The AI Rally Is Back: Ceilings Break and 5 Stocks to Watch

The AI trade was stuck sideways for four to five months starting in May, and into mid-September many asked why the tape still choked. Now four benchmarks are answering together: SOXX clearing 555 to 568 , SMH pushing through 600 to 604-605 , AIS breaking 74-75 to 76.5 and AIQ moving above 65 to 66.6 . When those ceilings fall in unison Luke Lango reads it as a melt-up regime – the long consolidation is over and an upside resolution is underway.

How the ceilings broke and why now

Technically the pattern is classic: sharp rally, deep pullback, tightening tops that form a ceiling, then a volume-backed breakout. The shelf built around July-August highs finally gave way in September. Lango ties the timing to liquidity , not headlines. A viable path to de-escalate the Iran file, oil back under $80 and potentially toward $70 , headline CPI creeping below 3% and core toward 2% , less need for the Fed to hike, and the 10-year yield sliding toward 4.5% or lower together create room for multiple expansion alongside an earnings recovery .

First stop is AMD , breaking above 580 toward 620 . On September 21 the stock touched $615.52 and pushed the company above $1 trillion in market cap , capping a move that is up ~188% year-to-date versus Nvidia up ~22% , a 166-point gap that Benzinga flagged as AMD's widest lead of the AI boom. The September ~10% price lift across accelerators, consumer GPUs and chipsets shows demand turning into pricing power. The core call is CPUs, not GPUs : as inference scales, orchestration, data movement and memory coordination demand a central processing unit , and the story of Lisa Su – unknown on the racetrack a decade ago, now leading a trillion-dollar company – has become the symbol of that shift.

Next is Meta and its Muse assistant , a test of how fast usage becomes dollars . The 11-day 2.6 million claim approaches ChatGPT's 3.5 million record, while Sensor Tower offers a sober cross-check that still confirms a record: 730,000 downloads in five days , over 902,000 in six days and more than 2.5 million in total , making Muse the No.1 free app on US iOS ahead of ChatGPT versus 773,000 for the prior Meta AI app in the same window. With no cloud business to recycle compute into revenue , Meta's AI spending looked harder to justify; now the chart pressing $744 toward the 750-800 shelf alongside 12.9x forward EBITA and 20x forward earnings discounts gives both technical and fundamental momentum. Lango's point is simple: if anyone can turn engagement into money , it is Meta.

AI CPUs step into the spotlight

ARM jumped from a 290 ceiling to $333 after reclaiming both the 100- and 50-day averages from a long low near 150 , a straight-line break that looks attractive on the chart. Fundamentals back it: a record $1.49 billion quarterly revenue , licensing up 29% to $819 million , royalty up 11% to $671 million and full-year $4.92 billion with royalty +21% and licensing +25% , marking a third straight year above 20% growth since the IPO . Lango quotes an accelerating 24%, 36%, 30%, 63%, 53% revenue trajectory that matches the official acceleration, but the new engine is silicon : the AGI CPU delivers up to 2x performance per rack and up to $10 billion in capex savings , with more than $2 billion of demand for fiscal 2027-2028 and a path to $15 billion by 2031 . The company sees 4x CPU capacity per gigawatt as agentic workloads scale, margins easing from 97% licensing toward high-40s to low-50s EBITDA , and a $500 year-end target that assumes 40-60%+ earnings growth holds.

SanDisk looks like a contradiction: 6.5x forward EBITA, 4x forward earnings and 6.6x forward EBIT multiples imply a business heading for collapse, yet fiscal Q4 revenue was $8.97 billion up 51% sequentially , GAAP net income $6.90 billion or $43.97 per share , full-year revenue $20.248 billion up 175% year over year , datacenter $5.153 billion up 437% and edge $12.160 billion up 195% . Valuations have already repriced – P/sales 13.36x from 8.35x , EV/Revenue 13.00x from 8.02x , free cash flow +1,577% – with R&D to revenue down to 6.56% and forward P/E around 8.16 showing efficiency gains. Technically Lango waits for confirmation, and he gets it: retaking the 100-day as support with a series of higher lows and higher highs , which underpins the call for above $2,500 within months ; without that confirmation a cheap multiple alone would just trap capital.

The sleeping giant wakes

Lango labels Qualcomm the sleeping giant of the trade. A new AI data-center interconnect architecture with Lumen and Corning plus an imminent Snapdragon Summit have made the data-center line suddenly tangible as a near-term profit driver , waking the stock. Technically the bounce retook the 200-, 100- and 50-day averages with higher highs and lows off the late-July trough . The long-term lever is physical AI : as intelligence moves to robots, glasses, cars and every edge device , Snapdragon is positioned as the H100-class backbone for on-device AI . With $44.28 billion in last reported revenue still driven by core franchises, the added data-center kicker creates a short- and long-term one-two punch that Lango thinks can double the stock in roughly 12 months, if not in three .

The macro glue tying the five names together is liquidity . Iran reportedly approaching the United States with a proposal to reopen the strait if the blockade eases, and talk of ending the conflict after the midterms , gives a viable path to prolonged de-escalation whether via ceasefire or settlement. Markets may not differentiate – either path could put oil back under $80 toward $70 , ease inflation, let CPI slide below 3% and core toward 2% , keep the Fed on hold, drag long Treasury yields lower toward 4.5% or less , and revive liquidity and multiple expansion alongside earnings. That, Lango argues, is why the AI trade that chopped from May into the third week of September is finally resolving higher now, and why it is a good time to be shopping .

The SOXX versus SMH split helps explain why breadth matters. SMH holds Nvidia at 15.1% and TSMC at 9.5% in a concentrated 25-name basket, while SOXX caps its largest names – Micron 11.3% and AMD 9.1% – across ~30 holdings. In early July both fell – SOXX -10.63% and SMH -8.54% month-to-date – yet flows stayed positive with SOXX +$6.13 billion and SMH +$1.60 billion ; the Roundhill Memory ETF (DRAM) fell 13.83% to show how memory beta overshoots. The $6.3 billion 59% year-over-year growth in data-center and AI revenue flagged in earnings season gave a fundamental anchor, and the current breakouts at SOXX ~568 and SMH ~604-605 now read as price confirmation of those flows .

Putting it together, the tape is constructive but layered , with each name proxying a different AI layer: CPU orchestration at AMD , usage-to-cash conversion at Meta , silicon scale at ARM , memory-cycle pricing power at SanDisk and edge-AI proliferation at Qualcomm . The calendar of catalysts is also spread: Muse monetization at Meta , AGI CPU ramp at ARM , edge-device rollout at Qualcomm , volume-price sustainability at SanDisk and pricing-power durability at AMD . The caution is symmetric: estimates beyond 2028 are already wavering , with the assumed peak pushed from 2027 to 2028 and then to 2028-2029 ; a major regulatory shift is the main downside Lango assigns low odds to, but fragility in out-year forecasts makes chart confirmation before chasing fundamentals the key discipline.