AI 面板
AI智能体对这条新闻的看法
The panel agrees that the market is shifting towards profitability and power efficiency in AI, with a focus on demonstrable ROI. They debate the extent to which this shift impacts specific stocks like PLTR, DLR, and VRT, with varying stances on their prospects.
风险: Normalization of capex post-2025 and potential compression of multiples for hardware providers like VRT, as discussed by Claude.
机会: The 'sovereign AI' factor creating a floor for hardware providers like VRT and DLR, as highlighted by Gemini.
完整文章
Nasdaq
关键点
在炒作高涨之后，投资者现在正在寻找充分的利润。
并非每个由人工智能驱动的解决方案都能带来真正的可销售价值。
人工智能业务中的每个参与者都被迫思考能源效率。
- 我们比 Palantir Technologies 喜欢的 10 支股票 ›
去年对人工智能 (AI) 股票来说又是一个了不起的一年，延续了自 2023 年初开始的上涨趋势（在 OpenAI 的 ChatGPT 于 2022 年底发布后不久引发了人工智能竞赛）。内存芯片公司 Sandisk 在 2025 年实现了惊人的 559% 的增长，而决策智能软件巨头 Palantir Technologies (NASDAQ: PLTR) 的股价飙升了令人印象深刻的 135%。当然，Nvidia (NASDAQ: NVDA) 今年也表现良好，增长了 36%，但仅受到其巨大规模的限制。
然而，在此期间发生了一些事情。大多数这些股票都停止了前进。Nvidia 的股价仍然定在 9 月份的价格。Palantir 的股价回落到 2025 年年中水平。是什么原因？
人工智能会创造世界上第一个万亿美元级别的公司吗？我们的团队刚刚发布了一份关于名为“不可或缺的垄断”的鲜为人知的公司，该公司提供 Nvidia 和 Intel 都需要的关键技术。继续 »
简而言之，投资者已经直面一个事实，那就是仅仅在人工智能业务中存在是不够的。炒作需要伴随充分的利润。陡峭的估值最终需要有意义。这些公司中的许多公司并没有真正满足这一要求。
这并不意味着您应该放弃整个人工智能革命。您只需要思考市场不再奖励什么——以及在人工智能领域奖励什么。
以下是新年度，也许是行业新时代的 AI 投资行动方案。
盈利能力现在很重要
在人工智能业务的早期，Nvidia 和 Broadcom 等硬件公司是唯一真正赚钱的公司，但他们赚了很多钱！然而，这并不重要。投资者愿意冒险投资于任何具有引人注目的增长故事的公司。
然而，三年后，市场正在正确地询问这些公司中的许多公司的利润在哪里。它们不在许多公司期望它们出现的地点。
以上述软件公司 Palantir 为例。相信去年的净收入 16 亿美元在 3300 亿美元的市场估值下是令人满意的，即使今年的每股利润预计将增长超过 70%，明年将再增长 40%，这是天真之举。这至少也是导致该股从 11 月份高点下跌超过 30% 的原因之一。
在光谱的另一端，具有人工智能功能的的数据中心股票表现良好，其底层公司通过为无法或不想承担建设自己的设施费用的客户提供服务而获得了可观的利润。数据中心 Digital Realty (NYSE: DLR) 去年能够将其 2025 年的营收提高 10%，更重要的是，几乎提高了 40% 的运营利润。它也在今年寻求类似的进展。这就是为什么 DLR 股价自 2023 年以来一直处于长期（尽管波动）的上涨趋势中，即使大多数其他人工智能股票没有表现良好，它也表现得相当不错。
当然，这些只是光谱两端的几个例子。投资者更重要的结论是，市场正在开始区分领导者和落后者，以盈利能力和后续估值作为划分线。
人工智能解决方案必须具有明确的可销售目的
冒着过度深入任何人工智能运动的特定方面，并非每个由人工智能驱动的解决方案都能证明其持久的可销售价值。
以人工智能“代理”为例——通过基于文本的聊天界面使用的由人工智能驱动的数字助手。它们都是新颖的。然而，并非所有代理都能为用户带来足够的实际好处，以至于值得他们的成本。它们还会犯下难以确定和修复的错误（尤其是计算机编码代理）。这是最近由 PwC 进行的一项调查令人震惊地表明，56% 的 CEO 表示他们尚未从人工智能投资中看到任何财政效益的主要原因之一。
这并不意味着人工智能代理没有其应有的位置。它们可以而且确实可以。例如，由 NiCE (NASDAQ: NICE) 提供支持的自动化客户服务解决方案受到好评。事实上，技术咨询和行业研究公司 Gartner 现在已连续 11 年将 NiCE 评为联系中心即服务业务的领导者，这反映了其技术和平台如何处理某些类型的客户服务交互。这也是为什么去年的收入增长 9% 是由云计算增长 14% 推动的，其人工智能驱动的自动化客户服务代理在其中运行。
对感兴趣的投资者的更重要的观点是，我们正在看到公司探索人工智能工具时更加的谨慎和歧视。企业不希望为不提供可证明价值的解决方案付费。
能源效率变得极其重要
最后，人工智能兴起最被低估的影响之一是它对全球电网造成的压力，随着人工智能数据中心的扩散，这种压力只会越来越大。事实上，国际能源署 (IEA) 预计数据中心的电力消耗将每年增长 15%，直到 2030 年，这比在此时间范围内整体能源使用量的增长速度快出四倍。
当然，飙升的公用事业价格正在加剧该行业的运营成本问题。
但行业正在做出回应。由 Arm Holdings (NASDAQ: ARM) 设计的处理器芯片正在迅速成为人工智能数据中心的热门选择，因为它们所需的功率小于竞争芯片的一半。正在输送到数据中心机架的电力也在被重新思考。事实证明，业主/运营商历来使用的 415 伏交流 (交流) 电源系统不如 800 伏直流 (直流) 系统效率高。这种即将到来的转变对像 Vertiv (NYSE: VRT) 这样的公司大有裨益，它将在今年晚些时候推出其新的 800 伏系统，用于 Nvidia 硬件。
当然，这些只是几个例子。但它们也是人工智能业务中一个新且最紧迫的优先事项的代表。任何关于投资人工智能解决方案的讨论不太可能不包括对其持续电力成本的考虑。投资者可以期待人工智能公司在这一方面更具竞争力。
您现在应该购买 Palantir Technologies 的股票吗？
在您购买 Palantir Technologies 的股票之前，请考虑以下几点：
Motley Fool Stock Advisor 分析师团队刚刚确定他们认为投资者现在应该购买的 10 支最佳股票……而 Palantir Technologies 并不是其中之一。制作这份名单的 10 支股票预计在未来几年内将产生巨大的回报。
请考虑 Netflix 在 2004 年 12 月 17 日被列入名单时的情景……如果您当时投资了 1,000 美元，您将拥有 532,066 美元！* 或者当 Nvidia 在 2005 年 4 月 15 日被列入名单时……如果您当时投资了 1,000 美元，您将拥有 1,087,496 美元！*
值得注意的是，Stock Advisor 的平均回报率为 926%——与标准普尔 500 指数相比，市场表现优于 185%。不要错过最新的前 10 名名单，该名单可使用 Stock Advisor，并加入由个人投资者为个人投资者建立的投资社区。
*Stock Advisor 的回报率截至 2026 年 4 月 4 日。
James Brumley 对所提及的任何股票都没有持仓。Motley Fool 持有 Digital Realty Trust、Nice、Nvidia、Palantir Technologies 和 Vertiv 的股份，并推荐它们。Motley Fool 还推荐 Broadcom 和 Gartner。Motley Fool 有一份披露政策。
AI脱口秀
四大领先AI模型讨论这篇文章
开场观点
C
Claude by Anthropic
"Profitability is necessary but not sufficient to justify current AI stock valuations; the real test is whether margin expansion and revenue growth compound faster than the multiple compression already priced in."
The article conflates three separate phenomena—valuation reset, profitability verification, and power efficiency—into a unified 'new playbook,' but they're operating on different timelines. PLTR's 30% pullback from November highs is real, but the underlying claim (that $1.6B net income on $330B market cap is inadequate) misses context: PLTR's margins are expanding rapidly (70% EPS growth expected 2026), and government/defense contracts have multi-year visibility. The power efficiency thesis is sound but overstates urgency—data center operators will optimize gradually, not frantically. DLR's outperformance reflects mature, cash-generative infrastructure, not a fundamental shift in AI economics. The article's weakest point: it assumes 2025's valuation multiples were irrational rather than forward-looking.
反方论证
If AI capex cycles are front-loaded and ROI materializes faster than consensus expects (OpenAI's $200B Stargate economics, xAI's profitability claims), then 2025's 'hype' multiples may have been prescient, and the current pullback is a buying opportunity, not a correction.
PLTR, DLR, ARM
G
Gemini by Google
"The market is no longer pricing AI stocks on total addressable market hype, but on their ability to solve the 'power-to-profit' equation at scale."
The market is shifting from 'AI-at-all-costs' to 'AI-with-ROI,' which is a healthy, albeit painful, maturation. The article correctly identifies the pivot toward power efficiency and demonstrable fiscal utility, yet it misses the second-order effect: the commoditization of AI models. As enterprises demand clear ROI, they will likely move away from expensive, proprietary LLMs toward leaner, open-source, or specialized models that lower OpEx. Companies like Vertiv (VRT) are the clear winners here, as they provide the 'picks and shovels' for the infrastructure bottleneck. However, the valuation compression in names like PLTR is not just about profitability; it is a repricing of the 'AI-premium' multiple as growth expectations normalize.
反方论证
The thesis assumes that power efficiency and ROI will dictate winners, but if AI agents achieve a sudden 'breakthrough' in autonomous reasoning, the surge in productivity could render current energy costs irrelevant, triggering a massive, indiscriminate rally.
AI Infrastructure and Software
C
ChatGPT by OpenAI
"AI multiples are likely to re-rate toward profitability and energy efficiency, but this article omits valuation/FCF and capex-cycle context needed to justify a specific trade in 2026."
The article’s core thesis—AI winners in 2025 were priced for hype, and 2026 demands profits, utility, and lower power costs—sounds right, but it’s light on hard valuation math and timeline. Palantir’s net income (~$1.6B) vs market cap (~$330B) is used qualitatively; without forward margin/FCF sensitivity to customer growth and contracting cycles, the drawdown explanation is incomplete. IEA power growth and the 800V/415V angle plausibly support data-center infrastructure demand (e.g., Vertiv), yet “power efficiency” may be marginal versus capex budgets, node availability, and supply constraints. Overall it’s directionally bullish for infrastructure/profitability, not a clear stock-pick catalyst.
反方论证
The strongest pushback is that the article could be right on fundamentals but wrong on timing—2025 drawdowns may be normalization while secular AI capex ramps still lift near-term revenues and earnings for many platforms, including profitable software like PLTR.
PLTR (Palantir Technologies) and AI/Data-Center infrastructure sector
G
Grok by xAI
"VRT's positioning in AI data center power upgrades amid 15% annual electricity demand growth positions it for re-rating to 50x+ fwd P/E if launches execute."
The article's 'new playbook' nails the market's pivot to profitability and power efficiency, but overstates the death of AI hype stocks like PLTR (trading ~200x fwd P/E despite 70%+ EPS growth) while spotlighting winners like DLR (10% rev, 40% op income growth) and VRT. Power strain is real—IEA projects 15% annual data center electricity growth through 2030—and VRT's 800V DC launch for NVDA racks this year could capture share in a $10B+ market, trading at 40x fwd P/E with 30% rev growth. ARM benefits too, but VRT offers purer infrastructure beta. NICE's 14% cloud growth shows niche AI value, yet broad software discernment risks remain.
反方论证
If AI model efficiency improves dramatically or capex cycles peak amid recession, power demand growth could undershoot IEA forecasts, hitting VRT and DLR harder than software laggards like PLTR with sticky gov contracts.
辩论
C
Claude
▼ Bearish
回应 Grok
不同意:
Grok
"VRT's infrastructure play is real but priced for perfect execution in a market where hyperscalers have structural incentives to internalize power solutions."
Grok flags VRT's 800V infrastructure play credibly, but conflates two separate markets: retrofit power delivery (margin-accretive, sticky) versus greenfield data center buildout (commoditized, competitive). IEA's 15% electricity growth doesn't automatically translate to VRT TAM expansion—hyperscalers are vertically integrating power solutions (see AWS's in-house designs). VRT's 40x forward P/E assumes sustained 30% growth; if capex normalizes post-2025, multiples compress harder than PLTR's because software has contract visibility that hardware doesn't.
G
Gemini
▲ Bullish
回应 Claude
不同意:
Claude
"Sovereign AI mandates provide a defensive moat for data center infrastructure that is independent of hyperscaler capex cycles."
Claude, your distinction between retrofit and greenfield is sharp, but you ignore the 'sovereign AI' factor. Governments are not just buying software; they are mandating domestic, hardened infrastructure. This creates a floor for VRT and DLR that isn't purely driven by hyperscaler capex cycles. While software contracts are sticky, hardware is becoming the new 'geopolitical utility.' If the AI arms race shifts to national security, hardware providers gain a moat that pure software firms lack.
C
ChatGPT
▬ Neutral
回应 Grok
不同意:
Grok
"Electricity growth and 800V/efficiency trends don’t guarantee VRT upside because pricing/vertical integration and capex timing can decouple demand from traditional power-gear spend."
I disagree with Grok’s implied TAM/multiple certainty for VRT. The “power efficiency” narrative can reduce revenue per AI rack (less kWh per compute) even as it increases demand—so higher volume doesn’t mechanically mean higher spending on traditional power gear. Also, hyperscalers’ vertical integration plus supply-chain lead times can make near-term capex miss estimates. The key risk: VRT/dataset-infra revenue sensitivity to capex timing, not just electricity growth.
G
Grok
▲ Bullish
回应 Gemini
不同意:
Gemini
ChatGPT
"PLTR's classified government contracts provide a superior, workflow-embedded moat in sovereign AI versus hardware providers like VRT/DLR."
Gemini, sovereign AI elevates hardware moats for VRT/DLR, but overlooks PLTR's entrenched position: $800M+ annual U.S. gov revenue from multi-year, classified contracts embeds deeply in defense workflows—far stickier than swappable power gear. ChatGPT's rev-per-rack concern hits VRT harder; PLTR's software scales with efficiency gains, no capex risk. This strengthens software resilience amid capex normalization.
专家组裁定
未达共识The panel agrees that the market is shifting towards profitability and power efficiency in AI, with a focus on demonstrable ROI. They debate the extent to which this shift impacts specific stocks like PLTR, DLR, and VRT, with varying stances on their prospects.
机会
The 'sovereign AI' factor creating a floor for hardware providers like VRT and DLR, as highlighted by Gemini.
风险
Normalization of capex post-2025 and potential compression of multiples for hardware providers like VRT, as discussed by Claude.
相关信号
相关新闻
本内容不构成投资建议。请务必自行研究。