Time for a catch-up!

I am thinking about possibly renaming this the OpenAI newsletter, since they seem to be making headline news with any day containing the letter “Y”, and there’s enough just talking about their activities to probably fill two or three newsletters each week at the moment.

The scale of their expansion and news of the tools that OpenAI are releasing is mind-boggling, coming in a week when they were declared the most valuable private company in the World, with a value of $500 billion and 800 million weekly active users. Given the fact that 99.99% of the World’s population didn’t even know who they were 3 years ago, that’s impressive growth by anyone’s standards.

However, I did find it ironic when it was revealed this week that OpenAI warned EU antitrust authorities that dominant tech firms like Google might misuse their infrastructure advantages to stifle competition in AI, particularly given their own expansionist moves over the last 6 months.

Anyway, let’s dive right in on all of the news from last week and find out more about those moves:

Big Sharks vs. small fish

OpenAI continued on from last week’s disruption of the news cycle with their annual Developers Day. Sam Altman announced four new development tools, which I summarised earlier this week in a LinkedIn post. The short version is as follows:

The first two new tools were about apps. For the end-user the interesting one is the ability to call a particular app whilst you’re actually in a chat. You can either request a specific one or have ChatGPT suggest one. For me, this again is about OpenAI building out their ecosystem, so that you never have to go “outside” it.

The second tool is an App SDK (software development kit) aimed at programmers. This will allow them to easily build apps that can then be placed in a public directory and subsequently monetised – a little bit like the Custom GPT library that already exists.

Then there were developments about APIs, which again will have an effect, but it’s the technical changes that the public (including me) don’t fully understand or really care about.

The last one though was the significant one. This is an Agent Builder, a Connector Registry (linking to your data sources), ChatKit that allows you to add chatbots to your products, and finally Evals 2.0 that helps evaluate your agents. Currently this is aimed at developers, but I am certain that in time it will be rolled out to the general userbase.

For me, this is again about OpenAI trying to build out an ecosystem of products, providing a complete package to ensure that customers don’t want to go anywhere else.

However, they then announced a multi-billion-dollar partnership with AI chipmaker AMD as part of their ongoing efforts to accumulate as much AI infrastructure as possible (see their earlier complaint about Amazon, Meta . This clearly was good for AMD shareholders, who saw their stock soar by 40% after the news, and it also puts AMD in a stronger position in their competition with Nvidia.

As part of the contract AMD will supply high‑performance AI GPUs and support up to 6 GW of compute over time. OpenAI also receives warrants to buy up to 10% of AMD’s shares, contingent on milestones. AMD also had other news this week with the news of ROCm 7.0, a major release of its open-source software platform for GPU computing, designed to accelerate AI and HPC workloads on modern GPUs (I’ll explain the relevance in a minute).

But let’s talk some more about OpenAI. This year so far we have seen them partner with NVIDIA for up to $100 billion for AI chips and systems, Oracle for $300 billion in relation to data centres over the next five years, CoreWeave for $22.4 billion in relation to cloud computing, and Hitachi for an undisclosed sum for power systems and cooling technology.

If that wasn’t crazy enough, this year alone they’re planning to spend $16 billion just on renting computer servers, but by 2029, they want to increase that up to $400 billion annually.

Yes, you read that correctly - $400 billion every year.

They are clearly trying to compete with Google and Amazon, and they’re trying to reach the same scale in 4 – 5 years, that it took those two companies more than 20 years to achieve.

The big question for me though, is how they will generate the power to run these data centres (hence why the more energy-efficient AMD chips are important !!), a topic that I’ve mentioned before, which carries significant risks.

There was some other news not involving OpenAI though, including Google introducing Gemini Enterprise, a platform that brings together all its AI tools for easier use by organizations.

The platform gives access to Veo 3 video AI, Gemini 2.5 Pro, and other Google AI models with smart agents, and adds a new Data Science Agent automates data handling and model training. The new model enables AI to navigate the web browser like a human, filling out forms, dragging UI elements, clicking, etc., rather than via APIs. It also includes a workbench for data analysis and workflow automation inside companies. This again underlines my assessment that all of the major players are building complete ecosystems in an effort to maintain market share.

That wasn’t the only major player with enterprise level news, with IBM announcing a new partnership with Anthropic, adding the LLM into some if their products, starting with their integrated development environment. IBM also announced it created a guide in partnership with Anthropic on how enterprises can build, deploy, and maintain enterprise-grade AI agents.

As part of this news, IBM also announced a deal to bring Claude to consulting giant Deloitte, who boast a global workforce of almost 500,000. Deloitte probably needs new AI tools, as it was announced this week thatthey had to partially refund the Australian government for a $440,000 AUD ($290,000 USD) 237-page report, which reviewed the Department of Employment and Workplace Relations’ IT systems, because it contained fabricated quotes, non-existent academic references, and other AI-generated errors.

Not wanting to be outdone Amazon made a big announcement about humanoid robotics, with their FAR team unveiling OmniRetarget.  This data generation engine is built to train humanoids in complex, full-body loco-manipulation. What does that mean – well think walking, climbing, lifting, and jumping with human-like precision. They demonstrated this with a short video, but it’s clear that robotics is also developing at breakneck speed.

Finally, amongst the big players, Meta signed a $14.2 billion, multi‑year supply agreement with CoreWeave through 2032, in order to secure computing capacity. 

There were also other releases, which were over-shadowed by the OpenAI announcements. This included ElevenLabs’ brand new on-device TTS speech language model, which allows for a realistic speech LM that runs locally on your device. This potentially unlocks a whole new category of voice agents, assistants, toys, and compliance-safe apps. Also, Supermicro unveiled new systems optimised for both edge AI and data centre workloads, again competing in terms of energy efficiency and compute capacity.

To end this section, there was potentially some bad news for OpenAI, with reporting suggesting its partnership with Jony Ive of Apple fame is running into problems, with them facing significant technical challenges as they work to develop a screen-less, AI-powered device. It now seems unlikely that they will launch the new product as scheduled in 2026.

Legislation, policy and other news

Following on from California signing the Transparency in Frontier Artificial Intelligence Act last week, Senators Hawley and Blumenthal introduced the Artificial Intelligence Risk Evaluation Act to Congress. It would require advanced AI systems to undergo review via the Department of Energy before public deployment. 

The EU also announced its “Apply AI” strategy, to reduce dependence on U.S. and Chinese AI. The European Commission plans to promote European AI tools in public administration, defence, and industry, backed by €1 billion in funding. 

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

Regards

Tom Carter