Nvidia beats expectations, but Wall Street wanted more

February 27, 2025

Good morning. We’re not going to talk about Alexa Plus — a generative-AI-enhanced version of Alexa — today. We’re also not going to talk about the rumors that the launch of GPT-4.5 is “imminent.”

We’re going to talk about Nvidia, and what it all means for the broader market.

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

In today’s newsletter:

⚕️ AI for Good: Precision medicine
💻 Report: Most workers don’t use chatbots
📊 Nvidia beats expectations, but Wall Street wanted more

Source: Unsplash

One of the most significant promises of machine learning technology is something called precision medicine, a medicinal approach that centers around treatments targeted to the biological quirks of a given individual. Since the basis of precision medicine involves finding patterns in massive quantities of data, machine learning is something of an ideal tool, here.

ReviveMed, an MIT-spinout, is one of several companies exploring the commercialization of precision medicine. The team has spent years creating a generative AI platform designed specifically to target metabolites, an approach that was first verified in 2016.

How it works: Metabolites refer to small molecules that are products of metabolism, so think amino acids, for instance.

ReviveMed’s idea is that metabolite data is underutilized; by measuring it, scientists can begin to understand both the root causes of disease, as well as the biological details that enable some people to respond better to treatment than others.
The startup gathered a vast “mega-network” of biological data on metabolites, a network that includes how these molecules interact with other molecules. The system then examines a patient’s data against that mega-network to find metabolite connections that indicate disease, something that enables ReviveMed to “discover effective therapeutics to target that.”

In January, the team announced the launch of mzLearn, an algorithm specifically designed for metabolite signal detection. In trials, the system demonstrated a high level of metabolite signal detection, indicating that early and precise diagnosis and targeted therapeutics are starting to become possible. The paper demonstrating the new algorithm has not been peer-reviewed.

Shop top-tier e-Bikes from brands like Aventon, Specialized, and Cannondale—certified, warrantied, and up to 60% off retail.

Upway makes it easy to buy or sell your e-Bike with fast delivery and expert support.

Plus, take $150 off your next ride with code thedeepview.

Upgrade your ride today.

Source: Unsplash

Considering the fact that we’re about to talk about Nvidia’s earnings, which, among other things, are indicative of the breadth and success of the AI ecosystem, I figured it would make sense to take a look at a recent poll from Pew Research Center on the use of generative AI in the workplace.

The details: In a representative survey of more than 5,000 Americans designed to explore the reality of generative AI adoption in the workplace, Pew found that it’s, at best, moving slowly.

The findings: Significantly, the survey found that 40% of those polled never use generative AI chatbots for work, and an additional 29% haven’t even heard of chatbots. 7% use chatbots a few times a month, and 9% use chatbots somewhere between every day and a few times a week.

Among those few that do use chatbots, the most common applications involve generating or editing documents, summarizing information, conducting research and analyzing code. Younger workers (18-29-year-olds) are significantly more likely to use chatbots than older workers.
Among those who use chatbots, only 40% find them extremely helpful in allowing them to do work more quickly; only 29% said chatbots improve the quality of their work. Among those who do not use chatbots, 36% said there “isn’t any use for them in their job.” 22% said they’re just not interested in using chatbots, 10% said they don’t know how and 9% said chatbots are prohibited by their employers.

More than half of those polled are worried about the use of generative AI in the workplace, and a third expect the tech will lead to fewer job opportunities for them.

But for now, more than two years into the ChatGPT-inspired chatbot craze, generative AI doesn’t seem to have taken over the working world, something that doesn’t seem to support the enormous investments being poured into building out generative AI infrastructure.

AI is everywhere in software testing—but is it actually speeding things up?

Rainforest QA surveyed 625 developers, engineers, and tech leaders to uncover what’s working (and what’s not) in test automation.

In their report, you’ll learn:

81% of teams now use generative AI in software testing—but has it delivered real speed improvements?
The surprising truth about AI’s impact on test creation and maintenance time.
Which automation tools and strategies consistently improve testing efficiency?
How smaller teams approach test automation differently from larger organizations.
When teams make the switch from manual to automated testing—and why.

Get the data-backed insights you need to ship reliable code faster. Download the full report now.

New Release: IBM launched a new, reasoning-enabled version of its Granite family of small language models, called Granite 3.2. IBM’s VP of AI Models, Dr. David Cox, told me about the pending release earlier in February.
In another win for small models: Microsoft announced the launch of its latest generation of small language modes, Phi-4 multimodal and Phi-4 mini.

The Federal Reserve's favorite recession indicator is flashing a danger sign again (CNBC).
China wants tech companies to monetize data, but few are buying in (Rest of World).
Utah bill aims to require officers to disclose AI-generated police reports (EFF).
The $19.6 billion pivot: How OpenAI’s 2-year struggle to launch GPT-5 revealed that its core AI strategy has stopped working (Fortune).
Ancient Mars wasn’t just wet. It was cold and wet (Science News).

Source: Nvidia

Before we get to Nvidia’s numbers, we have to talk about what’s been going on, lately.

Let’s start with the Magnificent 7, that coalition of Big Tech hyperscalers that includes Apple, Nvidia, Microsoft, Google, Amazon, Meta and Tesla. Way back before the term “AI” had entered the public vernacular (I’m talking about 2022), the tech sector had itself a very rough year.

But investors, excited by the prospect of generative AI represented by the bulk of the Mag 7, treated these Big Tech players far more kindly in 2023, a year that marked the beginning of a tech stock surge that seems, finally, to be running out of steam.

In 2023, the S&P 500 surged 26%. The Mag7 were responsible for nearly 20% of that growth.
In 2024, the Mag7 were responsible for more than half of the S&P 500’s gains, according to Citi research. By the end of last year, those seven stocks made up roughly 35% of the total value of the S&P.

This is a theme that has not continued into 2025. Investors seem to be growing increasingly concerned about the eye-watering capital expenditures being poured into AI infrastructure, and the lack of AI-related revenue to match, something that was recently compounded by DeepSeek’s release of high-performance, partially open-source generative AI models, models that are significantly less costly to both operate and use compared to entrenched options.

As such, it is perhaps unsurprising that the market has not treated these names all too kindly in the wake of their first-quarter earnings reports.

In the past month, Tesla is down 27%, Google is down more than 10%, Microsoft is down 7% and Amazon is down 7%. Nvidia is up less than 2% over the past month, while Apple and Meta are both up around 6%.

This has dragged both the S&P and the tech-heavy Nasdaq down in kind; over the past month, both indices are down more than 1%. Things became somewhat concentrated over the past week, with the S&P on Wednesday finally — and barely — breaking a four-day losing streak by notching a measly gain of .01%.

While tech has been bleeding — unsurprising, due to investors’ impression that these names are pretty heavily over-valued — signs are also appearing that all is not well with the U.S. economy. The Consumer Confidence Index slipped seven points for February, the largest monthly drop since August of 2021, something that coincides both with mass government layoffs and President Donald Trump’s affirmation that tariffs against Canada and Mexico “will go forward” in March.

Now, Nvidia has become well-known for reporting enormous levels of growth and revenue, always above Wall Street’s expectations. A highly positive report from Nvidia could be enough to send tech stocks back into the green, since it acts as a barometer for the AI trade; if the trade is intact, developers will be buying lots of chips, those ‘picks and shovels’ of the AI race. But lately, expectations have been so high that high performance is no longer enough.

Ok, that said, here’s how they did:

Nvidia reported quarterly revenue of $39.33 billion, a 78% year-over-year increase that came in above the $38.05 billion expected by analysts.
The semiconductor giant reported adjusted earnings of $.89 per share, compared to analyst expectations of $.84 per share.

These record numbers were, yet again, “driven by demand for … generative AI,” according to Nvidia’s CFO outlook letter. Sales of Nvidia’s latest Blackwell chip were “led by large cloud service providers which represented approximately 50% of our Data Center revenue,” something that’s indicative of the circular nature of the AI economy.

For the first quarter of this year, Nvidia expects around $43 billion in revenue, above the $41.78 billion anticipated by analysts.

On the call, CFO Colette Kress assured investors that the efficiency gains represented by models like DeepSeek’s V3 or R1 will not make Nvidia irrelevant; on the contrary, the Chain-of-Thought reasoning approach we’ve been hearing so much about “can require 100 times more compute per task compared to one shot inferences,” meaning developers will still need those picks and shovels.

CEO Jensen Huang said that demand for the Blackwell chip is “extraordinary,” adding that the company has (finally) “successfully ramped up” production of the chip. Huang also said that the era of, first, agents, then physically embodied AI and robotics is just beginning, a possible boon to Nvidia’s business.

The stock, which closed the day up around 3%, fell more than 1% in after-hours trading — a big earnings beat and strong first-quarter guidance were not, it seems, enough for Wall Street.

For context, Nvidia’s market cap was $530 billion this time in 2023 and $1.78 trillion in February of 2024. Today, the stock has a market cap of $3.2 trillion, a valuation that is well above its full-year revenue of $130 billion.

Deepwater’s Gene Munster called the results “great but not extraordinary.”

Last night, we heard from Nvidia what we’ve been hearing for two years, now — record revenue, insane demand and the promise of more record revenue and even more insane demand next quarter.

But the stock is no longer rallying in the face of these pronouncements the way it did in the early days.

This has run for two years, and with inflationary pressures edging nearer all the time, coupled with the fact that investments in AI are far, far outpacing revenue, confidence in Nvidia seems to be flagging.

I expect that today, Nvidia will bleed. I don’t think this report was enough to turn this ship around.