Forgot your password?
Close
typodupeerror
AI

Anthropic Raises Claude Code Usage Limits, Credits New Deal With SpaceX (arstechnica.com) 43

Posted by BeauHD from the supply-and-demand dept.
An anonymous reader quotes a report from Ars Technica: At its Code with Claude developer conference on Wednesday, Anthropic announced a deal with SpaceX to utilize the entire compute capacity of the latter's data center in Memphis, Tennessee. On stage at the conference, CEO Dario Amodei said the deal was intended to increase usage limits for Anthropic's Pro and Max plan subscribers. The announcement was accompanied by an increase in those usage limits; Anthropic doubled Claude Code's five-hour window limits for Pro and Max subscribers, removed the peak-hours limit reduction on Claude Code for those same accounts, and raised API limits for its Opus model. The table [here] outlining the Opus changes was shared in the company's blog post on the topic.

Anthropic claims the deal gives the company access to more than 300 megawatts of new compute capacity. For its part, SpaceX focused its announcement on the capability of the Colossus 1 supercomputer that's at the center of the deal. "Colossus 1 features over 220,000 NVIDIA GPUs, including dense deployments of H100, H200, and next-generation GB200 accelerators," SpaceX wrote. Additionally, Anthropic "expressed interest" in working with SpaceX to build up "multiple gigawatts" of orbital compute capacity, tying into a recent (but unproven) focus on exploring orbital data centers as an answer to the problem that "compute required to train and operate the next generation of these systems is outpacing what terrestrial power, land, and cooling can deliver on the timelines that matter." "I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed," Elon Musk said on Wednesday. "No one set off my evil detector."

Anthropic Raises Claude Code Usage Limits, Credits New Deal With SpaceX More | Reply

Anthropic Raises Claude Code Usage Limits, Credits New Deal With SpaceX

Comments Filter:

  • RONTHFLMFASO x 10^42 power

    • The radioactive Geiger Counter paradox. "Everything is radioactive or nothing?"

    • Re: (Score:3, Informative)

      by Turkinolith ( 7180598 )
      He should have pointed the "evil detector" back at himself.

      • Re: (Score:1)

        by Twinbee ( 767046 )
        No. Neuralink is helping paralyzed people already. Tesla cars have saved so many lives because of their safety. He also cared about the evil EFF party in South Africa when almost nobody else cared.

  • meh (Score:2)

    by nomadic ( 141991 )

    Huh, haven't seen a change in my Pro subscription. Though I hit the weekly model limit -- if that's not changing then it's not really a big deal.

    • You are correct in noticing that they did not increase the weekly limit. Therefore you can do the same, faster, per week but not more, faster, per week.

  • FlashAttention (Score:3)

    by bill_mcgonigle ( 4333 ) * on Thursday May 07, 2026 @12:41PM (#66132360) Homepage Journal

    I did some math the other day on running local AI models and the net result is most homes can't afford to run the current median models.

    They don't just need 80GB of VRAM, they need newer architectures - to be supported by CUDA, to be supported by pytorch, etc.

    These problems may well be solvable with more clever use of hardware, MoE, acceptable quantization, etc., but today you're in for several grand and something north of 100W idle to use what is effectively a $20/mo plan.

    A small enterprise can afford local, so that's good. We paid more than that for one SGI machine back in the day.

    The point of the exercise was to plot the position on the curve. We're at something like 2006 YouTube where nobody could afford the drives or bandwidth that YouTube/Google was giving away for free (aka with VC money). Eventually hard drives got cheaper, people got gigabit at home, FlashServer was replaced with h.264/HTML5, phones could stabilize video locally, etc.

    So it looks like these AI companies need to stay alive for about seven more years giving away product at a loss, or at least highly oversubscribed, to turn a profit. Hence the low token allowance, the banning of OpenClaw, etc.

    On the other hand, I read the blog of a security researcher yesterday who found an exploit with (IIRC) Claude, tried to refine the PoC, but got dinged on "out of tokens" before he could finalize it. So he just deleted the work and moved on.

    It sounds like they're trying to not lose money at such a velocity and are trying to find a sweet spot where people don't just declare it too underpowered to use.

    A global energy depression may well take out the supermajority of the companies that believe they can burn investment money for seven more years. There is circular financing money, then there is real return on capital money. One is to fool the markets, the other is grounded in current physics.

    • Re: (Score:2)

      by 0123456 ( 636235 )

      > So it looks like these AI companies need to stay alive for about seven more years giving away product at a loss, or at least highly oversubscribed, to turn a profit. Hence the low token allowance, the banning of OpenClaw, etc.

      You're assuming that tech will continue to improve as fast as it did over the last twenty years... when we're past the end of Moore's Law.

      At a minimum that would likely involve moving away from silicon to something we can run at much higher clock speeds, or finding a much more eff

    • Re: (Score:3)

      by allo ( 1728082 )

      Nah. If you don't absolutely challenge Opus you can get away with 180 Watt peak and 16 GB VRAM graphics cards like 5060 Ti (probably the best value starter card for AI at the moment). Using MoE models helps and getting up-to-date with new releases is important as nobody will deploy the latest and greatest model if you don't do it yourself. You won't beat Opus, but you're more or less on par with Sonnet, in particular when your client fits your use case well.

      • Don't many of these models process tokens at like 5 / sec on something like a 5060ti?

        • GPT-OSS-20B gives me 34 tokens/sec on a 4 yr old gaming laptop with a 3080 RTX - smaller models come in faster still.

        • Re: (Score:2)

          by CAIMLAS ( 41445 )

          I'm getting about 35 tokens/s on a 3060ti 12GB with Qwen 3.6 MoE Q4, KV 8/8, and would get about 50-60 with KV 4/4. That's with 128K context.

    • This gives people more time to become addicted to the dopamine rush from using AI and to bond with their chatbots. Anthropic is like the drug dealer who gives out free samples to get more customers.

      AI is the performance enhancing drug for tech workers. Someday we'll realize that we need to limit AI time just as we try to limit screen time.

  • I hate that it uses jet engines to supplement power of the data centers. Makes me think they should scale back the token usage on them.

    • Jet Engines are as efficient as the steam turbine generators used at power plants.

      • Jet Engines are as efficient as the steam turbine generators used at power plants.

        Sure, if the goal is to produce thrust. If it's to produce electricity, then no, because you have to spend thrust to achieve compression. Steam plants produce compression with a pressure vessel, not with thrust.

  • Now instead of hitting your token limit in 10 minutes, hit them in 20 minutes!
  • How.
    Are.
    You.
    Going.
    To.
    Cool.
    It.
    In.
    Space.
    Though?

     I feel like someone needs to keep tweeting this at Musk until he gets it. There's no air in space. You need infrared radiators. Those weigh A LOT and don't work very well.
  • Your own AI system is number 1 at neonazi propaganda and revenge porn. When you put on an act that you care about societal ills, absolutely *nobody* buys it. It's not a good look. You're not convincing anyone.

    Buddy, I respect your accomplishments. You deserve your spot as the world's richest man. And, overall, I understand that you're trying to sincerely push humanity forward, since that goal lines up well with your empire-building. But, don't try and act like a caring leader. It's not a good look on yo

  • I'm still rocking a 2009 iMac. I don't need much, check my email, pay my bills, slice some .STLs, play some older games; thanks to Steam sales I have a lot of older games. When it's not being used I let BONIC [berkeley.edu] run in the background. For those who don't know, BONIC is a distributed compute platform which helps Colleges and Universities across the globe leverage unused compute cycles from computers people have volunteered to let the researchers use when their machines are idle.

    Over the years my old iMac has

  • Anthropics remains a toxic, opaque company that treats its customers with disdain. They change usage limits and terms of use regularly without informing users, banning users for life without explaining the cause.

    The competitors might not be much better, but at least they seem to be more transparent. We better hope that Anthropic does not win the AI race without a change of policies.

    • Re: (Score:2)

      by CAIMLAS ( 41445 )

      Precisely why I jumped to local inference, and spent about $20/mo checking the work and planning on cloud models. You don't need much hardware to effectively replace claude.

  • Shouldn't this datacenter be utilized by xAI (which is part of SpaceX now)? Why would they be leasing the capacity to a competitor instead of using it for their own internal AI corp?

    Also is this datacenter built with the NVidia processors that were originally purchased by Tesla (for their AI project: Self Driving), before Musk decided to redirect them to his competing AI company xAI that he started because Tesla wouldn't guarantee him the controlling ownership stake he demanded "or else he would build AI e

    • Nobody uses xAI except maybe X bots. I doubt X employees use X to do infra/code tasks. Companies are throwing $$ at anthropic for Claude Code (mainly) and I think Anthropic will lease compute power from anyone as long as it keeps their frontier models humming along.

  • I hate this new so-called unit of measure. Make it less efficient, and you get a higher number!

  • his evil detector is already off the scales from his own evil self

Slashdot Top Deals

The best things in life go on sale sooner or later.

Close