最近的一项分析表明，虽然企业人工智能支出是真实的，“需求骤降”的说法是不准确的。相反，重点在于分配，一小部分高使用率客户为 OpenAI 和 Anthropic 等公司贡献了大部分收入。正在实施控制措施，例如禁用快速模式和默认使用成本较低的模型，目前编码应用程序占 OpenAI 和 Anthropic 两家公司收入的 70% 以上。 AI
影响 表明人工智能 API 提供商的收入将继续强劲增长，由重度用户驱动，并强调了成本管理策略的重要性。
排序理由 对企业人工智能支出模式和用户行为的分析，而非直接公告或产品发布。
AI 生成摘要 · Google Gemini · 来自 6 个来源。 我们如何撰写摘要 →
Budgeting is real. A demand cliff is not. We estimate 90th+ percentile customers drive most revenue. No material 2H26 AI-budget risk. We expect Anthropic/OpenAI API businesses to maintain current net-new m/m growth rates. (6/6) https://t.co/tgNssBBGVk
Our model: coding drives >70% of OpenAI + Anthropic ARR today; Anthropic is 90%+ B2B vs OpenAI ~60%. Next: cyber (Mythos re-release dependent), then white-collar. TaaS providers exceed $4B ARR; our Bedrock estimate puts total AWS growth above street this quarter. (5/6)
"No cap" can mean too little access for cost to matter—or enough output to justify spend. Controls go beyond allowances: Opus 4.8 + Fast-Mode off, Sonnet default. Employees can draft in unmetered M365 Copilot before metered Claude/Codex. The meter starts with routing. (4/6) https…
Spend concentrates twice. Ramp annual spend/employee: Median: $136 P90: ~$7,300 P99: ~$90,000 (~660× median) Inside firms, most employees stay below caps while a few consume heavily. At one aerospace firm, power users burned a $250 monthly allowance in 4 days. (3/6) https://t.c…
Meta's "Claudeconomics" dashboard ranked 250 users: "Token Legend," "Cache Wizard," agents researching for hours to burn tokens. It was shut down 2 days after The Information's report. Uber set a $1,500/mo/employee cap. (2/6)
Was widespread Tokenmaxxing ever really here? Meta burned 60T+ tokens in 30 days; one employee used ~280B. Uber burned its annual Claude Code + Codex budget in four months. We spoke with 50+ enterprises. The real story was not a budget wall, but allocation.👇️ (1/6)🧵 https://t.c…