<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Spacex on ICE-ICE-BEAR-BLOG</title><link>https://ice-ice-bear.github.io/ko/tags/spacex/</link><description>Recent content in Spacex on ICE-ICE-BEAR-BLOG</description><generator>Hugo -- gohugo.io</generator><language>ko</language><lastBuildDate>Wed, 06 May 2026 00:00:00 +0900</lastBuildDate><atom:link href="https://ice-ice-bear.github.io/ko/tags/spacex/index.xml" rel="self" type="application/rss+xml"/><item><title>Anthropic, SpaceX Colossus 1 통째로 임대 — Claude 사용 한도 인상의 진짜 의미</title><link>https://ice-ice-bear.github.io/ko/posts/2026-05-06-anthropic-higher-limits-spacex/</link><pubDate>Wed, 06 May 2026 00:00:00 +0900</pubDate><guid>https://ice-ice-bear.github.io/ko/posts/2026-05-06-anthropic-higher-limits-spacex/</guid><description>&lt;img src="https://ice-ice-bear.github.io/" alt="Featured image of post Anthropic, SpaceX Colossus 1 통째로 임대 — Claude 사용 한도 인상의 진짜 의미" /&gt;&lt;h2 id="개요"&gt;개요
&lt;/h2&gt;&lt;p&gt;2026년 5월 6일, &lt;a class="link" href="https://www.anthropic.com/" target="_blank" rel="noopener"
 &gt;Anthropic&lt;/a&gt;이 두 가지 발표를 한 묶음으로 내놨다. (1) &lt;a class="link" href="https://claude.com/product/claude-code" target="_blank" rel="noopener"
 &gt;Claude Code&lt;/a&gt;와 &lt;a class="link" href="https://claude.com/platform/api" target="_blank" rel="noopener"
 &gt;Claude API&lt;/a&gt;의 사용 한도 인상, (2) &lt;a class="link" href="https://www.spacex.com/" target="_blank" rel="noopener"
 &gt;SpaceX&lt;/a&gt;와의 컴퓨트 파트너십. 두 번째가 첫 번째의 원인이다. 헤드라인은 &amp;ldquo;한도 인상&amp;quot;이지만 실제 뉴스는 &lt;strong&gt;Anthropic이 경쟁사 &lt;a class="link" href="https://x.ai/" target="_blank" rel="noopener"
 &gt;xAI&lt;/a&gt;가 운영하던 Colossus 1 슈퍼컴퓨터 전량을 통째로 임대했다&lt;/strong&gt;는 사실이다.&lt;/p&gt;
&lt;pre class="mermaid" style="visibility:hidden"&gt;flowchart LR
 SpaceXAI["SpaceXAI &amp;lt;br/&amp;gt; Colossus 1 (Memphis)"] --&gt; Compute["220K+ NVIDIA GPU &amp;lt;br/&amp;gt; 300MW+ 전력"]
 Compute --&gt; Anthropic["Anthropic 추론 capacity"]
 Anthropic --&gt; ClaudeCode["Claude Code &amp;lt;br/&amp;gt; 5h limit 2배"]
 Anthropic --&gt; API["Claude API &amp;lt;br/&amp;gt; Opus RPM/TPM 상향"]
 Anthropic --&gt; Sub["Pro / Max 가입자 &amp;lt;br/&amp;gt; 체감 capacity 증가"]&lt;/pre&gt;&lt;h2 id="발표-내용--한도-변경-3건"&gt;발표 내용 — 한도 변경 3건
&lt;/h2&gt;&lt;p&gt;&lt;a class="link" href="https://www.anthropic.com/news/higher-limits-spacex" target="_blank" rel="noopener"
 &gt;공식 글&lt;/a&gt;이 명시한 &lt;strong&gt;즉시 효력&lt;/strong&gt; 변경:&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;항목&lt;/th&gt;
 &lt;th&gt;변경&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Claude Code 5시간 rate limit&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;2배 인상&lt;/strong&gt; — &lt;a class="link" href="https://claude.com/pricing" target="_blank" rel="noopener"
 &gt;Pro&lt;/a&gt; · &lt;a class="link" href="https://claude.com/pricing/max" target="_blank" rel="noopener"
 &gt;Max&lt;/a&gt; · &lt;a class="link" href="https://claude.com/pricing/team" target="_blank" rel="noopener"
 &gt;Team&lt;/a&gt; · seat 기반 &lt;a class="link" href="https://claude.com/pricing/enterprise" target="_blank" rel="noopener"
 &gt;Enterprise&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Claude Code 피크 시간대 감산&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;제거&lt;/strong&gt; — Pro / Max 계정&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Claude API rate limit&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;Opus 계열 대폭 상향&lt;/strong&gt; — 자세한 변경은 &lt;a class="link" href="https://platform.claude.com/docs/en/api/rate-limits" target="_blank" rel="noopener"
 &gt;API rate limits 문서&lt;/a&gt; 참조&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;&lt;a class="link" href="https://www.anthropic.com/claude/opus" target="_blank" rel="noopener"
 &gt;Opus 모델&lt;/a&gt;에 적용된다는 점이 중요하다. Sonnet / Haiku는 명시 대상이 아니다. Opus가 가장 비싼 모델이자 frontier reasoning 워크로드에 쓰이는 라인이라, &lt;strong&gt;새로 들어온 GPU 용량이 가장 비싼 추론을 가장 먼저 풀어주는 데 쓰인다&lt;/strong&gt;는 뜻이다.&lt;/p&gt;
&lt;h2 id="새-컴퓨트--colossus-1-전량-임대"&gt;새 컴퓨트 — Colossus 1 전량 임대
&lt;/h2&gt;&lt;p&gt;핵심 수치:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;300MW+&lt;/strong&gt; 신규 capacity&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;220,000+ NVIDIA GPU&lt;/strong&gt; — H100 / H200 / 차세대 GB200 혼합&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;한 달 내 가용&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;위치: &lt;a class="link" href="https://capacityglobal.com/news/anthropic-secures-full-capacity-of-spacex-data-centre/" target="_blank" rel="noopener"
 &gt;멤피스 Boxtown 지구의 옛 Electrolux 공장 부지&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;이 클러스터는 원래 &lt;a class="link" href="https://en.wikipedia.org/wiki/Colossus_%28supercomputer%29" target="_blank" rel="noopener"
 &gt;xAI가 자사 Grok 모델을 위해 record-time으로 세운 것&lt;/a&gt;이다. 같은 날 발표된 &lt;a class="link" href="https://x.ai/news/anthropic-compute-partnership" target="_blank" rel="noopener"
 &gt;SpaceXAI 측 글&lt;/a&gt;이 이를 확인한다:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;&amp;ldquo;SpaceXAI has signed an agreement with Anthropic to provide access to Colossus 1&amp;hellip; Anthropic plans to use this additional compute to directly improve capacity for Claude Pro and Claude Max subscribers.&amp;rdquo;&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;&lt;a class="link" href="https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/" target="_blank" rel="noopener"
 &gt;xAI는 Colossus 2 구축에 집중&lt;/a&gt;하면서 1세대 클러스터를 직접 경쟁사 Anthropic에 통째로 넘긴 셈이다. Elon Musk의 &lt;a class="link" href="https://www.tomshardware.com/tech-industry/artificial-intelligence/musks-spacex-has-rented-out-access-to-its-supercomputers-220-000-nvidia-gpus-and-300-megawatts-of-ai-compute-power-to-rival-anthropic-musk-says-no-one-set-off-my-evil-detector-antrhropic-also-interested-in-orbital-data-centers" target="_blank" rel="noopener"
 &gt;코멘트&lt;/a&gt;: &amp;ldquo;No one set off my evil detector.&amp;rdquo;&lt;/p&gt;
&lt;h2 id="anthropic-전체-컴퓨트-포트폴리오"&gt;Anthropic 전체 컴퓨트 포트폴리오
&lt;/h2&gt;&lt;p&gt;이번 SpaceX 건은 Anthropic이 6개월간 쌓아온 megadeal 시리즈의 가장 최근 조각이다.&lt;/p&gt;
&lt;table&gt;
 &lt;thead&gt;
 &lt;tr&gt;
 &lt;th&gt;파트너&lt;/th&gt;
 &lt;th&gt;규모&lt;/th&gt;
 &lt;th&gt;시기&lt;/th&gt;
 &lt;th&gt;출처&lt;/th&gt;
 &lt;/tr&gt;
 &lt;/thead&gt;
 &lt;tbody&gt;
 &lt;tr&gt;
 &lt;td&gt;Amazon (&lt;a class="link" href="https://aws.amazon.com/ai/machine-learning/trainium/" target="_blank" rel="noopener"
 &gt;Trainium&lt;/a&gt;)&lt;/td&gt;
 &lt;td&gt;최대 &lt;strong&gt;5GW&lt;/strong&gt;, 2026년 말까지 ~1GW 신규&lt;/td&gt;
 &lt;td&gt;진행 중&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.anthropic.com/news/anthropic-amazon-compute" target="_blank" rel="noopener"
 &gt;공식&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Google (&lt;a class="link" href="https://cloud.google.com/tpu" target="_blank" rel="noopener"
 &gt;TPU&lt;/a&gt;) + &lt;a class="link" href="https://www.broadcom.com/" target="_blank" rel="noopener"
 &gt;Broadcom&lt;/a&gt;&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;5GW&lt;/strong&gt;, 2027년 가동 시작&lt;/td&gt;
 &lt;td&gt;미래&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.anthropic.com/news/google-broadcom-partnership-compute" target="_blank" rel="noopener"
 &gt;공식&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;Microsoft + &lt;a class="link" href="https://www.nvidia.com/" target="_blank" rel="noopener"
 &gt;NVIDIA&lt;/a&gt;&lt;/td&gt;
 &lt;td&gt;Azure에 &lt;strong&gt;$30B&lt;/strong&gt; 규모&lt;/td&gt;
 &lt;td&gt;전략적&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.anthropic.com/news/microsoft-nvidia-anthropic-announce-strategic-partnerships" target="_blank" rel="noopener"
 &gt;공식&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.fluidstack.io/" target="_blank" rel="noopener"
 &gt;Fluidstack&lt;/a&gt; (미국 인프라)&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;$500억&lt;/strong&gt; 자체 투자&lt;/td&gt;
 &lt;td&gt;다년&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.anthropic.com/news/anthropic-invests-50-billion-in-american-ai-infrastructure" target="_blank" rel="noopener"
 &gt;공식&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;tr&gt;
 &lt;td&gt;SpaceX / xAI&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;300MW+&lt;/strong&gt;, 220K GPU&lt;/td&gt;
 &lt;td&gt;&lt;strong&gt;즉시 (~1개월 내)&lt;/strong&gt;&lt;/td&gt;
 &lt;td&gt;&lt;a class="link" href="https://www.anthropic.com/news/higher-limits-spacex" target="_blank" rel="noopener"
 &gt;공식&lt;/a&gt;&lt;/td&gt;
 &lt;/tr&gt;
 &lt;/tbody&gt;
&lt;/table&gt;
&lt;pre class="mermaid" style="visibility:hidden"&gt;graph TD
 Anthropic["Anthropic"] --&gt; AWS["AWS Trainium &amp;lt;br/&amp;gt; 5GW"]
 Anthropic --&gt; GCP["Google TPU &amp;lt;br/&amp;gt; 5GW (2027~)"]
 Anthropic --&gt; Azure["Azure NVIDIA &amp;lt;br/&amp;gt; $30B"]
 Anthropic --&gt; Fluid["Fluidstack &amp;lt;br/&amp;gt; $50B (US)"]
 Anthropic --&gt; SpaceX["SpaceX Colossus 1 &amp;lt;br/&amp;gt; 300MW+ 즉시"]&lt;/pre&gt;&lt;p&gt;세 가지 가속기 — &lt;a class="link" href="https://aws.amazon.com/ai/machine-learning/trainium/" target="_blank" rel="noopener"
 &gt;AWS Trainium&lt;/a&gt;, &lt;a class="link" href="https://cloud.google.com/tpu" target="_blank" rel="noopener"
 &gt;Google TPU&lt;/a&gt;, NVIDIA GPU — 위에서 Claude를 학습·운영한다고 공식 글이 명시한다. &lt;strong&gt;단일 칩 종속이 가장 큰 리스크라는 인식&lt;/strong&gt;이 깔려 있고, SpaceX 건은 그 중 NVIDIA 라인을 즉시 보강한다.&lt;/p&gt;
&lt;h2 id="rate-limit-구조-복습--어디로-들어오나"&gt;Rate limit 구조 복습 — 어디로 들어오나
&lt;/h2&gt;&lt;p&gt;Anthropic API의 한도 체계를 짧게 짚어두면 이번 인상이 어디에 꽂히는지가 명확해진다.&lt;/p&gt;
&lt;p&gt;&lt;a class="link" href="https://platform.claude.com/docs/en/api/rate-limits" target="_blank" rel="noopener"
 &gt;Rate limits 문서&lt;/a&gt;는 두 층을 분리한다:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;Spend limits&lt;/strong&gt; — 월 단위 소비 한도. Tier 1 ($100) → Tier 2 ($500) → Tier 3 ($1,000) → Tier 4 ($200,000) → Monthly Invoicing (무제한).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Rate limits&lt;/strong&gt; — 분 단위 RPM / TPM (requests/tokens per minute), 모델별로 다름.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;그 위에 별도로 &lt;a class="link" href="https://platform.claude.com/docs/en/api/service-tiers" target="_blank" rel="noopener"
 &gt;Service Tiers&lt;/a&gt; 가 얹힌다:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong&gt;Priority Tier&lt;/strong&gt; — committed spend 대가로 가용성·예측 가능 가격 보장. 응답 헤더에 &lt;code&gt;anthropic-priority-input-tokens-limit&lt;/code&gt; 같은 별도 카운터.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Standard&lt;/strong&gt; — 기본.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Batch&lt;/strong&gt; — 비동기, 정상 capacity 밖 워크로드용.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;이번 발표가 명시적으로 손댄 곳은 &lt;strong&gt;Standard Tier의 Opus RPM/TPM&lt;/strong&gt;과 &lt;strong&gt;Claude Code 5시간 윈도우&lt;/strong&gt;다. Priority Tier 자체 변경은 언급되지 않는다 — Priority는 이미 capacity가 보장된 라인이고, 이번에 풀린 GPU는 &lt;strong&gt;Standard 가입자의 체감 한도&lt;/strong&gt;를 위로 끌어올리는 데 우선 배정됐다고 읽힌다.&lt;/p&gt;
&lt;pre class="mermaid" style="visibility:hidden"&gt;flowchart TD
 Public["Public API (Standard)"] --&gt; T1["Tier 1-4 spend limit"]
 Public --&gt; RPM["모델별 RPM/TPM"]
 Priority["Priority Tier"] --&gt; Commit["Committed spend"]
 Priority --&gt; SLA["가용성 SLA"]
 Batch["Batch"] --&gt; Async["Async, off-peak"]
 Dedicated["대형 enterprise / dedicated"] --&gt; Custom["커스텀 협상"]
 Compute["Colossus 1 신규 capacity"] --&gt; Public
 Compute --&gt; ClaudeCode["Claude Code Pro/Max/Team"]&lt;/pre&gt;&lt;h2 id="같은-시기에-회자된-비교--경쟁사-megadeal"&gt;같은 시기에 회자된 비교 — 경쟁사 megadeal
&lt;/h2&gt;&lt;p&gt;대형 LLM 벤더가 capacity 딜을 마케팅 자산으로 활용하는 패턴은 새롭지 않다.&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://news.microsoft.com/source/2025/01/21/openai-microsoft-stargate-project/" target="_blank" rel="noopener"
 &gt;OpenAI · Microsoft&lt;/a&gt; — Stargate. &lt;a class="link" href="https://www.oracle.com/" target="_blank" rel="noopener"
 &gt;Oracle&lt;/a&gt; · &lt;a class="link" href="https://www.softbank.jp/en/" target="_blank" rel="noopener"
 &gt;SoftBank&lt;/a&gt; 합류해 &lt;a class="link" href="https://openai.com/index/announcing-the-stargate-project/" target="_blank" rel="noopener"
 &gt;수십 GW 규모 추진&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://openai.com/index/openai-and-amd-strategic-partnership/" target="_blank" rel="noopener"
 &gt;OpenAI · AMD&lt;/a&gt; — 다년 GPU 공급 + AMD 지분 워런트.&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://openai.com/index/openai-and-broadcom-strategic-partnership/" target="_blank" rel="noopener"
 &gt;OpenAI · Broadcom&lt;/a&gt; — 자체 AI 가속기 공동 개발.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;각 발표의 공통 문법: (a) GW 단위 capacity 수치, (b) 다년 약정, (c) 모델 가입자 경험 개선 약속. 이번 Anthropic 발표는 같은 문법을 따르되 한 가지가 다르다 — &lt;strong&gt;타사 라이벌 클러스터를 그대로 받아쓴다&lt;/strong&gt;는 점.&lt;/p&gt;
&lt;h2 id="무엇이-뉴스이고-무엇이-아닌가"&gt;무엇이 뉴스이고 무엇이 아닌가
&lt;/h2&gt;&lt;p&gt;&lt;strong&gt;뉴스인 것:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;경쟁사가 만든 frontier 슈퍼컴퓨터를 통째로 임대하는 모델이 시장에 성립한다는 점. AI 인프라가 vendor-neutral commodity처럼 거래되기 시작했다.&lt;/li&gt;
&lt;li&gt;&amp;ldquo;분기 단위가 아니라 한 달 안에&amp;rdquo; 300MW 신규 가용이라는 속도. 이건 보통 새로 짓는 데 18-24개월 걸린다.&lt;/li&gt;
&lt;li&gt;Anthropic이 Trainium · TPU · NVIDIA 3축을 모두 보유하면서, 그 위에 &lt;strong&gt;유동적 임대 capacity&lt;/strong&gt;까지 얹는 4축 전략을 명확히 했다.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;뉴스가 아닌 것:&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;모델 업그레이드 아님. &lt;a class="link" href="https://www.anthropic.com/claude/opus" target="_blank" rel="noopener"
 &gt;Opus&lt;/a&gt; · &lt;a class="link" href="https://www.anthropic.com/claude/sonnet" target="_blank" rel="noopener"
 &gt;Sonnet&lt;/a&gt; · &lt;a class="link" href="https://www.anthropic.com/claude/haiku" target="_blank" rel="noopener"
 &gt;Haiku&lt;/a&gt; 자체에는 변화 없음.&lt;/li&gt;
&lt;li&gt;가격 인하 아님. &lt;a class="link" href="https://claude.com/pricing" target="_blank" rel="noopener"
 &gt;pricing 페이지&lt;/a&gt;는 그대로.&lt;/li&gt;
&lt;li&gt;Enterprise 전용 신규 SKU 아님. &lt;a class="link" href="https://platform.claude.com/docs/en/api/service-tiers" target="_blank" rel="noopener"
 &gt;Priority Tier&lt;/a&gt; 변경 없음.&lt;/li&gt;
&lt;/ul&gt;
&lt;h2 id="궤도orbital-컴퓨트--한-줄-더"&gt;궤도(orbital) 컴퓨트 — 한 줄 더
&lt;/h2&gt;&lt;p&gt;공식 글 마지막 단락에 &amp;ldquo;&lt;a class="link" href="https://www.anthropic.com/news/higher-limits-spacex" target="_blank" rel="noopener"
 &gt;궤도 AI 컴퓨트 capacity 다중 GW 개발 의향&lt;/a&gt;&amp;rdquo; 표현이 들어갔다. SpaceX 측은 &lt;a class="link" href="https://x.ai/news/anthropic-compute-partnership" target="_blank" rel="noopener"
 &gt;더 직접적으로&lt;/a&gt; 표현한다:&lt;/p&gt;

 &lt;blockquote&gt;
 &lt;p&gt;&amp;ldquo;SpaceX is the only organization with the launch cadence, mass-to-orbit economics, and constellation operations experience to make orbital compute a near-term engineering program rather than a research concept.&amp;rdquo;&lt;/p&gt;

 &lt;/blockquote&gt;
&lt;p&gt;가까운 미래에 들어올 deliverable은 아니다. 다만 데이터센터 전력·냉각·부지 한계를 &lt;strong&gt;궤도 &lt;a class="link" href="https://www.starlink.com/" target="_blank" rel="noopener"
 &gt;Starlink 인접 인프라&lt;/a&gt;&lt;/strong&gt; 로 우회한다는 시나리오가 양사 공식 문서에 들어간 첫 사례다.&lt;/p&gt;
&lt;h2 id="인사이트"&gt;인사이트
&lt;/h2&gt;&lt;p&gt;이 발표를 한 줄로 요약하면: &lt;strong&gt;&amp;ldquo;가입자의 한도를 풀어주려고, 라이벌의 슈퍼컴퓨터를 통째로 빌렸다.&amp;rdquo;&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;이게 의미하는 바는 세 가지다.&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;&lt;strong&gt;AI capacity는 이제 commodity처럼 거래된다.&lt;/strong&gt; GPU·전력·냉각·네트워크가 모두 갖춰진 운영 중인 frontier 클러스터를, 라이벌이 한 달짜리 SLA로 받아쓸 수 있다는 사실 자체가 시장의 성숙 신호다.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;단일 칩 종속을 회피하는 다축 전략이 표준이 됐다.&lt;/strong&gt; Anthropic은 Trainium · TPU · NVIDIA · 임대 capacity의 4축. 단일 사고로 서비스가 끊기지 않게 하는 동시에, 가장 빨리 들어오는 라인을 즉시 사용자 한도로 환산하는 라우팅 유연성이 생긴다.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;사용자 입장에서는 단순하다.&lt;/strong&gt; Pro / Max 가입자가 Claude Code를 더 오래 끊김 없이 돌릴 수 있다는 것. 5시간 윈도우 2배 + 피크 감산 제거 + Opus API 상향, 세 가지가 한꺼번에 들어왔다.&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;다음으로 볼 만한 신호: (a) Standard Tier RPM/TPM 표 자체가 &lt;a class="link" href="https://platform.claude.com/docs/en/api/rate-limits" target="_blank" rel="noopener"
 &gt;docs&lt;/a&gt;에서 실제로 갱신되는지, (b) &lt;a class="link" href="https://platform.claude.com/docs/en/api/service-tiers" target="_blank" rel="noopener"
 &gt;Priority Tier&lt;/a&gt; 자체에도 동일한 capacity 가용성 개선이 따라 나오는지, (c) &amp;ldquo;&lt;a class="link" href="https://x.ai/news/anthropic-compute-partnership" target="_blank" rel="noopener"
 &gt;orbital compute&lt;/a&gt;&amp;ldquo;가 구체 일정으로 나오는 시점.&lt;/p&gt;
&lt;h2 id="참고"&gt;참고
&lt;/h2&gt;&lt;p&gt;&lt;strong&gt;1차 발표&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/news/higher-limits-spacex" target="_blank" rel="noopener"
 &gt;Anthropic: Higher usage limits for Claude and a compute deal with SpaceX&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://x.ai/news/anthropic-compute-partnership" target="_blank" rel="noopener"
 &gt;xAI/SpaceXAI: New Compute Partnership with Anthropic&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Anthropic 컴퓨트 megadeal 시리즈&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/news/anthropic-amazon-compute" target="_blank" rel="noopener"
 &gt;Anthropic × Amazon, 최대 5GW&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/news/google-broadcom-partnership-compute" target="_blank" rel="noopener"
 &gt;Anthropic × Google × Broadcom, 5GW&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/news/microsoft-nvidia-anthropic-announce-strategic-partnerships" target="_blank" rel="noopener"
 &gt;Anthropic × Microsoft × NVIDIA 전략 파트너십&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/news/anthropic-invests-50-billion-in-american-ai-infrastructure" target="_blank" rel="noopener"
 &gt;Anthropic의 $50B 미국 AI 인프라 투자 (Fluidstack)&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/news/covering-electricity-price-increases" target="_blank" rel="noopener"
 &gt;데이터센터발 전기 요금 인상분 충당 commitment&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Anthropic 플랫폼 문서&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://platform.claude.com/docs/en/api/rate-limits" target="_blank" rel="noopener"
 &gt;API Rate Limits&lt;/a&gt; · &lt;a class="link" href="https://platform.claude.com/docs/en/api/service-tiers" target="_blank" rel="noopener"
 &gt;Service Tiers (Priority/Standard/Batch)&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://claude.com/pricing" target="_blank" rel="noopener"
 &gt;Pricing&lt;/a&gt; · &lt;a class="link" href="https://claude.com/pricing/enterprise" target="_blank" rel="noopener"
 &gt;Enterprise plan&lt;/a&gt; · &lt;a class="link" href="https://claude.com/pricing/max" target="_blank" rel="noopener"
 &gt;Max plan&lt;/a&gt; · &lt;a class="link" href="https://claude.com/pricing/team" target="_blank" rel="noopener"
 &gt;Team plan&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://claude.com/product/claude-code" target="_blank" rel="noopener"
 &gt;Claude Code&lt;/a&gt; · &lt;a class="link" href="https://claude.com/product/claude-code/enterprise" target="_blank" rel="noopener"
 &gt;Claude Code Enterprise&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.anthropic.com/claude/opus" target="_blank" rel="noopener"
 &gt;Models: Opus&lt;/a&gt; · &lt;a class="link" href="https://www.anthropic.com/claude/sonnet" target="_blank" rel="noopener"
 &gt;Sonnet&lt;/a&gt; · &lt;a class="link" href="https://www.anthropic.com/claude/haiku" target="_blank" rel="noopener"
 &gt;Haiku&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;Colossus 1 / 멤피스 배경&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://www.tomshardware.com/tech-industry/artificial-intelligence/musks-spacex-has-rented-out-access-to-its-supercomputers-220-000-nvidia-gpus-and-300-megawatts-of-ai-compute-power-to-rival-anthropic-musk-says-no-one-set-off-my-evil-detector-antrhropic-also-interested-in-orbital-data-centers" target="_blank" rel="noopener"
 &gt;Tom&amp;rsquo;s Hardware: SpaceX rents Colossus to Anthropic, Musk on &amp;ldquo;evil detector&amp;rdquo;&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://www.datacenterdynamics.com/en/news/anthropic-to-use-all-of-spacex-xais-colossus-1-data-center-compute/" target="_blank" rel="noopener"
 &gt;DCD: Anthropic to use all of SpaceX-xAI&amp;rsquo;s Colossus 1&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://capacityglobal.com/news/anthropic-secures-full-capacity-of-spacex-data-centre/" target="_blank" rel="noopener"
 &gt;Capacity: Anthropic secures full capacity of Memphis data centre&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://en.wikipedia.org/wiki/Colossus_%28supercomputer%29" target="_blank" rel="noopener"
 &gt;Wikipedia: Colossus (supercomputer)&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong&gt;비교 — 경쟁사 megadeal&lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a class="link" href="https://openai.com/index/announcing-the-stargate-project/" target="_blank" rel="noopener"
 &gt;OpenAI · Microsoft · Oracle · SoftBank — Stargate&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://openai.com/index/openai-and-amd-strategic-partnership/" target="_blank" rel="noopener"
 &gt;OpenAI × AMD 전략 파트너십&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://openai.com/index/openai-and-broadcom-strategic-partnership/" target="_blank" rel="noopener"
 &gt;OpenAI × Broadcom 전략 파트너십&lt;/a&gt;&lt;/li&gt;
&lt;li&gt;&lt;a class="link" href="https://news.microsoft.com/source/2025/01/21/openai-microsoft-stargate-project/" target="_blank" rel="noopener"
 &gt;Microsoft 보도자료: Stargate Project&lt;/a&gt;&lt;/li&gt;
&lt;/ul&gt;</description></item></channel></rss>