Tech Catch Up – AI & GlobalFoundries

Misunderstood data

• The markets wobbled yesterday with the publication of a paper from MIT, which claims that most AI implementations are falling short, but this is not because AI is no good but because it is not being used properly.
• The report (see here) assessed 300 AI deployments at companies and concluded that 95% of them had had no tangible impact on productivity or generated any return on investment.
• This headline paints a pretty black picture for the AI market, but looking a bit deeper reveals a very different picture.
• The researchers concluded that the main reason for the failure to deliver a return is that the models have not been implemented properly.
• This makes sense as RFM Research (see here) has long concluded that the best use of an LLM in an enterprise is to use it to index and cross-reference all of the unstructured data that the company has.
• This requires the enterprise to implement the LLM properly and to fine-tune it on all of its corporate data, which is not what is happening in most cases.
• Most companies give their employees a generic chat prompt and expect them to triple their productivity overnight, leading to the mismatch between expectations and reality.
• Furthermore, most companies are investing in sales and marketing, while the researchers concluded that the highest ROI was to be found in automating back-office tasks.
• The instances where success has been seen are ones where the AI is being used to address a single problem and where the vendor works together with the client to implement its AI service properly.
• The net result is that rolling out ChatGPT or Gemini to all of one’s employees is unlikely to deliver very much in contrast to a specific application, such as automating record-keeping, where the AI has been fine-tuned to execute that particular task.
• Hence, I do not see this as a sign that AI is failing, but more of a sign that it is simply not that intelligent and has to be explicitly taught what to do.
• This is yet another sign (along with the disappointing launch of GPT-5) that super-intelligent machines remain a pipe dream, and what is being delivered is a useful and very valuable tool.
• This will deliver great productivity and economic gains over time, but the human race is not about to be retired by an army of robots taking either its jobs or its lives.

GlobalFoundries – Location, location, location

• GlobalFoundries has announced a deeper relationship with Apple in another sign that the geographic location of its foundries, once seen as a liability, is now a great asset.
• The agreement appears to cover wireless connectivity and power management, and in practice looks like better visibility for GlobalFoundries in terms of volume commitments, meaning that it can invest more aggressively in its facilities in the USA.
• This is clearly part of Apple’s commitment to invest $500bn in the USA, where moving some chip manufacturing from Asia to GlobalFoundries is an easy choice to make.
• For many years, the focus has been to move manufacturing to Asia, but as the strategic rivalry between the USA and China has intensified into an ideological struggle, having all of one’s eggs in China’s backyard has become a problem.
• Here, GlobalFoundries’ position with fabs in the USA, Germany and Singapore plays to its advantage, which it has leveraged with good effect.
• This is why I have long viewed GlobalFoundries as one of the winners from the current geopolitical situation, as although it does not have leading edge, it has capacity built and ready to go.
• This means its clients do not have to wait years for capacity to be built, which I think has been and continues to be a significant advantage.
• Deepening ties with Apple secures it as an anchor client that is likely to be a factor when selling its services to other clients.
• At 20.3x 2025 PER and the promise of better growth in 2026 as the geopolitical advantage makes itself felt, makes this one to keep an eye on.