Image Credit: Noam Shazeer of Character.AI has returned to Google, along with his model. Winni Wintermeyer for The Washington Post / Getty Images.
Hey Everyone,
Read this post on a web browser for the best reading experience.
I wanted to ground some of my thoughts around what I’m witnessing in BigTech, at OpenAI and the Generative AI startup ecosystem. Covering AI also means its positioning in the real business world, on the stock market and in how major business models like Cloud computing are impacted by it. A correction in the Generative AI bull market and a slew of exits of the startups in the space, along with a delayed Nvidia “Blackwell” B200 AI chips shows an odd situation.
Generative AI and OpenAI itself, appear to be at a significant cross-roads.
In a big move, Character.AI co-founder and CEO Noam Shazeer is returning to Google after leaving the company in October 2021 to found the a16z-backed chatbot startup. Read more.
Microsoft finally has acknowledged to shareholders that OpenAI is now a competitor in AI and search. Even while they get 75% of their revenue of a startup that’s losing $5 Billion just to “chase AGI”. Go deeper.
OpenAI have released potentially impactful products in SearchGPT, advanced Voice Mode, and a very affordable “GPT-4o Mini” – fairly great potential products. But its path to profitability needs an AGI miracle and remains fairly bleak even if you are a Techno-Optimist.
Other macro factors are beginning to take shape that might have an impact here:
Shahm Rules, not Sam Altman
The reality is this July, 2024 – The US economy may have entered a recession as the unemployment rate rose to 4.3% in July. Sahm Rule, a recession indicator. According to data from the Bureau of Labor Statistics released Friday, the unemployment rate rose to 4.3% in July, up from 4.1% in June and from recent lows of 3.4% in April 2023.
The increase officially triggered the Sahm Rule — a recession indicator developed by former Fed economist Claudia Sahm — which says that the US economy is in a downturn when the three-month moving average of the unemployment rate rises by 0.5% from its 12-month low. Read more about it.
It’s not clear with a recession, even a short one, what that might do to the enthusiasm and funding opportunities for players like OpenAI, Anthropic and their peers. Many of whom don’t have a very high probability of existing in five or even three years’ time.
I’ve been trying to wrap my head around BigTech capex investments in AI and how they will lead to ROI. The math is a bit fuzzy in mid 2024.
Hyperscale Bloat around CapEx in Generative AI and Datacenters
Earnings provided us with more data of how BigTech hyperscalers are investing in Capex around AI for the long-term, inspite of little ROI in 2024 and perhaps even in 2025. (source: This tweet)
Quarterly capex change for the hyperscalers:
🟣 Amazon: $11.4B -> $17.6B = 54% Y/Y Growth
🟢 Microsoft: $8.9B -> $13.9B = 55% Y/Y Growth
🟠 Google: $6.9B -> $13.2B = 91% Y/Y Growth
🔵 Meta: $6.2B -> $8.5B = 36% Y/Y Growth
Nvidia is reliant on the AI chip spending of Cloud players and those that can use Generative AI hype to grow their products faster like in Cloud and digital Advertising and expensive frontier models like xAI are developing. The gigacenters of AI compute required are enormous gambles for them ever providing business value and ROI. Can Grok save Twitter or magically reveal functional Tesla Robotaxis with Tesla and xAI’s huge datacenters? Sometimes it’s that far-fetched.
The problem also for Hyperscalers is not all of them are doing great, as Amazon missed guidance badly. Generative AI is simply their ticket to faster Cloud and Ads growth. For BigTech, Generative AI is not useful in and of itself necessarily, it appears in blunt terms mostly useful for shareholder swag and Cloud credibility (keeping growth higher for longer). Meta’s advertising sales in the second quarter increased 22%, double the growth rate at rival Google, which is faster than even Amazon’s surging Advertising business (20%).
How much Meta’s capex in AI impacted their Ads growth is very debatable. Zuckerberg said AI is “improving recommendations and helping people find better content, as well as making the advertising experiences more effective.” It’s not clear however, if that is actually true or related to their Generative AI efforts at all.
Capital expenditure in Datacenter Expansion is Crucial
Since Microsoft and Amazon are so reliant on revenue from the Cloud, Capital expenditures in datacenters is not an investment for 2025, but an investment for the decade ahead. You can listen to the Microsoft Earnings call (transcript here) and try to make sense of the statement of their CFO.
Amy Hood Microsoft’s CFO said of the capex: “Cloud and AI-related spend represents nearly all of our total capital expenditures. Within that, roughly half is for infrastructure needs where we continue to build and lease data centers that will support monetization over the next 15 years and beyond. For Microsoft, don’t expect ROI by 2025 or likely in the 2020s at all.
CFO Amy Hood pointed out that a lot of Microsoft’s AI spending will be monetized “over 15 years and beyond.” It’s less clear if shareholders will be that patient.
It’s not clear to many analysts how Capex in Generative AI and datacenters scales well if Generative AI has an ROI problem. If Generative AI isn’t generating revenue that’s rational compared to the capital expenditures in the AI chips, datacenters, , models, products and Cloud growth of MAGO (Microsoft, Amazon, Google, Oracle), there will be an implosion of value in the space eventually. Not just by Nvidia’s falling stock due to less sales, but on the Hyperscale valuations side as well.
Source: Brian Potter, here
It’s not clear how OpenAI Survives
“OpenAI is inextricably tied to Microsoft. It is bound to use Azure, Microsoft’s cloud compute platform, both by its agreements and the fact that the majority of its funding is in credits that can only be used on Microsoft Azure. Microsoft sells access to GPT through Azure, while also directly competing with it with its own upcoming model, and takes three-quarters of any of the (theoretical) profit that would come out of OpenAI’s services.” – Ed Zitron.
As hyperbolic as ChatGPT was a Venture Capital baked narrative, and as impressive as OpenAI sales are, they aren’t enough. How can it be worth $80 Billion on operating costs of $8.5 billion on AI training and staffing? A $5 Billion loss that doesn’t count the most of its actual revenue (75%) goes to Microsoft? Nothing really adds up here for how investments and startups should work.
Ed Zitron recently asked how does OpenAI Survive in a 31-min read. It’s probably not the time for Sam Altman to keep making nebulous statements about AGI. OpenAI would have to successfully fend off challengers like Google, Anthropic and others that may emerge from China or xAI.
OpenAI would need to raise more money, a lot more money that would further dilute its shares.
GPT-5 (2025) and GPT-6 (2026) would have to be successful and have Agentic AI relevance in work automation.
Timing of GPT-5 and GPT-6?
I believe GPT-5 comes out in Early 2025
But you could make the case it would arrive even later.
OpenAI Needs a Breakthrough by 2027
To truly survive its financial difficulties OpenAI would need a breakthrough in what GPT-5 or GPT-6 enables us to do with Generative AI and LLMs.
What does OpenAI’s Board Think about AGI?
According to OpenAI board member Adam D’Angelo, that milestone is likely to happen “within five to 15 years,” Seeking Alpha reported Monday (July 29). Unfortunately while OpenAI needs something by 2027, it looks more probably even in its own sad definition of AGI, it might take anywhere from 2029 to 2039.
That’s not a good omen.
OpenAI needs GPT-6 to be Actionable in Agentic AI
OpenAI likely needs serious revenue from SearchGPT in Ads by 2027 to become profitable. It may also need GPT-6 to be fairly smart in doing tasks that accelerates new revenue opportunities. Assuming here GPT-6 comes out in 2026, as the head of Microsoft AI has insinuated.
To thrive as a future entity OpenAI would also need to:
Navigate antitrust regulation, multiple lawsuits and global regulatory factors, including probes by the EU
Handle Microsoft’s complicated demands and competitive pressures
And finally have a marked impact on work via agentic AI to quote Zitron:
“Have these use cases be ones that are capable of both creating new jobs and entirely automating existing ones in such a way that it will validate the massive capital expenditures and infrastructural investment necessary to continue.” Read more.
Now of course it’s doubtful OpenAI can manage to do all of these things, it’s still possible it survives in some form.
OpenAI and Nvidia are Poster Children of Generative AI
While OpenAI has a very impressive valuation and Nvidia is now worth many Trillions, how they do in the next few years also is symbolic of the entire value that Generative AI is able to bring to the world in more concrete terms. ChatGPT and AI chip sales are impressive, but revenue and ROI would be better for a stable future for the movement.
Blackwell B200 AI Chips Delayed
Nvidia has reportedly told Microsoft and at least one other cloud provider that its “Blackwell” B200 AI chips will take at least three months longer to produce than was planned.
The delay is the result of a design flaw discovered “unusually late in the production process,” according to two unnamed sources, including a Microsoft employee, cited by the The Information. It’s not how clear how serious this might be for Nvidia’s late 2024 or 2025 revenues.
Nvidia’s Blackwell family is encountering major issues in reaching high volume production. This setback has impacted their production targets for Q3/Q4 2024 as well as the first half of next year. This affects Nvidia’s volume and revenue… – Semianalysis.
Character.AI selling out to Google
There’s been quite a few startups getting acquired of late or going for exits that don’t show strength in the B2C Generative AI market. Character.AI co-founder Daniel De Freitas is also joining Google with some other employees from the startup. Google is getting access to their models in yet another curious licensing deal that looks like reverse acqui-hires.
It’s distressing if you once had high hopes for the likes of Inflection (Microsoft), Character.AI (Google), Adept AI (Amazon), Leonardo AI (Canva), and others. Stability AI is somehow reborn, and always new startups enter the frying pan. What many of us predicted would happen regarding BigTech cannibalizing Gen AI startups and their talent is now occurring far more frequently. There was once the idea that Google would fund Character.AI a few months ago, but since then both Meta and Google have been cloning their concept, it led to a different outcome altogether.
So a lot of this recent news in this article points to a slowing if not failing Generative AI ecosystem. While VCs write their warped boosting narratives, a few are a bit more honest with what they have noticed.
Gartner Predicts 30% of Generative AI Initiatives Will Fail by 2025
New research from analyst firm Gartner suggests that at least 30% of generative AI businesses currently testing will be abandoned after proof of concept by the end of 2025. I can assure you it’s way higher than that because for the most part this tech does not deliver ROI as expected, promised or hoped.
Gartner found that early generative AI adopters are struggling with escalating costs and said that deployments can range in cost from $5 million to $20 million.
For example, designing a custom generative AI model, like fine-tuning a Llama model on industry-specific data, would cost a business $5 million to $6 million upfront and up to $11,000 in recurring costs.
In the world of business and real companies, this sort of adoption mostly will not make much sense. Unless you are JP Morgan and have a huge Tech budget and for R&D.
Generative AI in Finance, Banking and FinTech
JPMorgan Chase has given its asset and wealth management employees access to a generative AI platform, LLM Suite, which will work as a research analyst.
According to an internal memo seen by the Financial Times, employees of the investment bank will have their own version of a “Chat-GPT-like product,” that is to be used for “general purpose productivity.”
Generative AI is not a general purpose technology, but sure, it can save Finance some time and work-hours.
Doomers vs. Accelerationists – Crazy Expectations
While most common people trust AI less due to their hallucinating tendencies than before ChatGPT arrived, signalling a major trust problem, cheerleaders like Sam Altman whether you like him or not, also have been pushing for AGI, or the commercial version of such a Techno-optimist device.
Source: Peter Gostev
Sam Altman and the Rotten Narratives
Do and Ed Zitron have a point on Sam Altman, and that the AI he and his team have created are as manipulative potentially as him and this cult like movement?
For all the potential value created by Generative AI and their systems, we don’t talk about the dangers created nearly enough to society, cybersecurity and rule of law. Perhaps that’s why OpenAI itself captures our fascination being the best funded of such AI startups with an AGI narrative that is supposed to inspire awe and speculation indefinitely. At least that has worked.
Sam Altman may be to AI, what Sam Bankman-Fried was to crypto. A situation where so much deception and over-promising leads to many moral and ethical failings. OpenAI is also the manipulation of Silicon Valley over-promising personified, a common VC tactic of over-promising and repetition that stretches back well before the dawning of the internet.
Generative AI after two years doesn’t seem nearly as impressive as some of us would have imagined or hoped it would have be and OpenAI deserves a fair amount of the blame. There’s no tangible sense that AGI is around the corner and will be anytime soon.
OpenAI is not the Intel of this Age
Watching Intel crash on the stock market in the past week is a sign that the world moves us rapidly. Intel received as $8.5 billion in direct funding from the federal government as part of the CHIPS Act, has had to cut more jobs rather than create new ones.
Intel shares had their biggest drop since 1974 on Friday after the chipmaker reported a big miss on earnings in the June quarter and said it would lay off more than 15% of its employees. The stock is trading at 2013 levels.
So is OpenAI more like the WeWork or Napster of this era instead of the Cisco or Intel of the Generative AI age? Sam Altman knew how to hype a company, but it seems not how to run one.
While Warren Buffett raised Berkshire cash level to record $277 billion, Sam Altman’s OpenAI is at least $5 Billion in the red for 2024 alone, if you count the 75% they give to Microsoft, that number is likely closer to $8 Bn. OpenAI was founded in 2015 and can now barely compete with models from its spin-off Anthropic. While it has created an “OpenAI Mafia”, it is not and may never be profitable. What is the value to society of such a startup? And how can it continue to exist in such a state? It cannot.
AI bros just got really good at hyping their products in the early 2020s. But Generative AI is not producing nearly the economic value or boost to productivity we were told it would. In fact, it’s fairly likely BigTech are wasting Billions and the energy (and water) this gamble for AI supremacy requires, a commitment of Capex that’s nearly unimaginable to previous tech cycles.
If Facebook pushed for a schizophrenic Metaverse that never existed, what will the push for Generative AI be compared to years from now? We didn’t ask for this.
The Business fundamentals of OpenAI and their peers is bad, it’s worrisome bad. The likes of OpenAI and Anthropic will need to raise more funds and keep diluting their valuation, each time becoming more owned by the powers fated to acquire and swallow them. Monopoly capitalism does not only pick the winners in such an American tech economy, it eats their creations as well.
If you are an AI evangelist, it turns out it’s much more profitable to be an Ethan Mollick (boosting) than a Gary Marcus (harping), but which one is telling more the truth? And why does being objective here matter? It matters because people have a lot of financial incentives riding on the Generative AI hype. The business reality, ethics and financial matters tell a sobering story. If Techno-optimism is manufactured by VCs and BigTech for commercial purposes, it’s not real.
I have come across several pseudo-academic papers masquerading as studies in support of OpenAI and its products that are more like public relations than real science. That’s not to say that OpenAI isn’t trying or impressive on some advances.
OpenAI’s Product Efforts
Meanwhile OpenAI is constantly releasing new updates, new products and new features for its developers and users. While extremely unprofitable, controversial, closed and BigTech backed, their product efforts show top global talent trying to innovate.
OpenAI rolls out advanced Voice Mode to some users. OpenAI plans to bring the new mode to all ChatGPT Plus users in the fall.
OpenAI has rolled out a wait list for SearchGPT, that is their Perplexity clone with multiple deals with news publishers.
OpenAI along with many others have kept up the momentum of smaller, cheaper and more capable language models. GPT-4o Mini is fairly good.
OpenAI’s killer product, ChatGPT, has gotten really good traction in the early 2020s. They are the only Generative AI app that has really gone mainstream.
OpenAI’s GPTStore appears to have more or less failed to add much value, but perhaps it retained some ChatGPT users, at a certain point in time.
OpenAI’s ChatGPT Enterprise has likely been a commercial success due to Microsoft’s partner network and the perceived value of frontier models at big firms.
OpenAI is increasingly working with the DoD, Pentagon, the U.S. State and even, the Los Alamos National Laboratory for bioscience research. None of this should make you feel comfortable.
Is Goldman Sachs Right?
Investment giant Goldman Sachs published a research paper about the economic viability of generative AI which notes that there is “little to show for” the huge amount of spending on generative AI infrastructure and questions “whether this large spend will ever pay off in terms of AI benefits and returns.”
Hyperscaler BigTech might end up spending Trillions on Generative AI and have little killer products to show for it. Due to how shareholder pressures work Tech Titans need to appear like they leaders in AI, due to how many times they have mentioned “AI” on their earnings calls in the past few years. Microsoft’s Q2 Earnings showed revenue from Azure and other cloud services grew 29% during the quarter, some of it due to their leadership in Generative AI.
Generative AI can Boost Cloud Growth – Case Study – Azure
Of the 29% growth for Azure and other cloud services, 8 percentage points came from AI services, Microsoft said. That’s nearly 30% of the Cloud growth is coming from Generative AI which over many years it actually fairly lucrative. That’s not counting Microsoft’s revenue directly from OpenAI (75%) and its various Copilots including the evolution of Github Copilot Pro, including a lucrative Enterprise pro plan.
On this basis alone we can understand Microsoft’s capex on AI so far, while for Amazon, Google and especially Meta, it’s less clear to me. Microsoft has been the most aggressive in expanding new datacenters in South-East Asia and Europe for the 2024/2025 cycle.
Google has even more Generative AI products and influenced products than OpenAI, and likely with their ability to acquire and re-acquire talent have the most to gain in the long-run from the advances of Generative AI. However, their ability to execute products and marketing is less than convincing.
If OpenAI falters, Google and Anthropic are the big winners. Google also has funded Anthropic. As for Amazon, Meta, Tesla and xAI, their future in AI is far less certain to deliver ROI.
Due to its absurd capital expenditures and lack of a realistic path to profitability, OpenAI may end up not being able to move forwards so aggressively and may end up failing, most likely in the three to five year time period.
Things to Watch
If the U.S. enters a recession
If the Bull-Market in Generative AI is just correcting or enters a bear market
How China launches global products related to Generative AI
How regulatory, antitrust probes, lawsuits and regional differences in AI regulation impact OpenAI
The evolving relationship of Microsoft and OpenAI
If OpenAI can create another killer product ever again after ChatGPT.
Thanks for reading!
Read More in AI Supremacy
