I wonder how long it will take the software industry to re-learn the 2010s lesson, that basing your entire business on (and in this case, firing half of your employees and replacing them with) another company’s API is a bad business decision
Now we just need the RAM market to get back to normal. Or at least fine OpenAI for speculating on raw wafers. There's an article on the front page [0] with this passage that gives me hope that consumer access to VRAM may improve
> On the infrastructure side: OpenAI signed non-binding letters of intent with Samsung and SK Hynix for up to 900,000 DRAM wafers per month, roughly 40% of global output. These were of course non-binding. Micron, reading the demand signal, shut down its 29-year-old Crucial consumer memory brand to redirect all capacity toward AI customers. Then Stargate Texas was cancelled, OpenAI and Oracle couldn’t agree terms, and the demand that had justified Micron’s entire strategic pivot simply vanished. Micron’s stock crashed.
realistically any 'huge' frontier model that takes a rack of H100s to infer against is probably going to have downtime no matter who runs it.
downtime is always going to 'scale' poorly against loads that require a lot of hardware thrown at them, even with lots of good fail-over -- probably worse for the small vendors because they don't have the contracts supplying them with hardware first so availability is already at a premium for them.
so, I guess i'm saying yeah I hope frontier-level-models get out soon in the open arenas, but I suspect the same or similar level of exclusivity will exist as long as they take that much compute to operate.
What makes you say it is a bad business decision? It seems to be a fine decision to make for things like AWS, since when it goes down, a ton of websites go down and no one blames the site.
There is no way to know whether it is a good or bad business decision just because they can go down when a third party goes down. For example, if you save $50 million a year by firing half your employees and replacing them with AI, but you lose $10 million a year because your site goes down when Claude goes down, then you made a great business decision.
Oddly, I do not think you are wrong. In a pure money calculus exercise, this seems like a no brainer. Naturally, the math gets iffy the moment we are trying to capture something less tangible like 'customer may get sufficiently annoyed to drop us altogether' or 'we are no longer a respected company' or what MBAs would call 'unexpected goodwill extraction'.
I honestly don't care nearly as much as I used to, because I used to be more upset over this. Now, I simply wait to see how much is enough to rile up average Joe and Jenna.
On the other hand a competitor site that is up (or bricks and mortar competitors) might get a lot of business when AWS goes down. If you depend on AWS for operations it might be a lot more expensive than that.
Mostly I think its that management does not blame the person who picks AWS. Its another iteration of "no one got fired for buying IBM/Microsoft".
It is also an issue at other levels: if all a county's businesses rely on AWS (let alone its government) then that gives the US huge leverage over you (sanctions would shut down your economy).
Corporate leaders don't learn lessons. They follow trends, chase growth, reduce the perception of risk, diffuse blame, get their business acquired, and exit with money bags in both hands. No learning from experience necessary.
I am not sure how many people learned it the first time. To be fair, it's really hard to build a business without major dependencies. The key is to assume they will fail and have alternatives available.
we have a big dependency on AI, both for developers (can survive without it, mostly habits) and internal workflows (very hard to go without it). So we decided to unplug from cloud AI, rent our own GPU and use an open model for both scenarios. We have been very happy with it so far, 60% cheaper and around 50% faster
Is it important to be that self reliant? I wasn't in the workforce then, so I assume if you have an outsourced system with 99% reliability it would be an acceptable risk. Not sure if any AI system will reach that level, but for potential gains it could be worth it.
There's a decent chance on any given weekday that Claude will go down when US Pacific comes online while London is still working. It happens all the time. Kind of embarassing for a $300bn company.
I got 2 500s and quickly popped over to codex since I was just doing documentation and planning work and didn't neeeeeeed claude. He's already back for me. :)
To another's point, yeah, it's sad that Claude taking a brief hiatus just halts workflow. I imagine my whole team just started packing for a beach trip. Then again, I suppose if the CPA office computers go down, everyone doesn't dust off giant ledgers and quill pens and start working manually.
Don't worry this is not an outage. Claude is on another lunch break and just like last week [0] after taking a day off on Monday, forgot to tell the whole world first.
He just needs to finish his lunch in 30 mins before going back to work.
"aLl sYsTeMs oPeRaTiOnAl".... my brother in christ, the newest color bar for "claude.ai" on https://status.claude.com/ is red and every color bar is a mosaic, yet this is somehow considered "operational"
By now they would have done a sweep with Mythos to remove any existing bugs and protect it from any cybersecurity attacks. So what could realistically bring it down at this point? Internal espionage?
126 comments
> On the infrastructure side: OpenAI signed non-binding letters of intent with Samsung and SK Hynix for up to 900,000 DRAM wafers per month, roughly 40% of global output. These were of course non-binding. Micron, reading the demand signal, shut down its 29-year-old Crucial consumer memory brand to redirect all capacity toward AI customers. Then Stargate Texas was cancelled, OpenAI and Oracle couldn’t agree terms, and the demand that had justified Micron’s entire strategic pivot simply vanished. Micron’s stock crashed.
[0] https://adlrocha.substack.com/p/adlrocha-how-the-ai-loser-ma...
downtime is always going to 'scale' poorly against loads that require a lot of hardware thrown at them, even with lots of good fail-over -- probably worse for the small vendors because they don't have the contracts supplying them with hardware first so availability is already at a premium for them.
so, I guess i'm saying yeah I hope frontier-level-models get out soon in the open arenas, but I suspect the same or similar level of exclusivity will exist as long as they take that much compute to operate.
Which doesn't mean the open models will definitely succeed, it just means they have more of a shot than the open social networks ever did
>AI models are not really effected by that at all.
I don't know about that. More usage means more support, which means more docs and open source projects, wrappers, harnesses built around them etc.
Way less demand to build tooling around open weight models if they remain hobbyist.
1. Using AI helps as part of the training process.
2. All the prompts going to openai/claude is a gold mine.
There is no way to know whether it is a good or bad business decision just because they can go down when a third party goes down. For example, if you save $50 million a year by firing half your employees and replacing them with AI, but you lose $10 million a year because your site goes down when Claude goes down, then you made a great business decision.
I honestly don't care nearly as much as I used to, because I used to be more upset over this. Now, I simply wait to see how much is enough to rile up average Joe and Jenna.
Mostly I think its that management does not blame the person who picks AWS. Its another iteration of "no one got fired for buying IBM/Microsoft".
It is also an issue at other levels: if all a county's businesses rely on AWS (let alone its government) then that gives the US huge leverage over you (sanctions would shut down your economy).
It is a weird thing to respect come to think of it... Having an accurate status page should be the baseline, but that's the world we live in.
To another's point, yeah, it's sad that Claude taking a brief hiatus just halts workflow. I imagine my whole team just started packing for a beach trip. Then again, I suppose if the CPA office computers go down, everyone doesn't dust off giant ledgers and quill pens and start working manually.
He just needs to finish his lunch in 30 mins before going back to work.
[0] https://news.ycombinator.com/item?id=47662680
I took the time to stand up and do some stretching. Good use of time.
Anthropic: 9.
World: 9 nines? That's awesome!
Anthropic: No, a singular 9. 90-something percent uptime.