🚨 BREAKING:
🇺🇸 FED WILL ANNOUNCE U.S. INFLATION DATA TOMORROW AT 8:30 AM ET
IF INFLATION < 3.5%: MARKET WILL EXPLODE
IF INFLATION = 3.5%–3.6%: MARKET WON’T MOVE
IF INFLATION > 3.6%: HUGE CRASH IS COMING
ALL EYES ON THE RELEASE TOMORROW!!
📝UBS 发布了一份长篇报告:《Token Costs – What Customers are Saying》(Token 成本:客户反馈实录)。
💡企业确实感受到了更高的账单压力,但结论是,实际上没有人放慢 AI 使用。他们大多只是设置一些防护栏,并削减其他领域的支出,同时继续推进 AI 落地。
📌结论是,UBS 认为Google Cloud 和 AWS可能会因为自有芯片和模型而在这里获得成本优势。随着时间推移,这可能帮助它们从 Microsoft 和其他公司手中拿到一些份额。
📋UBS 核心结论:
📊We conclude: 1. This is quickly becoming an issue, but the risk to the “AI trade” may be overstated and some perspective is important here.
我们的结论:1. AI 成本问题正在快速凸显,但市场对 “AI 交易” 的风险可能被过度放大,我们需要更客观地看待这一现象。
企业设置成本控制机制,是正常的企业成本管控行为,并非出现大规模停止 AI 投入的恐慌性行为。
🔍The fact that enterprises are putting guardrails in place to control AI costs strikes us as normal enterprise cost-containment behavior. Not a single check was being an alarmist and talking about slamming on the AI brakes. That is just not happening.
企业为控制 AI 成本而设置 “护栏” 的行为,在我们看来是正常的企业成本管控行为 。调研中没有任何一家企业表现出恐慌,也没有企业表示要 “踩刹车” 停止 AI 投入。这种情况并未发生。
🚀Even Uber (admitting that they blew through their full year AI budget in a single quarter) remains full steam ahead in terms of its planned AI use and has set very high token limits per engineer.
即便像 Uber 这样承认 “单季度就花光了全年 AI 预算” 的企业,其 AI 应用计划仍在全速推进,且为每位工程师设置了非常高的 Token 使用上限。
🤝2. The AI ecosystem – AI model providers as well as hyperscalers – is likely to react and focus more on token efficiency.
2. AI 生态(包括 AI 模型提供商和超大规模云服务商)很可能会做出反应,并更加关注 Token 使用效率。
📉We wonder if this might limit near-term model and compute price increases and if this could impact hyperscaler share if the 1P chips/models from Google Cloud and AWS give them a token cost advantage over Microsoft, Oracle and others.
我们认为,这可能会限制短期内模型和算力价格的上涨。同时,如果 Google Cloud 和 AWS 的自有芯片 / 模型能为它们带来相较于微软、甲骨文等对手的 Token 成本优势,这可能会影响超大规模云服务商的市场份额格局。
💻3. In terms of the impact on SaaS/apps firms, we wonder if early-stage AI cost push-back could slow their transition to a usage-based pricing model (revenue accretive to software firms, but resulting in higher AI spend by customers) and/or motivate enterprises to cut elsewhere, creating an even tougher spending backdrop for core or non-AI software spend.
3. 对于 SaaS / 应用企业而言,早期阶段的 AI 成本压力,可能会减缓它们向基于用量的定价模式转型的速度(这种模式对软件企业有利,但会导致客户的 AI 支出增加),或促使企业削减其他领域的支出,为核心业务或非 AI 软件支出创造更严峻的环境。
Semianalysis 又给机构客户发了小作文:
“Our biggest Computex 2026 takeaway is that both CPO and 800VDC have a high likelihood of delays compared to original ramp expectations.”
“Industry chatter suggests that hyperscalers are pushing back on adoption of the single-ended 800VDC architecture”
“我们在2026台北电脑展(Computex 2026)最大的核心结论:共封装光学(CPO)与800伏高压直流供电(800VDC)两大技术,相较此前产业原定扩产放量时间表,大概率出现落地延期。”