing, airplane radar and satellite imagery between 2007 and 2018. 

• The researchers found that only 5% of the ice sheets are compressed, and so are consistent with previous modeling. The remainder of the ice sheets — called the “extension zone” — stretch as they get pulled out to sea, and so don’t follow the same physical laws. 

Why it matters: Antarctica’s melting ice sheets are directly contributing to a rise in sea levels, which is already causing flooding in some areas; understanding more precisely how and when the ice sheets will shift and shrink better enables researchers to work to protect coastal regions. 

Feeling Left Behind in the AI Gold Rush?

While others are cashing in on artificial intelligence, you might be wondering: "How do I turn this technology into actual income?" The AI revolution is creating millionaires daily, but without the right roadmap, you're missing out on countless opportunities.

Introducing "200+ AI-Powered Income Ideas" – your comprehensive blueprint for monetizing artificial intelligence, even if you're starting from scratch.

Inside this game-changing guide, you'll discover:

• Ready-to-launch strategies across content creation, e-commerce, and gaming

• Step-by-step implementation plans that work even for AI beginners

• Future-proof methods to build sustainable income streams

Click here to download your guide and start your AI income journey today.

Unlock Your AI Income Potential

Baidu launches new models to keep up with DeepSeek

DeepSeek’s launch of R1 at the beginning of this year spurred plenty of competition from the major Western players, driving a price war and leading a widespread push into Chain of Thought (CoT) reasoning. 

But DeepSeek has also spurred on competition within China. 

A week ago, it was Manus, the launch of a system that seems to function with a high degree of autonomy. This week, it’s another entry from Chinese search giant Baidu: Ernie 4.5 and Ernie x1. 

The details: Ernie 4.5 represents Baidu’s latest foundation model, a multi-modal generative AI model that appears to be highly proficient in areas ranging from texts, images, audio and visual content. 

• The model performs on par with or above DeepSeek’s V3 and OpenAI’s GPT-4.5 and GPT-4o on benchmark tests, though these results haven’t been independently validated. It likewise remains unclear what the model was trained on, what the technical architecture of the model is, how it was trained or the cost in dollars, electricity and carbon emissions both of training and operating the model. 

• Baidu said it will integrate these models into its products completely free for individual users, while enterprise users and developers can access them through an API. And therein lies the major edge these Chinese tech firms seem to have against their Western competition: Ernie 4.5, for example, “outperforms GPT-4.5 in multiple benchmarks while priced at just 1% of GPT-4.5.”

Ernie x1, meanwhile, is Baidu’s attempt at a reasoning model, something it achieved through progressive reinforcement learning and CoT reasoning. 

X1 will cost as little as RMB 0.002 per thousand input tokens and RMB .008 for per thousand output tokens. 

This latest entry in free or cheaply accessible GenAI models that remain on par with far more expensive frontier models shortly follows a policy proposal from OpenAI that advocated for the prohibition of Chinese AI infrastructure and “models that violate user privacy and create security risks.” 

New York-listed shares of Baidu $BIDU ( ▲ 3.31% ) surged 9% on Monday. 

Note on exchange rates: RMB 1.00 = $0.14. 

Enterprise AI without the overhead

Deploy secure, no-code AI agents for any team or role from one enterprise-grade platform that users love. That’s Sana Agents.

Discover more

• Language learning: Start speaking a new language in just 3 weeks with Babbel. With expert-crafted lessons and immersive content, you'll be ready for your next adventure in no time. Get 60% off while their Spring Sale lasts - start today! *

• Google taps MediaTek for cheaper AI chips (The Information).

• Dubai’s drone performances open new doors for creative workers (Rest of World).

• Zoom’s AI will soon help you do busy work, too (The Verge).

• DOGE’s cuts at the USDA could cause grocery prices to rise and invasive species to spread (Wired).

• Dow jumps 300 points, S&P 500 rises for second day as stocks try to rebound from rout (CNBC).

Infrastructure startup Crusoe has a plan to keep the data center lights on

Spoiler alert: it involves natural gas. 

AI infrastructure startup Crusoe — which is already working with OpenAI on a Stargate data center in Texas — said it has secured access to 4.5 gigawatts of power, in the shape of seven gas-powered turbines, according to The Information. 

The details: San Francisco-based investment firm Engine No. 1 in January partnered with Chevron to secure seven GE Vernova natural gas turbines, a multi-billion-dollar deal intended to provide power to data centers. 

According to The Information, Crusoe is getting access to all of it. 

• This comes amid a massive and persistent surge in data center electricity demands, not just to train GenAI models, but also to operate them. Demand, at this point, is outpacing production, contributing both to grid instability and carbon emissions. 

• Data center operators, stymied by slow-moving utility companies, have been increasingly looking for ways to power themselves while they wait to come online; in this vein, most of the Big Tech lineup has inked deals with nuclear energy companies. In the interim, gas turbines — which pollute heavily — have become popular; they’re in use at Elon Musk’s xAI facility in Memphis. 

“Because of that step function increase in overall power demand, you really have to bring your own power,” Chase Lochmiller, CEO of Crusoe, told The Information. “There’s no other way for you to be able to achieve that scale that quickly.”

Engine No. 1 and Crusoe are now working to identify sites where the turbines can be deployed, sites they can then sell to eager data center customers. 

Putting things in context: A single gigawatt — equivalent to one billion watts — is roughly enough energy to power 800,000 homes in the U.S., based on average numbers regarding electricity usage. 

We’re talking about a tremendous amount of power, allocated solely for consumption by AI companies.

Nvidia has a lot to prove at GTC

Last month, semiconductor giant Nvidia $NVDA ( ▲ 2.82% ) did what it always does: beat earnings expectations. But that now-anticipated beat wasn’t enough for Wall Street; the stock plummeted 9% on the day following its earnings report and has continued to slide. 

Shares of Nvidia, which rose 171% in 2024, are now down roughly 14% for the year so far. 

The reaction is the likely result of a combination of factors; one, macroeconomic concerns, which have broadly been pulling tech names into the red on fears of a pending recession; two, struggles with Nvidia’s ramp of its Blackwell chip; and three, fears — accelerated by macroeconomic factors — that the AI trade is running out of steam. 

Nvidia hopes to assuage these fears at its annual GTF conference in San Jose, a massive event — featuring a keynote speech from CEO Jensen Huang on Tuesday and a day devoted fully to quantum computing on Thursday — that runs from March 18th to the 21st. 

Here’s what we can expect: Analysts expect Huang to debut Nvidia’s next flagship chip, the Blackwell Ultra. Huang is also expected to talk about Nvidia’s coming pipeline of next-generation AI chips, which includes the Vera Rubin, as well as the company’s outlook considering geopolitical tensions, such as the ongoing tariffs that President Donald Trump has instated, alongside increasing competition with China. 

• We’ll likely hear about new products, and we’ll likely see demonstrations of some of them, like autonomous robotics or vehicles, in action. 

• TF International Securities analyst Ming-Chi Kuo said that Nvidia’s performance hinges on three things: “the continuing effectiveness of scaling laws, the production ramp-up of new AI servers and geopolitical uncertainties.”

If the company properly addresses these issues, Kuo said that “GTC could serve as a short-term catalyst for a share price rebound after the recent pullback in AI stocks. However, the sustainability of this momentum hinges on the conference’s ability to address investor concerns effectively.”

Though other AI-related stocks, including Intel $INTC ( ▲ 2.34% ) and AMD $AMD ( ▲ 0.71% ) rose Monday, shares of Nvidia retreated nearly 2%. 

Last time Nvidia hosted GTC, its stock rose around 7% during the week of the conference. 

But this year, in a lot of ways, is most unlike last year for Nvidia, the AI trade and the broader tech sector. 

At around the same time that Nvidia is painting San Jose green, the Federal Reserve will meet, with chair Jerome Powell set to speak on Wednesday. In terms of deciphering that geopolitical maze that companies, including Nvidia, find themselves currently subject to, Nvidia will have to do quite a lot to provide some solid ground for investors to grab onto. 

To steal an excellent thought from Rasmus Holst, CEO of Zensai, optimism is absolutely essential in this industry. 

The past two years have seen a massive boost in the valuations of just about every public company that has, or might have, a hand in AI, not over current or even near-term realities, but over optimistic expectations for the future. 

A number of factors could eat into that optimism, optimism that has already been degrading for the past several months, simply, since the costs of investment are so high. 

If money for AI projects starts drying up, and Nvidia therefore starts seeing a reduction in demand, we might see an unraveling of the market’s unswerving optimism in Nvidia. 

Though it is the second-largest company in the world by market cap, Nvidia is nowhere near the top when it comes to its price-to-earnings ratio (P/E), a measurement of a company’s stock price compared to its earnings results. The average P/E ratio is around 15. Amazon is in the low 40s, while Microsoft and Apple are each in the low 30s, fairly standard for tech companies. 

Nvidia’s P/E is 63.3. 

The combination of macroeconomic concerns, which could taint optimism and lock the AI trade into something of a downward spiral, plus unsustainably high valuations means that Nvidia, again, has a lot of work to do at GTC to get investors back on the bandwagon in the near-term.

🤔 Your thought process:

Selected Image 2 (Left):

• “Image 1 did a nice job reflecting the Northern Lights, but the reflection of the electric lights in the water just didn't look right.”

💭 A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

Here’s your view on Big Tech:

65% of you do not trust Big Tech.

Only 16% do.

No:

• “We all know that big tech means big money and intrusion into privacy. Must be regulated by a World Wide tight set of regs, not just USA.”

No:

• “Well, not completely anyway. At the end of the day, big tech are companies — which are always out to make a profit and therefore implicitly big tech has bias towards its own benefit.”

If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.