A Random Thought

On the importance of local infrastructure

Ever since the US announced tariffs on essentially everyone, I've been thinking about what would happen if they suddenly decided to go nuclear on the services industry. Imagine that the US suddenly decides to restrict access to key technologies for countries that aren't sufficiently aligned to their interests1. Drawing some parallels and imagining a more bipolar world, this might mean "countries that do business with China" or "countries that aren't giving us great deals". Not that hard to imagine if we go back 30-40 years, is it?

Now, how might that be an issue? Well, for one thing, this blog runs on infrastructure that the US has vast control over. So do most e-commerce platforms, payments providers, digital banks, company websites, apps. For the past 15-20 years, web infrastructure has migrated to the Cloud almost entirely2, with only a few holdouts.

Market Dominance, Market Control

The web's infrastructure is disproportionately operated by US-founded and US-based companies.3 This isn't without reason: US companies have led the innovation in this space since the beginning.

There are CDNs and hosting platforms (think Cloudflare), developer infrastructure (GitHub, GitLab). In terms of public clouds, there are a few EU companies - all much smaller than AWS, GCP, and Azure -, Alibaba, and then what? Of the hyperscalers4, only Alibaba (again), Tencent and Huawei are from another country - can you guess which? Then there are the datacenters themselves: the world's datacenters are unevenly distributed, with 54% of datacenter capacity in the US, 15% in Europe, 16% in China. You get it.

This gives the US some pretty great leverage. In progressively less tame scenarios, the US might: impose restrictions on more advanced hardware and services1; impose special taxation on services provided to specific countries; entirely block a country from hiring US services via broad sanctions5. It's been done before and it's happening right now. There are limited examples of each, but the threshold for harsher measures seems to be going down.

How companies should prepare

Any of these scenarios spells disaster for a company that depends solely on one provider. For all the talk of "cloud-agnostic" solutions, how many companies would be able to migrate off AWS or GCP within a month or two? Now might be the time to take the risk of "catastrophic failure" more seriously.

Companies should have data backed up in multiple providers and locations, and should ensure that they're able to replicate their services using separate providers. Goes without saying that it might not be enough to shift between Microsoft and Amazon. It's also time to seriously consider on-premise options - with careful evaluation of the time it would take to get the hardware and spin it up.

The estimates and tests won't be done in a day. Consider the (very famous) case of 37signals: DHH made waves by saying they were leaving the cloud all the way back in 2022. Did you notice it had been so long? They're finishing migrating parts of their infrastructure now, in May 2025!

How countries might prepare

That is the more interesting question, isn't it? There are plenty of considerations to make, so I'll spare this a separate post later. What I'm certain of is that there'll be national or regional pushes for infrastructure. It'll take large investments in datacenters, power and tech. It'll also take many, many years.

Not so easy on your own, is it?

Now that I've convinced you to think about it, what is there to do? It'll be a long process.

Advantages of Scale

There are many cons to competing with hyperscalers.

With large market share comes great pricing power and many opportunities to save. US hyperscalers have become much more efficient in the last decade by exploring board, rack, switch and datacenter topologies, innovating in cooling, networking, and much more. Local alternatives are likely to be pricier at first, and will need financial incentives for adoption.

Besides hardware efficiencies, there is a lot more to learn. These companies figured out how to build and sell attractive products on top of their infrastructure. They know how to help their customers spend more. They have entire teams dedicated to moving companies into their products and helping them make the best use of it. This takes time, resources, people, and requires great execution. People won't move to local alternatives that are shit to use. They'll continue using great products, and these hedges will be another variable in their decision-making.

The first 80% is hard, the last 80% is harder

The amount of layers we have to build to create large, sustainable cloud infrastructure is ridiculous. Start thinking about what you'd need to build to have a "fully local" stack.

It'll take great products, great infra, great engineers, great builders. But it can be done, and will be done in the next decade.

References

  1. US Export Restrictions: This has been done many times for technologies with "national security interests". For recent examples, see the AI hardware tierlist and the restrictions to 5G.↩

  2. Virtually everyone has "migrated to the cloud". Data has also been migrated at increasing speeds, further sedimenting cloud infra's hold on companies - migrating data is expensive, and also moves processes along.↩

  3. US hyperscalers dominate global cloud markets, with more than 60% market share for just AWS, Microsoft and Google.The same can be said for CDNs, with companies like Cloudflare, Akamai, apart from the public clouds.↩

  4. Hyperscalers: Companies with global presence in terms of compute, data management, and network infrastructure.Generally refers to the largest cloud providers, but I'd also include Meta, ByteDance, Huawei, etc as companies large enough to have a global-scale presence.↩

  5. Broad sanctions on Russia, Iran, etc. There have also been responses that limit the access of companies to certain territories.↩

#thoughts