Time for a catch-up!

OpenAI’s release of Chat GPT5 continues to dominate the headlines this week, although if you were to speak to the board of OpenAI, they would probably suggest for all the wrong reasons. After the initial euphoria and excitement, the mood turned quickly to disappointment, and in some cases outright anger.

Why? Well, despite excelling at many of the benchmark tests that LLMs are judged on, and despite having some amazing new capabilities, it seems everyone’s expectations surpassed the reality. Perhaps more damaging was the way in which the transition was managed.

People had established “relationships” with their AI, giving them a name and using them like some form of virtual emotional crutch. I won’t get into my views about how weird all of that feels to me, but it’s fair to say that OpenAI simply kicked that “crutch” away and watched as many of their most loyal customers face-palmed into the floor.

Gone was the option to choose which model you wanted to work with from a drop-down list. In its place you have a choice of GPT5 or … nothing. Yes, it’s technically more efficient, faster and capable than GPT 4o, but the lack of choice offended people. It reminds me of when Apple made the decision to upload the U2 album onto everyone’s device without asking. Were OpenAI’s marketing team asleep during that lesson at marketing school?

That then led people to start complaining that the upgrade in capabilities was not as much as they expected, referencing the “jump” from GPT 2 to GPT 4o. This then led down a whole new path of discussions around whether OpenAI had misled them, and whether AGI (Artificial General Intelligence) was as close as Sam Altman and others have ben suggesting.

This may not have been a “Dot-Com Bubble” bursting moment, but it certainly dented enthusiasm across the board for AI, particularly amongst the press. To his credit, Gary Marcus (excellent substack account worth following) has been warning for years that this was going to be the case, and I have to agree with him that this raises lots of uncomfortable questions about where we go next, since simply scaling up computing capacity is bringing incrementally smaller gains.

Anyway, I will come back to this conversation from another perspective later, but in the meanwhile let’s find out what else has been happening. With that in mind, let’s dive right in on the other news from this week …

Big Sharks vs. small fish

The other big news this week was connected to my other favourite topic of the moment, namely the impending battle between search engines and AI LLMs. In what can best be described as a bold and audacious move, Anthropic, the company behind Perplexity, made an offer to Google to buy its Chrome browser for $34.5 billion.

This was a cash offer that includes a $3 billion investment and a commitment to keep Chrome open and retain Google as the default search engine, the latter part raising antitrust and strategic implications. Given the fact that Anthropic itself is currently only valued at $18 billion, this was all the more surprising.

However, it was also a supremely clever move. Google are awaiting a judgement in the next few weeks on claims that it is a monopoly, that will decide whether they will have to sell off parts of their business, including the Chrome browser. OpenAI and xAI have also expressed an interest in the browser, but Anthropic’s move definitely stirred things up.

In other news, Google pledged $9 billion for AI and cloud expansion in Oklahoma over two years, scaling its AI and cloud infrastructure, boosting regional tech capacity and laying the groundwork for advanced AI deployment. Google Cloud and Wipro also rolled out 200 AI agents across industries. The strategic collaboration will deploy agentic AI solutions in sectors like healthcare, finance, insurance, retail, manufacturing, and IT to enhance efficiency and decision-making.

However, the fallout from the GPT 5 release again raises the question of the likelihood of investors achieving any return on investment. For now, it does not seem to have dented investor confidence in OpenAI, even if it was an unmitigated PR disaster.

However, it will be interesting to see if the investor mood changes, particularly given the fact that GPT 5 was released for free, which has to raise the question of when all of this investment will bring some actual profit.

A recent Economist article highlighted that for the big players, hard assets such as infrastructure, property and equipment now account for 60% of their equity book value, up from 20% over the last 10 years. Between them, the big players will spend $400 billion on data centres and infrastructure this year, and depending on who you listen to, they will spend either $2.8 trillion by the end of 2028 (Morgan Stanley), or $6.7 trillion by 2030 (McKinsey).

Not surprisingly, in a battle against these “big sharks”, the smaller “fish” struggle to make money or gain market share. This was illustrated this week when both C3.ai and BigBear.ai both missed their earnings targets, resulting in their stocks falling 13.7% and 29% respectively.

Legislation, policy and other news

In what is best described as an unusual move, the U.S. government has agreed to permit Nvidia and AMD to resume AI chip sales to China in exchange for a share of the revenue. The 15% levy on their export revenue may yield a few billion dollars, but long-term strategic risks, like strengthening Chinese chip independence, persist. 

In less positive news, research has revealed that modern AI chatbots may be prone to personal information exploitation, raising fresh data protection and privacy concerns as AI becomes embedded in daily workflows. Researchers also managed to hijack Microsoft’s copilot studio agents and got them to reveal their tools and share data from a CRM.

That’s all for now from me at the end of yet another hectic week … so stay informed, stay critical, and wherever possible - stay ahead.

Regards

Tom Carter