Marketing a CPU or any advanced processor not named Nvidia is getting much harder. Once upon a time there were two companies making data center CPUs, and marketing was simply “Ours is better than theirs”. Today, there are a dozen companies designing CPUs, and a few dozen more designing NPUs. How does any company position themselves in this mix, what is their narrative and who do they message against? There are no easy answers here, and the stakes are very high.

All marketing messages should start with a target audience, but here is where the trouble starts. About 60% of the revenue for the category comes from ten customers – the cloud hyperscalers. Almost all of these customers are working on their own custom AI accelerator chips, if not also designing CPUs. Suddenly, the message of “our chip is faster than the competitors'” carries a different connotation. At the same time, the merchant chip vendors absolutely need to convey the superiority of their chip, but they have to be very careful in wording any comparisons. Going back to the audience, at least part of the messaging needs to reach out to the broad organizations of their customers, not just the typical purchasing people in the data center. Every company designing their own chip today has a CFO asking uncomfortable questions about the cost of that development, and internal users who see their peers elsewhere using a wider array of solutions.

Then there is the question of how to position the product. Marketing often works best when a product can be positioned as ‘better’ than some alternative. These narratives are just easier for the human mind to process. But in this situation who do they position against? As we see it, there are really five camps in the CPU fight:

• Intel – they have not been looking great financially, but they remain the market share leader. If they can fix their manufacturing process (and there are no signs that this is slipping), then they can reinvigorate their market share. Lots of doubts around them, but by no means would we count them out.
• AMD – AMD has gained an awful lot of share from Intel in recent years. They continue to deliver on a compelling CPU roadmap.
• Nvidia – Nvidia is strongest in AI, but let’s face it, most of the conversations today are about AI. Their CPU/GPU bundle looks pretty powerful, and cannot be counted out despite having very small CPU share today.
• New entrants – top of this list is Ampere, but there are others out there, especially in the RISC V arena. These companies all make claims about performance and power consumption, however they are up against serious software optimization issues.
• Internal solutions – this is probably the biggest threat out there. The big customers all have their internally designed chips which are steadily eating into the addressable market, with the added benefit of built-in champions and close ties to actual software workloads.

Finally, there is the question of what is the messaging content. Over the past week we have seen a half dozen versions of the same slide showing one company’s solution represented as a server rack or fraction thereof compared to someone else’s multiple server racks. Everyone seem to have 3 times better total cost of ownership than the competitors. This is why choosing the target is so important – whose product do we compare to? How do we tell a massive customer that their internal product underperforms relative to the new product we offer?

Underlying all this conflicting data is the fact that actual results vary tremendously depending on the workload. This chip is better for these workloads, that chip is better for those workloads. The internal solutions all have a massive advantage here, but that advantage comes in the form of a strategy tax and may constrain innovation – AWS seems to be suffering from this right now when it comes to AI and Large Language Models. That being said, few customers want to deal with multiple vendors, the cost of managing all those solutions is highly taxing.

As we said, there are no easy answers here. The large companies probably need to focus on holding off the internal solutions of their customers. The new entrants need to make enough noise to garner toeholds wherever they can – for instance specializing on specific workloads at the expense of winning broader wins, something the large players cannot afford to do. If anything, the big hope here is that this Cambrian explosion of alternatives is likely to dampen the appeal of internal solutions. The rapid advances in AI mean that incumbent solutions all now look vulnerable, locking companies in at a time when the world is moving too fast for anyone to keep up. This is not a huge opening, but it does provide a glimmer of hope.