• The Deep View
  • Posts
  • ⚙️ Nvidia is on the brink of becoming the world’s most valuable company

⚙️ Nvidia is on the brink of becoming the world’s most valuable company

Author

Ian Krietzberg
October 15, 2024

Good morning. Yesterday, NASA launched its Europa Clipper mission, a 5.5-year, 1.8 billion mile-long journey to determine whether the oceans beneath the crust of the Jupiter moon Europa possess the necessary ingredients to create life.

The mission has the potential to return groundbreaking results.

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

In today’s newsletter:

AI for Good: Wildfire mitigation

Source: Unsplash

Wildfires have been steadily worsening for years. Certain applications of generative AI are increasingly being explored as a potential aid to human firefighters. 

The details: The intention is in highly accurate early warning systems that enable first responders to mitigate before things get out of control. 

  • Lockheed Martin is developing a series of firefighting AI tools that parse data from a number of sources — satellite imagery, air and ground sensors — to develop real-time actionable insights. 

  • The resulting system can quickly map a fire and predict its path. It can also predict locations that are highly susceptible to lightning-induced wildfires. 

Colorado recently advanced a bill that would appropriate $7.5 million in funding to continue its partnership with Lockheed Martin to develop and deploy AI-enabled tools in firefighting. 

A regularly complex element to the deployment of AI tools to help fight environmental disasters is the environmental toll the tools themselves exact; small, precise models, operating out of clean data centers, are the ideal here. Large, energy-intensive models operating out of dirty data centers could well cause more harm than good. 

Where Sophisticated Investors Access Private Markets

10 East allows qualified individuals to invest alongside private market veterans in vetted deals across private credit, real estate, niche venture/private equity, and other one-off investments that aren’t typically available through traditional channels. 

Benefits of 10 East membership include:  

  • Flexibility – members have full discretion over whether to invest on an offering-by-offering basis. 

  • Alignment – principals commit material personal capital to every offering.  

  • Institutional resources – a dedicated investment team that sources, monitors, and diligences each offering.

10 East is where founders, executives, and portfolio managers from industry-leading firms diversify their personal portfolios.

There are no upfront costs or minimum commitments associated with joining 10 East.

Adobe launches new video model 

Source: Adobe

Adobe on Monday launched a beta version of its Firefly video generation model, which is now accessible through a number of Adobe’s applications.

The details: Embedded directly in the video-editing software, Premiere Pro users can now employ something called Generative Extend, which can add up to two seconds of additional footage to the beginning or end of a given clip. 

  • The company also made text-to-video and image-to-video available on its Firefly web app; designed to help filmmakers visualize and complete their shoots, the model will shoot out clips no longer than five seconds in length. 

  • Adobe also rolled out a whole series of generative additions and tools across its suite of products, including Photoshop. 

Why it matters: Here, we have yet another publicly accessible generative video model while OpenAI’s heavily hyped Sora remains elusive. As I’ve said before, there is no moat. 

The release of this suite of generative tools comes amid ongoing copyright and ethical concerns regarding the formation and deployment of generative AI; Adobe claims that the model is trained using only “licensed and public domain” content, a marked difference from most developers who use the phrase “publicly available” content to refer to the vast reaches of the internet. 

Adobe additionally said that it would attach content credentials to all clips generated by the system to ensure creators and consumers can track the authenticity and provenance of the content being produced using Adobe’s tools. 

Learn from Nvidia, HP, Databricks, and more

Learn actionable insights, strategies, and techniques for generative AI stack design, governance, evaluation, and observability.

Register now for GenAI Productionize 2.0 – the premier conference for GenAI application development, featuring AI experts from leading brands, startups, and research labs!

But don’t take our word for it; here are real quotes from previous attendees:

"I'm blown away by the high quality and value of this event." - Ricardo B.

"Great event - worth getting up at 4am in the morning for!" - Sandy A.

"Spectacular and very insightful summit! Very well done!" - Chad B.

  • Trees and land absorbed almost no CO2 last year. Is nature’s carbon sink failing? (The Guardian).

  • Techno-optimism as digital eugenics (Cognitive Resonance).

  • S&P 500 hits record to start the week, traders look to key earnings: Live updates (CNBC).

  • The Internet Archive is back as a read-only service after cyberattacks (The Verge).

  • NASA probe Europa Clipper lifts off for Jupiter's icy moon (Phys Org).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

  • Enago Read: A tool to review literature for researchers with summarization and organization.

  • AI-Flow: Connect multiple AI models.

Study: Generative AI and the embers of autoregression 

Source: Created with AI by The Deep View

In determining the reasoning capabilities (or lack thereof) of Large Language Models (LLMs), a team of researchers recently decided to examine the models a bit differently. Arguing that studying LLMs through human benchmarks will not reveal any LLM-unique capabilities, they chose to study LLMs through the lens of the original task the models were designed to accomplish: autoregression, or next-word prediction over text. 

The details: The researchers predicted that LLMs would perform better when the correct output was highly probable, and when the correct task variance was highly probable. 

  • Inversely, they expected LLMs to perform worse when the correct output and correct task were of low probability. Through an exploration of linear functions, shift ciphers and sentence reversals, the researchers found their hypotheses to be accurate. 

  • “Our results show that we should be cautious about applying LLMs in low-probability situations,” one of the researchers wrote. “We should also be careful in how we interpret evaluations. A high score on a test set may not indicate mastery of the general task, (especially) if the test set is mainly high-probability.”

The same team recently published an addendum to this study that explored OpenAI’s o1, a model that was designed not for next-word prediction, but for reasoning. 

They found that, though it performed “substantially” better than the other LLMs they studied, it showed the same tendencies and capabilities around high (and low) probability scenarios. 

Nvidia on the brink of becoming the world’s most valuable company

Source: Nvidia

Shares of Nvidia opened strong on Monday, rallying nearly 3% by the end of the day to $138 per share, surpassing Nvidia’s previous highest-close record of $135.58, notched on June 18. 

The firm, whose stock has spiked some 180% this year alone — began a rally in early 2023 that has yet to, even slightly, diminish. 

Regarded by most as the most obvious way to invest in the artificial intelligence boom that has absorbed such a massive quantity of venture capital funding and investor attention alike, Nvidia is in an enormously powerful position. 

For a variety of reasons, the technology sector is spending a remarkable amount of money to establish AI infrastructure. As TD Cowen analysts recently wrote: “We believe the major companies in AI ... face an investment environment characterized by a Prisoner's Dilemma — each is individually incentivized to continue spending, as the costs of not doing so are (potentially) devastating.”

The biggest names — xAI, Microsoft, Meta — are talking about the purchase and application of hundreds of thousands of the chips that power generative AI. 

Nvidia supplies those chips. 

This is why the firm is considered by many on Wall Street to be the “picks and shovels” of the AI gold rush; even if there is no gold to be found, the business of picks and shovels is a hard cash business, and Nvidia’s picks are in very high demand. For now. 

  • Nvidia CEO Jensen Huang recently said that demand for the company’s next-gen Blackwell chip is “insane … Everybody wants to have the most and everybody wants to be first.”

  • The chip is expected to cost between $30,000 and $40,000 for a single unit; Nvidia’s current-gen chip, the H100, costs about $30,000 for a single chip. Elon Musk’s xAI recently brought a cluster of 100,000 H100s online, at a modest cost, in chip hardware alone, of around $3 billion. 

Meta said in January that by the end of 2024, it would have more than 350,000 of these H100 chips in its arsenal. 

Nvidia is expecting record revenue of $32.5 billion for its third-quarter results, the bulk of which will come from its data center business. 

What’s been going on with Nvidia points very well to the circular economy of AI. These Big Tech names are feeding billions of dollars directly into Nvidia and the resulting spend is elevating everyone’s stock prices. But the fruits of that expense are not, in turn, deriving much in the way of value, as most of Big Tech is right now subsidizing this enormous cost just to get users on board with generative tech. 

The circle is really between Nvidia and Big Tech, while Wall Street watches in excitement. 

It’s not a sustainable scenario. 

At some point, the companies that are paying into the Nvidia charitable fund will need to come up with a way to justify that expense. Before that happens, any sign of slowdown in the AI economy will be evidenced first by Nvidia as a reduction in demand and spending, a result that will likely send Wall Street jumping overboard.

This is why Nvidia’s earnings are so important to the broader ecosystem. If — which I don’t expect — Nvidia reports below its expectations, I would imagine that the investment environment of AI would suddenly become very, very challenging.

Which image is real?

Login or Subscribe to participate in polls.

🤔 Your thought process:

Selected Image 1 (Left):

  • “The shadows appeared more real, more complex and varied. The fake AI image was too perfect, the lighting was ideal and the shadow under the boot on the right appeared unreal.”

Selected Image 2 (Right):

  • “The leaves on the ground are more naturally weathered and less perfect than the leaves on the other photo. The boots also seem less perfect, with some spots that make them actually look like they have been used a bit.”

💭 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 Tesla’s robotaxi:

Your timeline for Elon’s robotaxis: 15% said two years, 30% said five years, 23% said 10 years and 25% said it’ll never happen.

Never:

  • “Technical challenges plus regulatory reluctance plus entrenched interests on the other side of autonomous plus inconsistent focus from Elon will make robotaxi uneconomical. Also, I'd assign a 65% chance that someone other than Tesla makes a breakthrough in the field and dominates whatever future robotaxi market exists.”

10 Years:

  • “Technical challenges remain to be solved, however regulatory challenges will take time for widespread adoption.”

Do you think there's a ceiling on Nvidia's growth?

Login or Subscribe to participate in polls.