Eight US-listed companies carry most of the AI trade, from the chips that train models to the software that sells them. Here's what each one actually does, what its numbers look like as of September 2026, and where the risk sits. None of this is a recommendation to buy or sell anything.
The rules before the list
This is a map of the AI stock sector, not a shopping list. We don't tell you what to buy, we can't predict where any of these prices go, and anyone who says they can is guessing.
Every figure below is as of September 2026 unless a different date is attached to it. Prices and market caps move daily, so treat the numbers as a snapshot, not a fact that stays true. Company revenue, capex guidance, and valuation figures come from earnings releases and reporting from major financial outlets; where a number is a media estimate rather than a company disclosure, we say so.
One more thing worth stating plainly: AI equities swing hard in both directions. The same stocks that doubled in a year can drop 30% on a single earnings miss. So how should you weigh names that behave so differently? Start with what each company actually does.
Nvidia: the chip supplier everyone else depends on
Nvidia designs the GPUs that train and run most large AI models, and its data center business has become the sector's boom gauge. For the quarter ended July 26, 2026, the company reported revenue of $96.2 billion, up 106% from a year earlier, with data center alone at $89.0 billion. That's roughly a doubling in twelve months, which is why the stock sits near a $5.4 trillion market cap.
The bull case rests on lock-in. Nvidia's CUDA software is the default toolchain for AI developers, so switching to another chip vendor means more than a hardware swap. It means rewriting working code. Its next platform, Vera Rubin, is in full production, and the company guided third-quarter revenue to about $108 billion.
The risk is concentration. Roughly a handful of hyperscale cloud buyers account for a huge share of data center revenue, and Nvidia's own guidance assumes no data center compute revenue from China. If any of those large customers slows its buildout, the top line feels it quickly. The stock also trades at a premium that assumes several more years of the current growth rate.
Microsoft: the cloud and the office suite
Microsoft sells AI two ways. It runs Azure, one of the largest cloud platforms, where customers rent the compute to build their own models. And it bakes Copilot into Windows, Office, and Teams, selling AI as a feature on top of software millions of businesses already pay for each month.
The numbers are steady rather than explosive. In its recent quarter the company reported $90 billion in revenue, up about 18%, with Azure growing 43% and commercial remaining performance obligations, a measure of contracted future revenue, at $678 billion. That backlog is the part bulls point to: money already committed even if new demand cooled.
The flip side is spending. Microsoft is guiding to roughly $175 billion or more in calendar-2026 capital expenditure, mostly for AI data centers. That cash goes out now and pays back over years. If Azure growth slips below the mid-40s range analysts watch, the spending looks heavier.
Alphabet: models, cloud, and the search business
Alphabet develops the Gemini family of models and owns Google Cloud, YouTube, and the world's dominant search engine. That combination makes it one of the few companies that both builds frontier AI and controls huge distribution for it.
Google Cloud has been the standout. Its growth rate has run well ahead of the wider cloud market, and management raised 2026 capital expenditure guidance into the $195 billion to $205 billion range to keep pace with demand. Gemini has also been pushed into Search, Workspace, and enterprise products.
The central question is what AI does to the search business that funds everything else. If AI assistants replace some fraction of traditional searches, the ad model that has carried Alphabet for two decades faces a real test. Bulls argue Alphabet owns the assistant layer too, so the traffic stays in-house. Skeptics say that's unproven.
Amazon: AWS and the retail engine
Amazon is the largest cloud provider through AWS, and AWS is where most of the AI story lives for this company. In its recent quarter AWS sales grew 37% year over year, its fastest pace in years, driven by companies renting capacity to train and serve models. The retail and advertising businesses fund the investment.
The scale of that investment is the headline. Amazon lifted 2026 capital spending guidance to around $220 billion, mostly for AI infrastructure, a figure larger than many countries' annual infrastructure budgets. Management argues the demand is contracted and real; the market has punished the stock on capex days before.
The risk is payback timing. Every dollar of capex is a bet that AI workloads keep growing fast enough to fill the capacity. Amazon's cloud margins are strong, but a slowdown in AI spending would leave a lot of expensive hardware underused.
Where the bull case lives
- Nvidia's CUDA lock-in and doubling revenue
- Microsoft's $678B contracted backlog
- Google Cloud growth and Gemini distribution
- AWS's fastest growth in years
Where the bear case lives
- Customer concentration at Nvidia
- Heavy, years-long capex payback at all three clouds
- AI's unresolved effect on Google Search ads
- China revenue excluded from Nvidia's outlook
Broadcom: custom chips and networking
Broadcom doesn't sell a consumer AI product. It designs custom accelerators for large cloud companies that want their own silicon instead of buying Nvidia, and it makes the Ethernet networking switches that connect data center clusters. It's the second source of AI compute for hyperscalers, in effect.
The growth has been steep. In the quarter reported September 2, 2026, Broadcom posted $29.6 billion in revenue, up about 86% from a year earlier, with AI semiconductor revenue at $16.7 billion, more than tripling. Management guided next-quarter AI chip revenue to $21.7 billion and has talked about the AI segment doubling again in coming fiscal years.
The bet is that hyperscalers keep wanting to reduce reliance on a single chip vendor. The risk is that these are a small number of very large contracts. If one customer redesigns its roadmap or delays a program, the AI revenue line moves sharply. The stock has already run severalfold since 2022, which prices in a lot.
Taiwan Semiconductor: the factory behind the chips
TSMC manufactures the advanced chips that Nvidia, Apple, AMD, and Broadcom all design. Nobody in AI escapes it. It reported second-quarter 2026 revenue up roughly 36% year over year, and raised full-year guidance on AI demand, with August revenue up about 53%.
Its advantage is process leadership. The most advanced nodes, where AI accelerators are built, are essentially TSMC-only right now. Customers pay a premium and book capacity years ahead.
Its risk is geographic. TSMC is headquartered in Taiwan, and its most advanced plants sit there. That concentrates a critical piece of the global AI supply chain in one politically sensitive place. Analysts treat this as a structural risk that no amount of financial performance removes.
Meta: spending to keep up
Meta runs the Facebook, Instagram, and WhatsApp apps, and its AI work is split between recommendation systems that boost ad revenue and its own models and assistants. The company raised 2026 capital expenditure guidance into the $125 billion to $145 billion range, mostly for AI infrastructure.
The bull case is that AI improves ad targeting and engagement, which directly lifts the core revenue engine. Meta has also shipped AI features into its apps and a newer AI app, and the software distribution is unmatched.
The bear case is that this is the most aggressive spending relative to current AI revenue of the group. Meta's capex is close to its operating cash flow, and the market has sold the stock after capex raises. If AI features don't translate into measurably better ad performance, the spending looks like a cost without a matching return.
Palantir: the software layer
Palantir sells AI and data platforms to governments and large companies, most visibly its Foundry and AIP products, which let organizations run models on their own data with access controls built in. It's a software company, not a chip or cloud play.
It has posted the kind of revenue growth that draws attention, and analysts track it closely because software margins are high once a customer is onboarded. The stock trades near its all-time high, and it carries a valuation multiple that has skeptics asking how much future growth is already priced in. Insiders have sold shares at various points, which is common after big runs but always draws notice.
The risk here is valuation stretch more than business failure. Palantir is profitable and growing; the debate is whether the price anticipates years of flawless execution. Government contracts also come with political and procurement timing that a commercial-software investor may not be used to.
How to think about the group
If you're searching for the best AI stock to buy now, the honest answer is that there's no single one, because these names sit at different points in the chain. Chip and foundry companies (Nvidia, Broadcom, TSMC) sell the picks and shovels, and they rise and fall with the pace of data center building. Cloud and platform companies (Microsoft, Amazon, Alphabet, Meta) sell AI as a feature on top of existing businesses, which makes them steadier but ties their AI upside to spending that dents near-term profit. Software (Palantir) sells the last mile, with high margins and the highest valuation sensitivity.
The honest pattern across all of them comes down to the same tension. AI demand is real and growing fast by every disclosed measure, and the money spent chasing that demand stays enormous and borrowed against future returns. Both things are true at once, and which one dominates over the next few years stays genuinely uncertain.
If you take nothing else from this: match any decision to how much volatility you can actually stomach, verify current prices yourself since they change daily, and never let a single AI position grow large enough that a bad quarter hurts more than you can absorb. That's a framework, not a forecast.
