3 AI Chip Stocks, 3 Price Tags — And Why the Cheapest Grows Fastest

A "3 AI chip stocks to buy in September" list makes Nvidia, AMD, and Broadcom look like three flavors of the same bet — grab any of them and you've got AI exposure. They aren't. Line the three up in one snapshot and the market is charging a 4.7x price gap between the fastest grower and the slowest, and the fastest is also the cheapest.

The question that actually separates these names isn't "is AI good?" It's "what is the market asking me to pay for each company's growth, and does the quality of that growth justify it?" The price-to-earnings (P/E) ratio — the share price divided by a year of earnings — is the market's one-number answer. Lower is cheaper per dollar of profit; higher means the market is paying up, usually because it believes more growth is coming. Here are the three, as of the week of September 7, 2026:

	Nvidia (NVDA)	AMD (AMD)	Broadcom (AVGO)
P/E, trailing	27.9x	132.2x	45.3x
Revenue growth, YoY	~83%	~40%	~49%
Gross margin	~74%	~50%	~68%
Net cash / (net debt)	+$66B	+$10B	−$35B
Price vs. 200-day average	above	far above	below
Market cap	$5.4T	$851B	$1.73T

Now read down the P/E column against the revenue-growth column. Nvidia is growing revenue at roughly twice the rate of its two rivals and still trades at less than a third of AMD's multiple. A listicle buries that. The multiple isn't telling you which company is "better." It's telling you which one the crowd has already decided to pay up for — and which one it has decided to treat as the safe, unglamorous incumbent.

That distinction is the whole point, so let's separate the three, because each carries a different factor stack and therefore a different job in a portfolio.

Nvidia: the leader priced like a utility

Nvidia is the name everyone calls "too expensive" because it's worth $5.4 trillion. The multiple says otherwise. At about 28x trailing earnings it's cheaper than both AMD and Broadcom, and it's doing that while growing revenue roughly 83% year over year, holding gross margins in the mid-70s, and converting most of that revenue into cash — about $127 billion of free cash flow over the trailing year, with roughly $66 billion of net cash on the balance sheet.

The 28x is a trailing multiple, and for a company compounding earnings at this rate, trailing multiples tend to compress on their own as the earnings base grows — so the number is a moving target the business is actively pulling down, not a fixed price. Strip the multiple away and the unambiguous numbers say the same thing: it's the fastest grower, the highest-margin business, and the one with the cleanest balance sheet. That's the rare setup where valuation, growth, and profitability point the same direction instead of trading off against each other.

On timing, nothing is screaming. The stock sits above both its 50- and 200-day averages, its 14-day RSI is 54 (neutral, not overbought), and its recent daily volatility is the lowest of the three. Momentum is supporting the trend, not leading a froth. The narrative is doing real work here, too: Nvidia reported $215.9 billion of revenue for fiscal 2026, up 65% from the year before, and at its March GTC event it lifted its cumulative 2025-2027 AI-infrastructure revenue forecast to over $1 trillion — about double its earlier estimate. That's the number the 28x is built to underwrite.

The job this earns: a core quality-growth holding. It's the leg where you own the acknowledged leader without paying a catch-up premium.

AMD: paying for the catch-up, not the current report

AMD is the mirror image. At about 132x trailing earnings it's by far the most expensive of the three — and it's the slowest on revenue growth, at about 40%. Its gross margin is about 50%, well under Nvidia's, and its return on invested capital (about 8%) is the lowest. On almost every current number, the report card is the weakest.

So why 132x? Because the market is pricing the next year, not the last one. The stock is now trading near the top of its 52-week range, and the catalyst is concrete: a 6-gigawatt agreement with OpenAI, under which OpenAI also received warrants for up to 160 million AMD shares that vest as deployment milestones are hit, with the first gigawatt of MI450 accelerators slated to begin ramping in the second half of 2026, plus a similar deployment at Meta. AMD's data-center business just posted a record quarter, and the space is increasingly being framed as a "two-horse race."

That's the story the multiple encodes: AMD is no longer the underdog that can't quite make it. It's being priced as the #2 about to take real share from Nvidia. The risk is that you're paying today for a ramp that hasn't yet shipped at scale. If the MI450/Helios deployments slip, or if OpenAI's commitments convert to hardware more slowly than the warrants imply, the 132x has to re-rate — and at that level, a re-rate is a big move. The momentum is real, but momentum and revisions are timing tools, not a thesis. On the pure factor stack, AMD is the one where you're underwriting the narrative rather than the current numbers.

The job this earns: the high-multiple, high-beta catch-up — the momentum leg.

Broadcom: strong numbers, a tape that disagrees

Broadcom is the one where the tape and the fundamentals are pointing in opposite directions. The company sells the custom AI chips — the so-called XPUs — that the biggest cloud providers design for themselves rather than buy off Nvidia's shelf, and that business is growing fast: it reported $10.8 billion of AI-semiconductor revenue in a single quarter, up 143% from a year earlier, and beat the Street. Yet the stock is about 24% below its 52-week high, and it fell the day after that blowout quarter.

In the snapshot, that shows up as the only one of the three trading below both its 50- and 200-day averages, with a cooling (negative) MACD and an RSI of 43. The multiple, around 45x trailing, sits between Nvidia and AMD: more expensive than the incumbent, cheaper than the catch-up. And it carries the one real balance-sheet wrinkle of the group — roughly $35 billion of net debt.

This is the de-rating story. Strong growth at a mid-market multiple, but the market has been selling the stock into good news, which usually means one of two things: the multiple got ahead of itself earlier and is now coming back to earth, or the crowd is worried about something (customer concentration, the durability of custom-silicon demand) that hasn't shown up in the print yet. Either way, it's the opposite posture from AMD: here the price has lagged the numbers, so the asymmetry runs the other direction — you're buying strength the tape is currently discounting.

The job this earns: the lower-valuation way to own the AI-infrastructure buildout — the value anchor in the barbell, bought where the price and the report card have stopped agreeing.

Put together

That's the structure the snapshot reveals, and it's a cleaner way to think about AI chips than "buy the sector." The three aren't interchangeable; they're three different jobs.

Nvidia is the durable quality compounder, priced like a utility even though it's the fastest grower — the core you can hold without overpaying. AMD is the high-beta expression of "the #2 is finally arriving," where you're underwriting a ramp rather than a report card. Broadcom is the de-rated middle, where a mid multiple and a strong growth print have been discounted by a softening tape.

In an uncertain AI-capex environment, that maps to a barbell: hold the quality core that's cheap relative to its own growth, pair it with the higher-conviction momentum leg if you want more of the catch-up story, and look for asymmetry in the de-rated name where the price has lagged the fundamentals. What would change this read is the next set of print-versus-price checks — does AMD's ramp actually ship, does Nvidia's multiple stay compressed as earnings grow, and does Broadcom's tape and its AI-revenue print keep disagreeing. The factor stack updates on each one.