The 9 best-performing AI stocks of Q3 — and not one is a chipmaker

We ranked 41 US-listed AI stocks by how they actually performed this quarter. Not one of the nine best is a chipmaker.

That is not a stylistic observation. Across the eleven semiconductor names in our universe, the median return since June 30 is −14.6%. Across the five enterprise software names, it is +33.5%. The quarter that was supposed to belong to the people selling AI compute has instead belonged to the people selling AI outcomes.

This piece is a performance ranking with the arithmetic shown, not a buy list. Every figure below is measured from the last close of June to the close on September 11, the most recent session. Q3 does not end until September 30, so these are quarter-to-date numbers and they can still change.

Article Brief

Key Takeaways

5 Points30s Read

1. 01The headline finding–Across 41 US-listed AI stocks, not one of the nine best performers this quarter is a semiconductor company. The median semi is down 14.6%.
2. 02The leaders–Salesforce +58.1%, Palantir +43.3%, Super Micro +36.7%, ServiceNow +33.5%, Microsoft +32.9%, Dell +31.5%, Snowflake +29.3%, Arista +17.5%, C3.ai +16.0%.
3. 03The structure–Median return by layer: enterprise software +33.5%, platforms +7.7%, AI hardware −2.5%, semiconductors −14.6%, power −16.3%, neoclouds −23.9%.
4. 04The rotation is measurable–Rank correlation between first-half and third-quarter returns is −0.53. The top nine had a median first-half return of −21.1%; the bottom nine, +90.7%.
5. 05The caveat–This is a performance ranking, not a buy list, and Q3 has three weeks left to run. One quarter is one observation.

How we built the list

Ranking “AI stocks” is mostly an argument about the universe, so here is ours, stated plainly enough to disagree with.

We took 41 US-listed companies whose business is materially tied to artificial intelligence and sorted them into six layers: semiconductors and compute, platforms and hyperscalers, neoclouds, power generation for data centers, enterprise AI software, and AI hardware and networking. We pulled daily closing prices for every name, computed the return from the final June close to September 11, and ranked. Prices were cross-checked against two independent providers, which agreed to the cent on the names we spot-tested.

No screening on valuation, no analyst ratings, no forward estimates. The only input is what the price did.

The nine

1. Salesforce (CRM) — up 58.1%

The best-performing AI stock of the quarter spent the first half of the year going nowhere; it is still slightly negative year-to-date at −2.3%. What changed is that the agent business stopped being a slide and started being revenue. Combined Agentforce and Data Cloud annual recurring revenue reached roughly $3.9 billion, up more than 210% year over year, with Agentforce alone passing $1.5 billion in ARR. The stock jumped 14% in a single session in late August on a guidance raise and its Anthropic-powered “Claudeforce” tie-up.

Salesforce was the clearest test of the “SaaSpocalypse” thesis — the fear that AI would eat application software rather than enrich it. A 58% quarter is the market marking that fear down.

2. Palantir (PLTR) — up 43.3%

Palantir carried the enterprise-AI narrative through a hostile first half and is, like Salesforce, still roughly flat for the year at −0.4%. The quarter brought a steady drip of commercial validation, including an expanded alliance with PwC US covering enterprise AI deployment, which the market read as a direct rebuttal to a prominent short thesis.

3. Super Micro Computer (SMCI) — up 36.7%

The only server name in the top three, and the cheapest way to own AI server assembly. Super Micro carries the purest AI server mix of the major hardware vendors and trades at a substantial discount to Dell on trailing earnings, a gap the market attributes to governance history and thinner margins rather than to demand. It is up 29.5% year-to-date, so unlike the two names above this is continuation rather than recovery.

4. ServiceNow (NOW) — up 33.5%

The most quantified agent story of the group. Subscription revenue reached $3.88 billion, up 24.5% year over year, and the company said AI annual contract value crossed $1 billion with agentic deployments up ninefold in nine months. ServiceNow entered the quarter down more than 30% from January; it is still −10.1% year-to-date even after a third of its value came back.

5. Microsoft (MSFT) — up 32.9%

The largest company in the top nine by a wide margin, and the one whose 33% quarter represents the most absolute value created. Microsoft sits on both sides of the rotation: it sells the compute and it sells the applications. It also reported plans to more than triple data center capacity, and its expanded partnership with Snowflake put Azure OpenAI models inside another company’s data platform.

6. Dell Technologies (DELL) — up 31.5%

The outlier of the entire study, and the single most interesting name on the list. Dell is up 343.9% year-to-date. It rose 237.6% in the first half and then added another 31.5% in the third quarter, which almost nothing else in the universe managed.

The catalyst was a quarter that is hard to argue with: record AI server orders of $60.9 billion, a record $95 billion AI backlog, revenue of $47 billion up 58%, earnings per share up 273%, and a full-year revenue outlook raised by $25 billion to $192 billion. The stock added another 12% on September 11 alone, closing at $567.29.

7. Snowflake (SNOW) — up 29.3%

Snowflake is the name that arguably triggered the rotation. Its results were read across the sector as concrete evidence of AI-driven growth rather than AI-driven disruption, with chief executive Sridhar Ramaswamy attributing roughly half the company’s growth acceleration to AI. It is up 51.8% year-to-date, the second-strongest YTD figure in the top nine.

8. Arista Networks (ANET) — up 17.5%

The networking layer between the chips and the workloads. Arista is up 49.4% year-to-date, making it one of the few names that has compounded steadily rather than lurching. Its quarter was less about a single catalyst than about the arithmetic of the buildout: the switching and optical interconnect layer scales with deployed capacity regardless of which model wins.

9. C3.ai (AI) — up 16.0%

The most speculative entry and the one we would treat most cautiously. C3.ai is up 16% for the quarter from a very low base and remains −23.3% year-to-date, having fallen further than almost anything else in the first half. We could not identify a single dominant catalyst for the quarter, and a 16% move on a stock trading near $10 is a smaller absolute event than it looks. It qualifies on the numbers; it does not carry the same evidence as the eight above it.

The finding that matters more than the list

Rankings are the least interesting thing in this data. The structure underneath is where the research pays.

Sort the 41 names by layer and the quarter is close to monotonic:

• Enterprise software: +33.5% median
• Platforms and hyperscalers: +7.7%
• AI hardware and networking: −2.5%
• Semiconductors: −14.6%
• Power generation: −16.3%
• Neoclouds: −23.9%

Every layer that sells physical capacity fell. Every layer that sells software rose. There is no chipmaker, no power producer and no neocloud anywhere in the top nine — and the best neocloud in the universe, IREN, still lost 4.2%.

The rotation is measurable, not vibes

Here is the part we have not seen anyone else run.

For each of the 41 names we computed the first-half return and the third-quarter return, then measured the relationship between them. The rank correlation is −0.53 (Pearson −0.39). That is a strong inverse relationship for market data: the better a name did in the first half, the worse it tended to do in the third quarter.

The medians make it concrete. The nine best performers of Q3 had a median first-half return of −21.1%. The nine worst performers of Q3 had a median first-half return of +90.7%.

Look at what happened to the first half’s biggest winners:

• SanDisk rose 726% in H1, then fell 28.2%, with a 50% peak-to-trough drawdown inside the quarter
• Micron rose 266%, then fell 15.5%
• Intel rose 254.6%, then fell 26.3%
• Marvell rose 233.2%, then fell 20.7%
• Arm rose 209.0%, then fell 25.3%

Dell is the exception that proves how rare the combination is: up 237.6% in the first half and up again in the third quarter, alone among the six biggest H1 winners.

TECHi has been reporting the mechanics of this rotation from the other end all quarter. We covered the fund that held the right AI thesis with too much leverage and lost two-thirds of its value in July — its two largest positions were SanDisk and Micron, the first and second names on that list above. We covered Oracle funding its backlog with a $20 billion stock sale, and the circular financing behind Anthropic’s proposed listing. The quarter’s price action is what those stories look like in aggregate.

What this does and does not tell you

It does not tell you the next quarter. A −0.53 rank correlation across one quarter and 41 names is a real pattern, but it is one observation of one rotation, and mean reversion is the single most over-extrapolated signal in markets. Anyone reading this as “sell semis, buy software” is reading it wrong.

What it does tell you is where the earnings evidence landed. The software names that outperformed did not do it on narrative. Salesforce printed $3.9 billion of agent-related ARR. ServiceNow printed $1 billion of AI contract value. Snowflake’s chief executive attributed half its growth acceleration to AI. Dell printed a $95 billion backlog. In each case the market was paying for demonstrated AI revenue rather than for exposure to AI demand.

The capacity layers spent the quarter being asked a different question — not whether demand exists, but who funds the buildout and at what cost. That is the question our Oracle and leverage reporting kept running into, and it is the one the first half never had to answer.

Our prior ranking of the best AI stocks for 2026 was built on business quality rather than quarterly price action. This list is the opposite exercise: no judgement, just the tape. The two disagree in interesting places, and the disagreement is the point.

Three weeks remain in the quarter. Everything above can still change.