Buy vs. rent vs. build decision optimization is an eternal dilemma with only one universal answer: "It depends." Maximize wherever practicable and ethical:
The man has forgotten that those LLM models don't grow on trees in nature, freely available to anyone.
Don't be a sucker - go create your own model, train it for hundreds of millions of dollars. With your own money - you aren't a sucker to borrow from anyone. Fine tune a model, train it again.
Only then, when and if you're get done, will you be able to claim "I'm not sucker, I don't rent anything. And you shouldn't too!".
Until then, no model is yours. Even if you run it locally, you're inevitably engaging in renting it, if not for money directly, then for knowledge and expertise of those large companies. The ones that've created them for you. With an option to utilize them for free, locally.
I generally think there is a lot of good use for local models. Works like Lucy have a ton of potential to help drive tools, do basic reasoning. https://huggingface.co/Menlo/Lucy
But I think this week we are seeing a lot of extremely capable large models. DeepSeek V3 (670B) was joined by Kimi K2 (1T), and now Qwen3 Coder (480B). Trying to run these at home with any reasonable speed is very hard. Smaller models can be quite useful too but I feel like one absolutely wants to be able to tap these large models for a lot of core crucial work.
https://www.reddit.com/r/LocalLLaMA/
> For example let's look at Microsoft pricing. Using a single macOS runner for 1 year costs 0.08 * 60 * 24 * 365 = 42 thousand USD. Meanwhile, for Zig's CI testing we purchased a beefy mac mini for $3000 two years ago.
If your entire usecase is asking for a box to be turned on and nearly nothing else, sure. But I'm enjoying not having to think / care about app updates, migrations, ISP policies about hosting, etc. I'm glad to pay for someone else to do those things. Same like how I pay for someone to care about the email servers I use.
I'm afraid to host my web sites myself simply because I don't trust my ISP. It's rather reliable, but it does go down ~once a quarter, and relying on them would be highly inconvenient. Therefore, I have to choose between Hetzner vs VPS vs cloud (AWS, GCP, etc) vs colocation. After doing all the math and judging my ability to deal with hardware and software upgrades, at the end of the day hosting in the cloud isn't really that much more expensive and it's the least troublesome.
This all makes sense for personal projects, but if the fortune 500 you work for decides its cheaper to migrate to the cloud and give AI assistants to every developer, they decided to light their money on fire, and you should be happy to supply the gasoline and matches.
I've read a lot of blog posts making this same argument for a year "Cloud makes 80% margins! You're fool for paying them!" and I think they show a lack of curiosity as to why so few companies paying 9 figure cloud contracts feels incentivized. That's a huge incentive for a director looking to make their mark right?
In fact we have an example of a company that spent years moving off cloud - Dropbox - and the company has been stagnating since - so this isn't some "easy win".
If you talk to smart business leaders you'll hear the main 3 points come up over and over: cloud offers three huge advantages that are worth incredible margins to many companies:
1. Your business can scale existing or new workloads very quickly
2. It frees up organizational focus for your customer's problem
3. It's easier to hire cloud skills for than bare metal skills
You'll note that improvement on all of these are a trade of money for more time and lower risk. Put in those terms I hope it's not confusing why we continue to pay cloud provides.
You are only repeating the arguments of run-of-the-mill management, and most people reading Kelley's blog post will be familiar with them.
There's something else going on here.
After all, as a business manager with a duty to make the most for your shareholders, if you really anticipated scaling your compute needs, wouldn't you invest in the cheapest way to do more computing ?
If you were a board member of a trucking company, and your CEO told you "Managing trucks is a pain, always filling them with fuel and stuff, I outsourced it to another trucking company, they are only charging us 5 times our previous budget in equipment and drivers salaries. Now we can focus on what we do best, making our customers happy" you'd fire that CEO.
"It's easier to hire for cloud skills that bare metal skills" is self-fulling in part, but if you are doing self-hosting correctly, it's also meaningless because you will just run largely the same software infrastructure. You can go to Dell or HP and have them deliver you a rack or more that already have the cloud software you want installed.
The "it frees up organizational focus for your customer's problem" is the closest to being meaningful. Modern corporate management seems weak generally, other examples include the trend of not promoting internally and instead hiring from outside -- it's like the none of the corporations trust themselves to evaluate people themselves, and inherently feel more confident hiring someone another corporation has found valuable.
"We are bad at managing our core business which has a profit margin of 10%. Therefore, we will take an essential service in support of that, outsource it to a company which is known for algorithmically detecting customer lock-in and raising prices, and is extremely difficult to leave, and will charge us 5 times as much."
