Good morning. Today, we’re taking a deep dive into all the interesting complexities surrounding OpenAI’s $40 billion fundraise.

That’s it. That’s all I got for you. Happy Wednesday.

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

In today’s newsletter:

• 🎙️ Podcast: Tools, not gods: IBM VP on AI, neuroscience and the nature of intelligence

• 🔏 Study: Are LLMs actually good at math? 

• 💰 OpenAI raises record $40 billion — don’t call it a bubble

🎙️ Podcast: Tools, not gods: IBM VP on AI, neuroscience and the nature of intelligence

The latest episode of The Deep View: Conversations — an exploration into neuroscience and AI with IBM VP Dr. David Cox — is out.

Give it a listen here, or check it out on YouTube above.

“That arc of we build tools, we pass on knowledge — we built machines that take away physical labor; we built machines that could transmit information. I want that arc to continue and I don't see there being a direct line to AGI, whatever that is, or any kind of replication necessarily of human-like intelligence or all the other things that come along with human intelligence,” Cox said. “And I think there's something distracting about the AGI narrative.”

“It takes us to a place that's not obvious to me is the place we need to go.”

Fyxer AI: Gain 1 hour every day with an AI executive assistant.

Meet Fyxer, your AI Executive Assistant. Fyxer AI will get you back one hour, every day. Begin your day with emails neatly organized, replies crafted to match your tone, and crisp notes from every meeting.

• Email Organization: Fyxer prioritizes important emails, ensuring high-value messages are addressed quickly while filtering out the spam.

• Automated Email Drafting: Fyxer drafts personalized email responses in your tone of voice. It's so accurate that 63% of emails are sent without edits.

• Meeting Notes: Fyxer listens to your video calls and sends a meeting summary with clear next steps.

Fyxer AI is even adaptable to teams, learning from team members' communication styles and enhancing productivity across the board.

Setting up Fyxer AI is easy—it takes just 30 seconds to get started with Gmail or Outlook, no learning curve or training needed.

There's no credit card required for the 7-day free trial, making it a risk-free productivity booster. Start your free trial now.

Study: Are LLMs actually good at math? 

Benchmarks have become a core focus of the AI field, touted as some indication of the general capability of the models in development. But, often, benchmarks don’t measure what they purport to measure, and further, don’t indicate what they seem to indicate (benchmark results when training data is unknown don’t really say much). 

When OpenAI unveiled o3 and o3-mini, part of its big demo involved showcasing the model’s scores on a Frontier Math benchmark. Of course, it later came out that OpenAI was funding Epoch AI, the company behind the benchmark, and had access to its data, but Frontier Math is only one of the available benchmarks. 

There’s a whole website dedicated to it — the Math Arena — which ranks large language models (LLMs) based on their accuracy in solving the latest math competitions (problems that, because they’re new, aren’t likely to be found in their training data). 

What happened: But the competitions that the Math Arena tests LLMs on only “evaluate final numerical answers and do not require rigorous proof-based reasoning essential for most mathematical tasks,” according to a team of researchers at ETH Zurich and Sofia University’s Institute for Computer Science, AI and Technology (INSAIT). 

• The researchers tested six state-of-the-art LLMs on the 2025 USA Math Olympiad (USAMO), a competition consisting of six problems that require detailed proofs — not just final answers — for credit. Each LLM solution was graded by two independent expert judges. 

• Each problem is scored out of seven points, for a total of 42 points. And, across all 150 evaluated solutions from the models, not a single one attained a perfect score. In fact, the highest average score achieved by any model (R1) fell bellow 5%, with a total score of two out of 42. The researchers tested Google’s Flash Thinking, DeepSeek’s R1, OpenAI’s o3-mini and o1-Pro, Claude 3.7 Sonnet and QWQ. o3-mini was the lowest-scoring model, achieving a 0.9 out of 42. 

And, where humans tend to fail in solving these problems by, you know, not finding the answer, all of the models tested “consistently claimed to have solved the problems.” Where they failed was instead in “flawed logic,” “unjustified reasoning” and “misinterpretations of previous progress.” Their reasoning additionally lacked creativity, meaning “each model often attempted the same (and wrong) solution strategy across all attempts, failing to explore alternative approaches.” 

The area the models performed strongly in involved algebraic and arithmetic computations. 

Why it matters: The research — which, it must be noted, has not yet been peer-reviewed — highlights a fundamental flaw of LLMs: results, in this case, of a mathematical nature, “derived using these models cannot be trusted without rigorous human validation.” 

• Invest in Sembly AI, The Future of AI Workplace Tools. 60+ investors have already contributed over $370,000. 1000+ B2B clients, ProductHunt #1 of the day. Join us & help shape the future of work! *

• Alibaba preparing for flagship AI model release as soon as April (Bloomberg).

• US stocks suffer worst quarter since 2022 ahead of ‘Liberation Day’ tariffs (Semafor).

• This is how China skirts U.S. chip bans (Rest of World).

• ChatGPT isn’t the only chatbot that’s gaining users (TechCrunch).

• Investors hope April 2 could bring some tariff clarity and relief. That may not happen (CNBC).

OpenAI raises record $40 billion — don’t call it a bubble

OpenAI on Monday announced the closure of a record $40 billion funding round — the largest private tech funding round on record (by quite a large margin) — that valued the company at $300 billion, nearly double its previous valuation of $157 billion. 

It comes at a rather inauspicious moment, when the valuations of publicly-traded AI-related companies have been in full retreat through the first quarter as investor sentiment toward the Magnificant Seven has, rather violently, shifted.  

Here’s everything we know about the round: Fitting for a company with as strange a history as OpenAI, the terms of the fundraise are complicated. 

• The round will be led by Japan’s SoftBank, which will be contributing $30 billion. Microsoft and Thrive Capital are also joining in, at unknown amounts. 

• OpenAI, however, isn’t going to get it all at once. The initial funding — to be received in the next two weeks, according to SoftBank — will be $10 billion, while the remaining $30 billion will come by the end of 2025. 

SoftBank, however, added in a disclosure that its full $30 billion commitment is fully contingent upon OpenAI converting to a for-profit organization by the end of 2025. If it does not do so, SoftBank said it could slash its investment to just $10 billion. 

The terms are similar to those attached to the $6.6 billion funding round OpenAI secured as recently as October, 2024, a round that required OpenAI’s transition to a for-profit within two years, or investors can ask for their money back. 

SoftBank said it plans to finance the initial $10 billion investment through borrowings from banks and other financial institutions. SoftBank reported a net loss of $2.4 billion for the quarter ending Dec. 31.

This is all in addition to the $2.2 billion SoftBank has already invested in OpenAI beginning in September of last year. See, OpenAI is SoftBank’s “most important partner.” As SoftBank notes, the reasoning here hinges entirely on the advancement of artificial general, and then super, intelligence, two hypothetical, scientifically dubious pursuits whose potential existence has no grounding in evidence-based science. 

It’s also an addition to Project Stargate, the $500 billion U.S.-based data center buildout project that will be fronted by OpenAI and SoftBank. To that end, $18 billion of this $40 billion funding round will be going toward OpenAI’s Stargate commitments, according to CNBC.

The business: OpenAI reportedly has 20 million paying subscribers and 500 million ChatGPT users, though its method of measuring active users is quite unclear. Either way, fewer than 5% of its total subscription base is monetized. 

• As I mentioned yesterday, this round places OpenAI among the largest companies in the world by valuation, with a noticeable difference in that OpenAI is operating a money-losing business at a wildly extreme scale. 

• For example, both Verizon and AT&T have market caps of right around $200 billion, meaning OpenAI is now valued far above both of these companies. But both earned more than $120 billion in revenue last year, and both reported earnings above $2 per share. 

OpenAI, meanwhile — which lost $5 billion in 2024 on $4 billion in revenue — expects to burn a total of $44 billion between 2023 and 2028, before finally making a $14 billion profit in 2029, according to The Information. And that’s if all goes well. That profitability projection is based on an assumption that the company will generate more than $100 billion in revenue that year. 

The landscape: Beyond the challenging landscape of the public tech sector, OpenAI is actively facing a lawsuit from one Elon Musk over the very question of its pending conversion to a for-profit enterprise. A judge recently agreed to expedite a trial to Fall of 2025, the timing of which makes OpenAI’s planned conversion … interesting.

OpenAI did not return a request for comment on the above. 

• Then, in the midst of all of this, there’s DeepSeek and the rise of open-source generative AI. According to The Information, a number of larger enterprises, including Palo Alto Networks, are planning to spend less on GenAI from providers like OpenAI, since models from DeepSeek can “handle the same tasks for just 5% of the cost of the OpenAI models.” 

• This coincides, of course, with ongoing challenges in securing enterprise adoption due to concerns of privacy, security, reliability and bias. 

In recognition of this shifting dynamic, CEO Sam Altman said recently that OpenAI plans to release an open-weights model “in the coming months.” 

Much of the current environment — wildly high, speculative valuations based on technological optimism — seems to mirror the dot-com bubble of the early 2000s. 

OpenAI does not seem to be comfortably operational. In October, it secured $6 billion in funding and another $4 billion in debt; just a few short months later, despite all the supposed user growth, it’s getting another $10 billion injection with another $30 billion on the way.

And in recent days, Altman has posted multiple times about the “biblical demand” OpenAI’s been experiencing in light of its copyright-questionable Studio Ghibli viral trend, saying that “our GPUs are melting.” 

Altman said Tuesday that “we are getting things under control, but you should expect new releases from OpenAI to be delayed, stuff to break and for service to sometimes be slow as we deal with capacity challenges.”

The fresh funding round seems like it was vitally needed, which is wildly demonstrative of just how much of a cash burn this business is.

Taking the time to respond to a user who highlighted OpenAI’s flawed business model, Altman wrote: “as a business we are doing really great but thanks for your concern.” 

There are two sides to this field. Tools and gods. (We coincidentally highlighted this in the latest episode of our podcast). OpenAI is intent on building digital god, and SoftBank has decided that digital god is worth $30 billion (+). 

But, as much as that’s OpenAI’s focus, the company isn’t selling Hal 9000 (in part, because it doesn’t exist). It’s selling a powerful but inherently flawed chatbot at a time when chatbots — with or without reasoning — are becoming a commodity.  

There is no moat. There is no GPT-5. There is no super intelligence. And there is no evidence that one will eventually exist.

What there is, is a very large bubble.

🤔 Your thought process:

Selected Image 2 (Left):

• “In Image 2 the coyote’s coat is starting to shed. There’s atmospheric haze. Plants are emerging. Melting snow patches in the distance. It’s springtime!”

💭 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 Apple’s Health Coach:

32% are not excited about it, over data privacy concerns. 26% are. The rest, not so sure.

No:

• “This is a rapidly losing battle, but I would prefer it if Big Data didn't know more about me than I do.”

Maybe?:

• “Apple already knows everything about me — and so far — I don't feel that it has done nefarious things. I love my apple watch, I love using the fitness apps, and other health features. I think they are the only viable option to do this at scale, and at a reasonable price ...”

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.