Good morning. For months, I’ve been waiting for some clarity on what OpenAI’s balance sheet looks like. We roughly know how much revenue it’s making, but we’ve never really had an idea how expensive that revenue is.

Yesterday, we got the closest bit of information on that thus far, courtesy of The Information. We break it all down for you below.

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

In today’s newsletter:

• 🌏 AI for Good: Measuring reforestation efforts in northeast Asia

• 🚘 Google invests another $5 billion into Waymo

• 💻 Mistral is on Meta’s heels (sort of) with major new model release

• 💰 OpenAI is on track to lose $5 billion this year

AI for Good: Measuring reforestation efforts in northeast Asia

Yesterday, we talked about the two methods of carbon sequestration: biological and geological. When it comes to biological sequestration, it’s important for researchers to be able to measure and estimate the efficacy of reforestation efforts, and how they relate to forest carbon storage capacities. 

The challenge is that this measurement is difficult to manage in certain areas. 

What happened: In 2020, researchers at Purdue University led an international team to measure forest carbon capacities in northeast Asia. 

• Their research — a blend of remote sensing, fieldwork and machine learning — provided the most up-to-date and accurate estimate of carbon capture potential in South Korea, China and North Korea, a significant leap considering North Korea’s reclusive nature.

• The researchers found that carbon storage in the area increased by 20%-40% over the last 30 years, with around 76% of that “attributable to mass reforestation efforts made across China and the Korean Peninsula.” 

The area studied comprises roughly 0.01% of forested land around the world, but it has nearly 1% of global carbon storage. 

Introducing HumanX: The #1 Gathering for AI Leaders

Ditch the run-of-the-mill AI conferences and join us at HumanX, where we go beyond discussions to inspire and equip you with actionable insights for plotting your AI journey.

In just four months, we've built the world's best initial speaker lineup of over 200 CEOs and innovators from Snowflake, Applied Intuition, Google, and more, making HumanX not just a conference, but a community of leaders ready to act.

Don’t be left behind — buy your pass today with code “HX25n_TDV” to save an additional $150 on your ticket.

Your competitors are probably already coming.

Google invests another $5 billion in Waymo  

Google parent Alphabet said during its second-quarter earnings call that it will be investing a further $5 billion in Waymo, the self-driving unit that was born within Google back in 2009, over the next several years. 

“This new round of funding will enable Waymo to continue to build the world’s leading autonomous driving company,” Alphabet’s Chief Financial Officer Ruth Porat said. 

• Waymo was responsible for driving $365 million in revenue for the quarter, a significant increase from the $285 million it earned at this time last year. 

• But that revenue got more expensive; Waymo’s quarterly losses increased, jumping from $813 million a year ago to $1.13 billion this quarter. 

CEO Sundar Pichai said on the call that Waymo currently provides 50,000 paid trips each week, mainly in San Francisco and Phoenix. 

Why it matters: This additional bet on a thus-far scandal-free self-driving company comes as GM has indefinitely delayed production of its own autonomous vehicle, Origin. It also comes as Tesla’s highly-anticipated robotaxi remains elusive, with CEO Elon Musk pushing the company’s robotaxi event back two months to October. 

Musk said on Tesla’s earnings call that “anyone who doesn't believe that Tesla will solve vehicle autonomy should not hold Tesla stock.”

• Vaire Computing, a startup working on more energy-efficient silicon chips, raised $4 million in funding.

• Cohere raised $500 million in funding, giving the startup a $5.5 billion valuation.

• A day after raising $500 million, AI startup Cohere told staff it was laying off about 20 employees (Fortune).

• The ACLU fights for your right to make deepfakes (Wired).

• Google is the only search engine that works on Reddit now thanks to AI deal (404 Media).

• Meta removes 63,000 Instagram accounts in Nigeria over ‘sextortion’ scams (Reuters).

• Ireland’s data centers overtake electricity use of all urban homes combined (Guardian).

Mistral on Meta’s heels (sort of) with a major new model release 

Earlier this week, Meta released Llama 3.1 405B, a 405 billion parameter, somewhat open and freely-accessible model that is roughly on par with the closed competition from OpenAI and its peers. 

A day later, Mistral tossed its hat in the ring of high-performing open models with Mistral Large 2, a 123 billion parameter model that, according to its chief scientist, is “superior or on par” with Llama 3.1 405B, and is on par with GPT-4o and Claude 3 Opus.  

The details: Mistral said the model “sets a new frontier in terms of performance/cost of serving on evaluation metrics.”

• The model is available on Azure, Google Cloud and Amazon Bedrock. 

• Mistral said that the model provides “much stronger multilingual support.” The company has made the model weights available on Hugging Face. 

Mistral has not disclosed its training data or the carbon footprint associated with training this latest generation of Mistral Large. And I’ll just mention again the academic criticism of the release-by-blog post method, in which companies can make seemingly scientific — though unverified — claims about their models. 

In a big departure from the competition, the model is available “for research purposes only.” 

Invest smarter with Public and AI

If you're looking to leverage AI in your investment strategy, you need to check out Public.

The all-in-one investing platform allows you to build a portfolio of stocks, options, bonds, crypto and more, all while incorporating the latest AI technology — for high-powered performance analysis — to help you achieve your investment goals.

Join Public, and build your primary portfolio with AI-powered insights and analysis.

Analysis: OpenAI is on track to lose $5 billion this year

The Information on Wednesday reported that OpenAI — the current king of the generative AI hill — is on track to lose roughly $5 billion this year. 

The details: Based on previously undisclosed internal financial data and sourcing, The Information reported that as of March, OpenAI was on track to spend at least $4 billion this year solely on renting the Microsoft servers that power ChatGPT and its other models. 

• Then, there’s the cost of training. A source with knowledge of the matter told The Information that the company had planned to spend $800 million this year on model training (including data licensing deals), but, because it decided to ramp up model training faster than expected, OpenAI “ended up spending considerably more.” The Information estimated that OpenAI could end up spending as much as $3 billion on model training this year. 

• Then, there’s the cost of headcount. In 2023, OpenAI had a projected workforce cost of $500 million. During that year, the company doubled its headcount to 800. Now, it employs about 1,500 people and has nearly 200 job postings listed on its website. The cost of headcount could be about $1.5 billion. 

All told, OpenAI’s costs of operation could be in the region of $8.5 billion, a heavy backdrop to the $3.4 billion in annualized revenue the startup recently achieved. OpenAI did not respond to a request for comment. 

The Information said that the company, valued at $80 billion, will likely need to raise cash within the next 12 months. CEO Sam Altman has called OpenAI “the most capital-intensive startup in Silicon Valley history.”

• Venture capitalist Roger McNamee said in response that the “thesis of generative AI is that if you spend hundreds of billions of dollars, you will eventually create a valuable product w(ith) a monopoly.

• “The only part that is guaranteed is the cash burn.”

Some context: Anthropic, a major OpenAI rival, is expected to burn around $2.5 billion this year, all on only $800 million in annualized revenue. And after Tesla reported a lackluster second quarter, CEO Elon Musk asked his Twitter audience if Tesla should invest $5 billion in xAI, Musk’s AI startup, suggesting that it, too, is strapped for cash. 

• Meanwhile, Meta recently released a free, somewhat open-source model that is roughly on par with the models coming from these closed labs. 

• The move, according to AI researcher Gary Marcus, essentially eliminated any moat OpenAI might once have had. 

OpenAI — and Anthropic — have remained mum on the costs involved in operating their businesses. 

In the context of the recent Goldman report and Sequoia paper, each of which addresses the mismatch between spend and revenue that has become the norm of this industry, these numbers add a lot of credence to the idea that the industry is in a bubble. 

• OpenAI is regarded as the clear winner of the genAI race so far. With an annualized revenue of $3.4 billion, the company is clearly making some money. The issue is that it isn’t making enough to justify the massive expenditure forever. 

• What remains unclear is how long the generative AI industry — and OpenAI — can continue at this rate before investors decide they need to start seeing more meaningful returns (and what that might look like, i.e. advertising within chatbots?). 

OpenAI is on a mission to achieve artificial general intelligence. The belief is that such an achievement will allow them to tap into near-limitless resources, making the billions burned on ChatGPT well worth it. That excitement and hype have worked so far on investors, but I don’t know how much longer investors can hang on — and continue pouring more money down OpenAI’s drain — if the burn is real but the AGI remains perpetually somewhere on the horizon. 

And, as far as the science is concerned, there is every chance AGI won’t become any more real anytime soon.

Bubble, burst.

A poll before you go

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

We’ll see you in the next one.

Your view on open-source AI:

33% of you think it’s the best option; 30% of you don’t trust Zuckerberg.

15% of you think it’s super dangerous.

^{*Public disclosure: All investing involves the risk of loss, including loss of principal. Brokerage services for US-listed, registered securities, options and bonds in a self-directed account are offered by Public Investing, Inc., member FINRA & SIPC. Cryptocurrency trading services are offered by Bakkt Crypto Solutions, LLC (NMLS ID 1828849), which is licensed to engage in virtual currency business activity by the NYSDFS. Cryptocurrency is highly speculative, involves a high degree of risk, and has the potential for loss of the entire amount of an investment. Cryptocurrency holdings are not protected by the FDIC or SIPC.}

^{Alpha is an experiment brought to you by Public Holdings, Inc. (“Public”). Alpha is an AI research tool powered by GPT-4, a generative large language model. Alpha is experimental technology and may give inaccurate or inappropriate responses. Output from Alpha should not be construed as investment research or recommendations, and should not serve as the basis for any investment decision. All Alpha output is provided “as is.” Public makes no representations or warranties with respect to the accuracy, completeness, quality, timeliness, or any other characteristic of such output. Your use of Alpha output is at your sole risk. Please independently evaluate and verify the accuracy of any such output for your own use case.}