Two companies are spending billions of dollars to become the place your AI memory lives. One of them will probably win. Neither outcome is good for you.
Google and OpenAI are converging on the same prize from opposite directions. OpenAI is building memory from the conversation outward — you talk to ChatGPT, it remembers, and that memory becomes the spine of everything else they ship. Google is building memory from the data inward — it already has your email, your calendar, your search history, your documents, and now it just needs an assistant smart enough to stitch them together. Both want the same thing: a persistent, personal context layer that makes every other AI tool redundant by comparison.
The consumer angle on this is almost always wrong. It frames the race as a feature comparison — whose memory is smarter, whose is more accurate, whose syncs across more devices. That misses the point. The race is not about features. It is about who holds the running record of what you think about, and on whose terms you get to use it.
OpenAI's bet: capture through conversation
OpenAI's path is the cleanest. You tell ChatGPT things. It remembers them. The more you use it, the more it knows, and the harder it becomes to leave. Every "remember that I'm…" and "remember that my project is…" is a deposit into an account you can't withdraw from.
The strength of this approach is that the data is explicit. When you say "remember this," you are handing over a labeled, dated, high-signal piece of context. That is exactly the kind of data that makes a model better at being personal and, not coincidentally, better at being sold to. OpenAI's bet is that conversation is the highest-bandwidth way to capture intent, and that once they have a year of yours, you won't start over somewhere else.
The weakness is that it is opt-in by friction. You have to keep talking to ChatGPT for the memory to grow. The moment you spend a month in a different tool, the value of the ChatGPT memory decays and the lock-in weakens. OpenAI knows this. That is why memory is being pushed into agents, into background tasks, into everything that keeps you inside the surface area where the capture happens.
Google's bet: capture through infrastructure
Google doesn't need you to tell it anything. It already knows. Gmail knows your work. Calendar knows your schedule. Search knows what you were curious about at 2am. Docs knows what you're drafting. Android knows where you went. The bet is simple: no competitor can match this surface area, so the winner of the memory race is whoever already holds the pipes.
The strength is depth. A memory built from fifteen years of your email is richer than anything you would ever type into a chat box. Google's assistant doesn't need you to remember to tell it things. It just watches.
The weakness is the same as the strength. People are starting to feel the watching. The same instinct that made people uneasy about targeted ads is now attaching to the idea of an assistant that has read every email they've ever sent. Google's pitch is that this time it's fine because the data stays inside Google. The pitch has not aged well in the past, and it is not aging well now.
What both bets have in common
Strip away the strategy decks and both companies are making the same wager. They are betting that you will accept a model where your context lives on their infrastructure, under their terms, optimized for their roadmap, in exchange for an assistant that is more useful than one you could assemble yourself.
They are probably right. Most people will accept that deal, the way most people accepted it with social media and search and email. The product is genuinely better when it has your context. That is not the issue. The issue is what happens to the context once it's captured, and who gets to decide.
The honest version of both roadmaps is: we hold your memory, and we will use it to build the next thing we sell you, and the thing after that, and the policy that governs all of it is one we write and revise at our discretion. That is not a scandal. It is a business model. It is also the reason a memory you own is a different category of product, not a competitor on the same checklist.
The third option nobody is marketing to you
There is a version of AI memory that does not require you to hand your context to a company whose incentives diverge from yours the moment the data gets valuable. It looks like this: you capture what you think and plan, deliberately, in your own words. A tool you control organizes it. Whatever assistant you're using this month — ChatGPT, Claude, a local model, whatever comes next — comes to your memory to get context, instead of you uploading your memory into theirs.
That is what Eigen Mesh is. An open-source, Apache 2.0 personal AI memory. A second brain you own, portable across tools. Not a framework you assemble from parts — a product you install or use hosted. You decide what gets captured. You decide what gets shared. You can run it yourself and the trust question becomes moot.
The labs are not going to build this. Their entire business depends on the memory living with them. The only way you get a memory you own is if someone without a downstream data business builds it, and the only way that someone stays honest is if the code is open and you can leave with your data intact.
Eigen Mesh is open source under Apache 2.0 — github.com/hikaru90/eigen. Self-host with Docker Compose, or get early access to the hosted version. 100 free credits, then €3.99/month — less than a coffee. GDPR, EU servers, no training on your data. Your memory. Not theirs.