After Nvidia nears $6 trillion, can US AI stocks still be chased?

After Nvidia nears $6 trillion, can US AI stocks still be chased?

Imagine you're an office worker in Los Angeles or New York with some savings, wanting to ride the AI wave, but market sentiment is subtly shifting. Nvidia has surged to nearly $6 trillion market cap. Can it still double easily? Cathie Wood is selling some hardware stocks, preparing for new directions. Musk is no longer just selling cars, but putting robotaxis and Optimus front and center. What signals does this send? These questions are not to create anxiety but to help you see turning points early.

Today is May 25, 2026, Memorial Day in the US, US stock markets closed. Based on May 22 closing data: NVIDIA (primarily GPU chips, AI computing platforms and solutions) at about $215.33, market cap about $5.25 trillion; cloud service, about $102.13; AI supercomputer and dedicated chips, about $256.78; warehouse automation robot system provider, about $54.03; electric vehicle and clean energy solutions provider, about $426.01.

Over the past two years, the first half of AI was a gold rush for 'selling shovels'. Whoever provided GPUs, whoever built data centers, stood in the spotlight. CPU chip, holding the training entrance, became one of the world's most important companies, no doubt. But now, Wall Street smart money is asking a more realistic, sharper question: After spending so much on training models, can they actually make real money? After buying so many GPUs, can they turn into stable revenue and profit?

Today I'll cut to the core with three questions: First, if NVIDIA is already a $5 trillion behemoth, can it still double as easily as before? Second, if Cathie Wood isn't leaving AI but changing seats within AI, what has she seen ahead? Third, if Musk shifts focus from just selling cars to full self-driving, robotaxis, and Optimus robots, does this mean the real second half of AI has moved from training models to real-world monetization?

These questions sound simple, but the answers hold the key to ordinary investors' success in the next few years. AI is not over; it's moving from hardware euphoria into the second half: application deployment, inference consumption, and physical implementation. I'll break it down for you in plain language.

First, the big picture. From 2023 to 2025, the AI logic was simple: big companies bought GPUs, piled up computing power, trained large models. The market rewarded tool providers, capital expenditure expanded wildly. But entering 2026, investor sentiment changed. They no longer only look at who has the most computing power, but who can turn AI into a sustainable business that enterprises will keep paying for, who can translate the story into profit statement and cash flow. This is the core inflection point.

The actions of top-tier capital are telling. Cathie Wood's ARK fund trimmed some semiconductor hardware that had surged, while adding positions in NVIDIA, Tesla, Cerebras, and autonomous driving and AI application directions. This is not abandoning AI, but avoiding the most crowded, expensive track and turning to places that can truly 'dig gold and sell gold'. On Musk's side, Tesla's narrative is shifting from vehicle deliveries to physical AI. Physical AI is not chatbots on screens, but AI systems that can enter factories, warehouses, roads, drive cars, move goods, and do repetitive dangerous work. That's the source of long-term imagination.

In one sentence: The first half of AI competes on capital expenditure and computing power supply; the second half competes on commercial rollout and real cash flow. Today I'll focus on three main lines most worthy of ordinary investors, corresponding to three companies. Their risk-return profiles are completely different. Understand the difference, and you won't buy blindly. First main line: Enterprise AI applications. Many people hear Salesforce and think 'this company sounds boring'. It doesn't make robots, sell GPUs, or have flashy consumer products. It mainly handles enterprise back-end IT ticket management, customer service systems, internal approvals, compliance processes, and operations automation. Sounds dull, but this is precisely the track most easily overlooked and yet the first to really make money in the AI era.

Why? Ordinary individual users use AI out of novelty, switching between tools, low loyalty, unstable willingness to pay. But enterprises buy AI for real cost reduction, efficiency improvement, reduced labor waste, faster processes, lower operational risk. As long as the system helps the boss save money, they will sign long-term contracts, renew annually, and add budgets.

Salesforce's greatest strength is that it's already deeply embedded in core processes of many large enterprises. Employee computer problems to report, new employee onboarding permissions, customer complaints to assign tickets, financial approvals, security vulnerabilities... these used to be all manual. Now AI agents can directly participate: automatically judge problems, classify, generate solutions, handle repetitive work, and only transfer truly complex parts to humans with complete context.

For example, a large company with tens of thousands of employees may generate hundreds or thousands of IT tickets daily. People forget passwords, can't connect to VPN, can't install software. Each issue is small, but collectively they consume huge IT engineer time. After integrating Salesforce's AI, many standardized issues can be resolved automatically. The company saves not only salaries but also response time, employee waiting cost, and management pressure. This is where enterprise AI is most valuable – not making users feel new and fun, but letting the boss see bills actually decrease and the whole company run more efficiently.

More importantly, cloud computing is not selling AI from scratch. It's already in the customer's system, with massive enterprise data and workflow entry points. Other new AI companies are knocking on doors to sell, but it sits in the conference room like an old friend, saying to the boss: 'These processes that trouble you most every day, I can now automate a large part.' This embedded advantage is hard for many pure AI startups to replicate in the short term.

From real data, in the first quarter cloud computing disclosed 630 large customers with annual contract value over $5 million. This is not trivial; it's truly serving heavyweight enterprises willing to invest long-term, with very high cash flow quality and customer stickiness.

ServiceNow is more like a defensive core position in the second half of AI. Its explosive power is not the strongest, but it's stable, with good cash flow and deep moat, suitable for ordinary investors pursuing long-term stable compound interest. Of course, it also has risks: valuation not cheap; enterprise software sentiment can cause de-rating; big companies have long procurement cycles; Microsoft, Salesforce, Oracle are also pushing AI heavily, competition intensifying.

So operationally, don't buy at highs. Wait for clear pullbacks, trend stabilization, and continued earnings verification before gradually building positions. Track three signals: whether large customer count keeps increasing, whether AI features drive upgrades and add-ons, and whether renewal rates and remaining performance obligations remain strong.

Second main line: Inference computing power

Training large models is like building a school – huge upfront investment. Inference is the daily consumption after the model is trained, like water and electricity bills, long-term and ongoing. As more enterprises use AI and more agents start working, inference demand will snowball. This is the long-term core of the AI computing power track in the second half.

Cerebras didn't follow the traditional path of assembling many GPUs into clusters, but bet on a radical technology route – wafer-scale chip Wafer Scale Engine. It places huge computing cores and high-speed memory on one giant wafer, minimizing latency and energy loss from inter-chip communication. In certain ultra-large model high-speed inference, low-latency response scenarios, it can achieve significant advantages.

OpenAI and Cerebras cooperated to increase 750 megawatts of low-latency AI computing power; AWS also announced cooperation to bring Cerebras systems to the cloud. These moves show big companies are preparing for smoother AI experiences. User experience is real: if AI takes ten seconds to respond, people may give up; when AI Agent writes code, analyzes files, processes enterprise workflows, every lag affects commercial rollout.

My view: Cerebras is a high-elasticity offensive target. It has a differentiated competitive advantage in specific high-value inference scenarios, but NVIDIA's CUDA ecosystem, developer community, and mature deployment experience remain very strong. It won't fully replace but find niches to break through. Suitable for risk-tolerant, technology-tracking aggressive investors. Ordinary investors should build a small observation position, wait for OpenAI cooperation to translate into revenue, AWS channel to open more customers, and revenue growth and gross margin to improve before adding.

Third main line: Physical AI

Tesla Optimus humanoid robot has huge imagination space, potentially changing the labor market, but currently still on the road from demo to mass production. To find physical AI companies with real revenue, real customers, real profit improvement now, look at more pragmatic Symbotic.

Symbotic focuses on providing AI-driven warehouse automation systems for large retailers and supply chain enterprises. Using robots, AI vision, and smart scheduling software, it helps with goods storage, sorting, handling, and outbound. It directly addresses real pain points in the US warehousing industry: high labor costs, recruiting difficulties, seasonal labor shortages.

Its most important customer is Walmart, and cooperation has expanded to all 42 regional distribution centers. In fiscal Q2 2026, Symbotic revenue reached $676 million, up 23% year-on-year, achieving $9 million net profit (compared to loss last year), adjusted EBITDA of about $78 million, significantly improving operational quality. These data are more convincing than any sci-fi story.

Symbotic is a balanced value target. It has long-term demand (rising labor costs are irreversible), high system deployment barriers (once installed, switching suppliers is costly), and good order visibility. But project-based business has delivery fluctuations, and customer concentration needs improvement. Suitable for investors seeking rollout certainty and accepting some volatility.

Key tracking points: whether revenue guidance is met, whether gross margin and EBITDA continue improving, whether new customer breakthroughs occur.

How should ordinary investors handle the second half of AI?

If you're heavily overweight NVIDIA, don't rush to sell all. It's still a core AI asset with strong ecosystem and fundamentals. But if the position is too concentrated, consider keeping a core position, gradually take some profit, free up capital to observe second-half opportunities.

If you're empty now, patience is key. Don't chase highs. Build a small observation position first, wait for stock pullbacks, earnings verification, trend stabilization before slowly adding.

If you hold many second-tier AI hardware stocks, especially those with big gains, high valuations but unclear moats, reevaluate: Is there real revenue? Core large customers? Clear profit path? If only driven by concept and sentiment, consider gradually reducing position and shifting to performance-supported directions.

Finally, three investment iron rules: First, never use leverage to bet on AI; volatility is too high; one drawdown can wipe you out. Second, never chase pure concept stocks; look at real revenue, orders, cash flow, and moats. Third, never switch stocks frequently. AI is a long-term trend evolving over years. Patiently track earnings and control position size is the way.

What are the key inflection points in the second half of AI? Here are the latest signals: Can enterprise AI companies like ServiceNow continue to expand large customer count and AI add-ons? Can inference computing companies like Cerebras turn OpenAI and AWS cooperation into revenue and margin improvement? Can physical AI companies like Symbotic continue to deliver order fulfillment and new customer breakthroughs?

These three signals will determine whether AI can truly turn from story into profit.