Time for a catch-up!

Well, it’s finally another newsletter after weeks without one. My excuse, such as it is, is that I’ve been super busy, particularly following the release of the EU’s AI Omnibus, which I’ve already posted about, and I’ve been struggling with the recent flu bug.

My issue now is that there’s been so much happening, that like the famed Gordian knot, it’s difficult to know how to unpick all of the multiple intertwined stories. I had originally planned to try and produce a brief overview of key points and then publish a series of deeper dive posts, but whilst I may still do the deep dives, I will now try and summarise the last 4 weeks.

There’s been a lot of news about the size and value of various companies, to the point where it has been distracting. However, whilst the likes of Nvidia have published news of massive growth and increased revenue, other key players have started to seed the message that there might actually be an AI Bubble. Why the sudden admissions? Well I’ll give my views on that below.

The other key event, that attracted little media attention in my view, was the launch of the Genesis Project, described by the White House as the federal government’s new “Manhattan Project for AI”. For me, despite the glossy packaging, in reality it feels like its derived from the plot of a dystopian novel.

Anyway, let’s dive right in on all of the other news from the last few weeks:

Big Sharks vs. small fish

Well of course we have to start with OpenAI, which as always was in the news.

We’ve just passed the 3^rd anniversary since ChatGPT was released to the public and changed the World forever, as we all discovered generative AI for the first time. Up until recently, OpenAI have always led the pack, with the other big players (and occasional small one) rushing to catch up.

However, in a sign that things may be shifting, Google released Gemini 3 and Nano Banana, effectively upending the race for dominance. These products were so good and made such a big impact that Sam Altman declared a “code red” in response. This basically is the equivalent of “Huston we have a problem”. No drill, no games, but rather a serious threat to OpenAI’s business.

10 days later, and just over a month after releasing ChatGPT 5.1, they release 5.2 well ahead of their planned schedule. It has yet to load onto my account, but by all accounts, it has a greater focus on enterprise level business, rather than individual users.

GPT-5.2 Instant, Thinking and Pro have started rolling out for paid ChatGPT plans, with API access already live, and it is said to bring improvements on spreadsheets, presentations, coding, long documents, tool use and image understanding.

That is also interesting, as OpenAI released new data showing strong growth in enterprise use of its tools, with ChatGPT message volume rising eightfold since November 2024. The Ramp AI Index, states that 36% of U.S. businesses using enterprise AI tools now choose ChatGPT Enterprise, compared with 14.3% using Anthropic.

However, Google Gemini presents a real threat to OpenAI’s market share, both at the enterprise level, but also with consumers where OpenAI derives most of its revenue. When you them take a step back and look at the wider picture, you realise that Google have quietly put in place lots of the foundations to out-manoeuvre OpenAI and others.

News broke that Google is now reportedly in talks to supply Meta with billions worth of AI chips, and when combined with the news of its new Ironwood chip, this positions Google as also directly competing with Nvidia. It hasn’t happened yet, but this could be the start of a powershift, particularly at enterprise level, as it means that there is a real alternative to both OpenAI and Nvidia. The markets obviously agreed, as Nvidia shares dropped 6% on the news, whilst Alphabet rose by a similar amount.

That wasn’t all. There were multiple stories in the last few weeks about OpenAI going for an IPO with rumours of a potential valuation of over $1 Trillion. There were literally pages of reporting detailing the various monthly, quarterly and annual revenues with 101 variants on what it means for a future floating on the open market.

However, a majority of observers expressed concern that the numbers still don’t add up, and that the breakeven figures for 2030, which run into the hundreds of billions of dollars, are just not realistic or feasible.

This leads nicely into my own thoughts about the AI bubble. It has been noteworthy that numerous key players like Sam Altman from OpenAI, and Google CEO Sundar Pichai have all voiced a view that “it looks like a bubble’. Some, such as Hugging Face’s CEO, have suggested it only potentially impacts LLMs and not the wider infrastructure, but given how inter-connected and inter-related those are, any failure cannot help but have a knock-on effect.

My view is that the bubble will burst at some point in the next 6 - 12 months, and like with the Dotcom bubble, it is likely to have a massive negative impact, not only on shareholders and their savings, but the wider economy as well. However, just as the Dotcom bubble did not bring an end to the internet or computers, so a bursting AI bubble will not stop the progress and development of AI, but it is likely to kill-off multiple smaller companies that don’t have the cash reserves or revenue to survive the “storm”.

We then zoom out even further and look at how the US government in particular may react to such a market crash. Firstly, to understand the scale of the problem, you need to take into account the fact that according to JPMorgan, 75% from the share of S&P 500 returns since late 2022 that have come from AI-related stocks.

Given the current administration’s support for AI, and its close political ties to Silicon Valley, you have to wonder whether, as with the banks during the financial crisis, the view will be that some companies are simply “too big to fail” – this is particularly true when those companies are also deemed to be critical to national security.

Yet, currently, certain sectors are still booming and clearly are working in a strong marketplace with excellent returns in terms of revenue. Nvidia in particular continues to grow strongly, not only exceeding sales expectations, but also with order books that are helping to maintain that boom as demand outstrips supply, but it’s not just Nvidia.

In November Samsung increased some prices by up to 60% as demand collided with constrained supply, and manufacturers and cloud providers are reportedly panic-buying to secure supply. The squeeze is being driven by AI data centres hoovering up DRAM, leaving less for phones, laptops and cars. Analysts are also warning that consumer devices and auto makers could face higher costs and potential shortages into 2026.

Yet in the midst of this Samsung has announced a $310bn, five-year investment push into semiconductors and AI infrastructure, including a new chip factory and AI-focused data centres.

You then have Elon Musk talking about needing 100 - 200 billion AI chips per year for Tesla/SpaceX. Given the fact that Nvidia shipped only 4 million Hopper GPUs in the last 2 years, this is almost comical but again underlines the huge demand.

Everyone is clearly betting this demand is structural, not a blip, but like musical chairs, one wonders what happens when the music stops? Again, on top of all of this, we have an emerging problem that the demand for power (and water) is fast exceeding supply. This is going to place immense structural pressures on electricity prices, with the real danger of businesses out-bidding normal consumers – potentially driving up inflation and impacting the wider economy.

To illustrate the power problem xAI has told local planners in Memphis that it plans to build a solar farm next to its Colossus data centre, one of the largest sites in the world for training AI models. The solar site would cover 88 acres and is expected to produce around 30 megawatts of electricity, which is only about 10% of the data centre’s estimated power needs. Separately, xAI has announced a 100-megawatt solar project nearby, paired with large batteries to provide continuous power. The developer, Seven States Power Corporation, received $439 million not from the U.S. Department of Energy, but from the Department of Agriculture, most of it as an interest-free loan.

Back to software and it seems that all of the big players are adding “shopping” features to their LLMs. As part of a broader set of Gemini-powered travel features, Google rolled out its AI “Flight Deals” tool globally, baking it into Search.

Perplexity also launched an updated AI shopping assistant for users in the United States. Built with PayPal’s help, it gives people a smoother and more personal shopping flow right inside their chats.

Their assistant gives custom answers based on a user’s lifestyle, needs, and taste, with products appearing in clean info cards with pros, cons, key details, and data from trusted reviews. With Instant Buy, users can then pay right away using their saved PayPal details.

Last but not least is OpenAI who released a string of new features. They have partnered with Instacart to launch a full shopping and checkout flow directly inside ChatGPT. Users can plan meals, adjust diets, build carts, and check out without leaving the chat.

Instacart already used ChatGPT for recipes, but this takes it into full agentic commerce, with OpenAI earning fees from transactions. The grocery angle is less important than the strategic direction. ChatGPT is no longer about just helping you decide, it is now actively helping you to complete the task. Going back to the all-important enterprise market, you can start to imagine how AI agents could handle revenue tasks, not just insights.

OpenAI also now supports regional data hosting, letting enterprises pin ChatGPT data to specific geographic locations. Up until now that had been a massive problem for enterprises working in regulated industries.

They also added ChatGPT Voice directly inside text chats, making voice a seamless input method instead of a separate mode, which makes compliance conversations easier and opens up the potential for broader deployment. Don’t think of this as a consumer feature but rather imagine how it’s going to alter how internal tools, agents, and customer systems are designed.

I will note that Grok 4.1 and Claude Opus 4.5 were also launched in the last few weeks, and whilst both returned impressive results, I won’t get bogged down in the details, as indications are that all of the big players are likely to release yet more new models before the end of the year.

I will end this section with the news about the “start-up”, Project Prometheus. Given the fact that it is launching with $6.2bn in funding, it hardly fits the image of a “start-up”, but then you realise it is partially funded by Jeff Bezos, who will run the company with Vik Bajaj, a former Google X and Verily executive known for his work in science and advanced tech.

What we do know is that its focus is on AI that supports engineering and manufacturing across areas like aerospace, computing, and cars.

AI for physical science and engineering has become a growing area in the wider AI race, and I will attempt to delve into this and issues like robotics in a post in the coming days.

Legislation, policy and other news

So, from Project Prometheus let’s move onto the Genesis Project, where the US government are pushing for global dominance, but deciding to remove the guardrails in order to achieve it. Whilst it is presented as power-charging new developments in everything from medical care to business, and from climate to defence, it also potentially bypasses all of the safeguards that until now we’ve taken for granted.

This in part has been driven by the likes of Peter Thiel and the team behind Palantir. Palantir’s CEO Palantir's CEO even went on The Axios Show and made the argument that Americans should accept a surveillance state as the price of beating China in AI. His logic? "You will have far fewer rights if America's not in the lead."

The administration’s view is that this is about more predictable regulation, clearer procurement paths, and a shared baseline for what responsible AI should look like at the national scale. This was underlined by Trump’s recent Executive Order blocking States from setting local AI-related legislation, something that will be challenged in the Courts. Beyond the internal US politics and the rights of States, there is the real concern over human rights, where data becomes the key resource.

Staying on the topic of a “AI sovereignty”, German software giant SAP launched EU AI Cloud, a European-focused platform that gives organisations complete control over their AI and cloud infrastructure while keeping data within EU borders. The service includes partnerships with Cohere, Mistral AI, and OpenAI, offering multiple deployment options from SAP’s own data centers to on-premises installations for enterprises with strict compliance requirements.

This represents AI sovereignty as European organisations demand local control over critical infrastructure. With stricter EU data protection rules and growing concern over dependence on US cloud providers, SAP's approach shows enterprises a pathway to advanced AI capabilities without compromising regulatory compliance or operational control. The importance of this independence was underlined by outages in the last month involving both AWS and Cloudflare.

Talking of Amazon, they have also announced plans to invest more than $35 billion across all its businesses in India up until 2030, on top of the $40 billion invested in India so far, focusing on business expansion as well as three strategic pillars: AI-driven digitisation, export growth and job creation.  

So that’s all for now from me at the start, rather than the end of a busy week … which will include a major announcement and explanation of why I’ve been so busy recently … so stay informed, stay critical, and wherever possible - stay ahead.

Regards

Tom Carter