Predicting what the CMA will do on the IBM/Hashicorp Merger - an example using my cmabot

Tony Curzon Price 14/2/25

Back in September, I outlined a few routines that I had built on top of the OpenAI API to help me make a prediction about the Synopsys/Ansys merger. I promised to eventually put the tool on the web … Well … last week, I was again asked to forecast what the CMA would do, this time in the case of the IBM/Hashicorp merger. I get asked for these forecasts by hedge funds betting on regulatory decisions. They know that if they put the work in to make a slightly better forecast than the average punter, they can design trading strategies that will earn slightly higher returns. It is a great example of the process by which prices in markets come to reflect available information.

What better excuse than this commission to put a bit more effort into developing my CMABOT, which I am pleased to say is now live! Please feel free to play around with it and tell me when it works and when it doesn’t. This short note will outline how I have used the cmabot to come up with my forecast in this case. And yes, I will let you know what my forecast is at the end… (The tool does not generate a forecast itself … yet).

When I started out on this analysis, I had no idea what Hashicorp does for a living, and I had a vague sense of IBM, living off past patents and doing slightly dull sounding, but terribly important, system integration for huge companies, governments etc. In other words, I was probably pretty much where a Grade 6 or 7 at the CMA doing the legwork to come up with a recommendation on whether or not to take this merger to Phase 2. My task for the hedge fund is to work out with what subjective probability the deal will be referred on to Phase 2, and, if that happens, with what probability it is blocked.

The first thing I do is to get the cmabot to find me a list of IBM’s major products, which I do like this:

Then I do the same for Hashicorp (Company 2):

More about these products later… But first, let’s check whether there are any straightforward competitors between these products - I do that with “Load/Generate Horizontal Theory of Harm” on the right hand side of the screen, which gives me:

This table was generated with the prompt at the bottom of the table applied to all the products that the cmabot had returned for IBM and for Hashicorp. They are all pretty low values.

What this says to me is that if you’re IBM’s customer for one of its products, you are unlikely to think of these Hashicorp products as substitutes. This tells me that we are unlikely to be in the situation of a standard horizontal merger that threatens to take away competition by putting a substitute under the same roof. For example, if I put Oracle in instead of Hashicorp, I get this table, which has a good number of values over 0.5, which would make me think that there’d be some classic horizontal theories of harm to investigate in that merger:

I still don’t really know what Hashicorp does, but let’s keep going in this slightly blind, structural way for now. If the CMA is going to find against this merger, it is unlikely to be for horizontal theories of harm. To check vertical theories of harm - they are all some variation on “IBM uses its ownership of Hashicorp to make life harder for one of its competitors, thus taking greater profits” (or vice versa). In order to check how likely that is, I first need to see which are the relevant competitors to IBM here. For this, I run a relatively open-ended prompt, asking what are the competitors to IBM in various product lines, and how dependent are those competitors on any Hashicorp product. That gives me this table:

This is interesting. The hyperscalars seem to have some degree of Hashicorp dependency, but so do others, from VMWare in corporate data centers to other big system integrators like HPE, SAP and Salesforce. There is also the slightly odd matter of Red Hat, the IBM-acquired Linux distribution being on the list.

This is getting more interesting - Hashicorp, even if not absolutely necessary to many companies, clearly has a pretty wide customer base. (Remember, I still don’t know what the company does), and IBM, given its own breadth, competes with many of those customers. So this is starting to get more interesting from a vertical perspective.

To delve into the vertical issues, I need to give the cmabot the name of some IBM competitors, and it will check in more detail the degree of dependency that this competitor has to any of Hashicorp’s products.

Let’s start with AWS and see what gives:


The question being asked in the table above is: given that I am an AWS customer using one of these AWS products, how likely is what I am doing also going to require me using a Hashicorp product? In other words, are there any complementarities between Hashicorp and AWS products? What I find is that Terraform and to some extent Vault have a high degree of complementarity with quite a few AWS products. So this is the first basic clue that IBM might be acquiring some new market power over competitors like AWS in acquiring Hashicorp. If IBM really wanted to, they could make life harder for AWS customers once they controlled Terraform or Vault.

Let’s do the same with different types of competitor. First, VMWare:

Again, Terraform and to a lesser extent Vault are complementary to VMWare. So whatever good stuff it is Hashicorp is doing, it is not just for the hyperscalars. Let’s now try Accenture, as another big system integrator:

Again, Terraform and Vault seem to be pretty important to a general system integrator. Let’s do one more of these, checking Oracle:

Terrafrom looks important, but less so the other products.

Conclusion so far? - Terraform, and to some extent Vault, will give IBM some nice-to-have market power over quite a broad range of its competitors. I can see why this merger went to Phase 1: these are economically important and pretty complex markets. The CMA needed to understand why Terraform (and Vault) are so important to so many different types of company, and what interest IBM might have in flexing its muscle over any of those competitors. It’ll obviously needed more than a brief look around to decide whether this is OK or not. That is what Phase 1 gives you - the time to do a bit more analysis before deciding whether to go in-depth.

In order to make my prediction on whether the Phase 1 referral will conclude with a longer, in-depth, Phase 2 referral, I now need to find out more about what role Hashicorp actually plays in the cloud/integrator/entreprise software ecosystem.

Terraform offers a Hardware Configuration Language that allows cloud customers to become more cloud-provider agnostic. The language itself is open source, and there is a substantial community that has developed around it. All the major hyperscalars support it, as does VMWare. Terraform revenues come from Hashicorp value-added tools and services over and above the language. You can see why it is a must-have tool for so many different sorts of businesses. The cloud vendors need to support it; the system integrators will use it to remain cloud agnostic; the datacentre crowd use it for hybrid setups… The Vault product does “secrets management” - keeping keys for encryption to work between corporate intranets and public networks. Also a central, cross-cutting function where you would want to be cloud-independent. But the product seems less successful in that space.

This makes Terraform pretty important in maintaining what competition there is between hyperscalars- The CMA itself has just reported on competition in cloud, and it is concerned with what it found. Cloud services are, unsurprisingly, quite sticky. Multi-cloud makes this somewhat better, but it is hard:
Some customers can and do successfully multi-cloud but we have found that
technical barriers to multi-cloud negatively affect many customers' ability to use
and integrate multiple public clouds. This limits customers' ability and/or incentive to exercise choice of cloud provider.
This means that Terraform is part of the solution to the problems that the CMA has found in the cloud space. They will be keen to make sure that this solution is not blunted by the acquirer’s incentives.

So … what changes if Terraform goes from being the main product at Hashicorp to being one of many products at IBM?

Perhaps IBM would try to flex its muscle somehow - say by making some features in the value-added suite available only if you were also an IBM integration customer. What would Accenture or HPE do then? Clearly, we are interested in Hashicorp/Terraform substitutes: how much of a must-have is it, and how much of moat does Terraform really have?

These are questions that cmabot has a tool to help with: I put Hashicorp Terraform into Company 2, and I run the dependency on IBM amongst competitors to Hashicorp Terraform to generate this (the dependency column is of less interest to me here than the competitors and their product lines):


The hyperscalars are there - yes, they offer a substitute, but not one that enables multicloud… so they aren’t a relevantly good substitute. And there is Red Hat again in the list - there is going to be quite some untangling for the CMA to do there to understand if these are really substitute products. If they are, I think that there might be a request - or perhaps an offer from IBM - to divest the Ansible Automation Platform, either to a competitor or placing it in an open source foundation with funding promise from IBM. The last three, and perhaps Pulumi in particular, are contenders in this space,and their products may be attractive and technically better than Terraform … but my impression is they do not have the community or user base… So I would say they are, for now, only a moderate constraint on any exercise of market power by IBM through Terraform ownership.

(It has been pointed out to me that Open Tofu is an interesting contender, at Terraform fork being run inside the Linux Foundation. This was not generated by cmabot, so I obviously need to tweak the prompts a bit. It could be that the blindness came because it is not a business competitor to Hashicorp…)

But all this is very rough guess at this stage. Here is what I think will happen:

  1. There is enough complexity here for a phase 2 referral (80% chance on these grounds only)
  2. At phase 2, the CMA will disentangle the questions sketched out above, and IBM will provide assurance and some structural remedies (like the Ansible divestment) … because the Terraform acquisition makes sense in terms of IBM’s positioning as a “good custodian” of tools that have wide market benefit - they have shown this with Red Hat.
  3. The CMA will ask for guarantees, will get them, and will then allow the merger (90% chance).

This does not speak to all the politics around the CMA’s supposed “anti-growth” disposition, the recent replacement of the chair by someone from the cloud business (Doug Gurr, ex-Amazon), and the very recent “strategic steer” from Secretary of State Reynolds by which:

In all areas where the CMA has discretion over which reviews, studies or investigations to commence, and in all cases where the CMA is considering remedies, the CMA should give appropriate consideration to:

These recent moves will without a doubt complicate the CMA’s decision. I think that without these noises from Whitehall, IBM/Hashicorp would have been a very natural Phase 2 referral. But actually, I think the argument for referring may still win out - these more political considerations make me lower it from 80% chance to 60% - in other words, I would be mildly surprised (and disappointed) if the CMA does not take this one to Phase 2.