Despite reports to the contrary, the Ai buildout continues to grow!
The Great AI Rotation Fall 2026
10 Stocks Positioned for the Next Phase of AI Infrastructure
September 1, 2026 — A RetireFund Working Investment Report
The next leg of AI spending is increasingly moving beyond GPUs and into the bottlenecks surrounding them—memory, optical connectivity, Ethernet/copper networking, custom silicon, power delivery, cooling and physical data-center systems.
That thesis remains intact despite today's weak tape. On September 1, the S&P 500 fell 0.71% and the Nasdaq 1.03% as long-term yields rose and oil prices jumped. That macro backdrop argues for staged entries rather than chasing rebounds, particularly because several of the best AI-infrastructure names have already undergone significant corrections.
My conclusion after updating the numbers is that COHR, CLS and MU remain the three stocks I would begin with, but the full 10-stock ranking contains several opportunities that may be almost as attractive—particularly MRVL and QCOM.
Executive ranking
| Rank | Company | Sept. 1 price | AI role | Current stance | Alpha score |
|---|---|---|---|---|---|
| 1 | Coherent (COHR) | $272.03 | Optical interconnect / photonics | BUY | 9.6/10 |
| 2 | Celestica (CLS) | $296.24 US | AI servers / switches / hardware platforms | BUY | 9.5/10 |
| 3 | Micron (MU) | $933.44 | HBM / DRAM / NAND | BUY, staged | 9.3/10 |
| 4 | Marvell (MRVL) | $210.40 | Custom silicon / networking | BUY | 9.1/10 |
| 5 | Qualcomm (QCOM) | $166.61 | Edge AI + emerging data-center compute | BUY | 8.9/10 |
| 6 | Vertiv (VRT) | $255.97 | Power / cooling | BUY on weakness | 8.6/10 |
| 7 | Broadcom (AVGO) | $368.55 | Custom AI ASICs / networking | BUY after earnings clarity | 8.5/10 |
| 8 | Credo (CRDO) | $206.63 | High-speed copper/optical connectivity | Starter only | 8.2/10 |
| 9 | Applied Optoelectronics (AAOI) | $103.39 | Optical transceivers | Speculative BUY | 7.8/10 |
| 10 | Bloom Energy (BE) | $213.63 | Behind-the-meter data-center power | WAIT / starter only | 7.5/10 |
Prices are September 1 U.S. closes where available. COHR closed at $272.03, MU at $933.44, MRVL at $210.40, QCOM at $166.61, VRT at $255.97, CRDO at $206.63, AAOI at $103.39 and BE at $213.63. CLS closed around $296.
1. Coherent — COHR
My #1 near-term AI-rotation idea
At $272.03, COHR is roughly 38% below its $440 52-week high. That alone doesn't make something inexpensive, but the underlying business has continued moving in the opposite direction.
Fiscal Q4 revenue reached $2.05 billion, while non-GAAP EPS was $1.74. More importantly, management guided fiscal Q1 2027 revenue to $2.2–$2.4 billion and non-GAAP EPS to $1.85–$2.05.
That means the stock has experienced a major multiple reset while the company's operating trajectory remains positive.
The strategic attraction is optical connectivity. AI clusters require exponentially more bandwidth between accelerators, racks and data centers. As accelerator performance rises, connectivity becomes a limiting factor. That places optical transceivers, lasers, photonics and datacom components directly in the AI capital-spending stream.
My price map
| COHR | Action |
|---|---|
| $265–$280 | Buy first tranche |
| $245–$260 | Strong add |
| $225–$240 | Aggressive add if fundamentals intact |
| >$315 | Stop chasing; reassess |
I would expect the first meaningful rerating to take COHR toward approximately $320–$340. A successful FY2027 ramp could eventually justify $375–$425+.
The risk is that optical components remain cyclical. If hyperscaler capex suddenly slows, COHR's multiple can contract rapidly.
12-month working range: $360–$425.
2. Celestica — CLS
Best combination of quality and growth
Celestica may be the strongest business in this group at today's price.
The company has raised 2026 revenue guidance to $20.5 billion and adjusted EPS guidance to $11.30, representing expected year-over-year growth of approximately 65% and 87%, respectively. Management has also said it expects growth to accelerate again in 2027.
At approximately $296, that $11.30 EPS outlook implies a price-to-current-guidance ratio of roughly:
$296 ÷ $11.30 ≈ 26.2×
For a company producing that level of earnings growth, that is considerably more reasonable than the headline stock chart suggests.
CLS's 52-week range is roughly $183.66–$474.02, putting the shares almost 38% below the high.
My price map
| CLS — NYSE | Action |
|---|---|
| $285–$305 | Buy |
| $265–$280 | Strong add |
| $245–$260 | Exceptional opportunity if outlook unchanged |
| >$340 | Don't chase |
For Canadians, the TSX listing is particularly useful. The investment thesis is the same; currency simply changes the quoted share price.
12-month working range: $375–$450.
I would not require CLS to revisit its previous $474 high for this investment to work very well.
3. Micron — MU
The most unusual dislocation of the group
This is where our hedge-fund discussion matters.
Micron was indeed one of the enormous positions held by Situational Awareness, the highly leveraged AI-focused fund that suffered a 67% July loss and was pressured to liquidate most of its roughly $16 billion public-equity portfolio. Citadel purchased much of that portfolio and subsequently disposed of more than 80% of its acquired risk. Reuters confirms the forced unwind.
More specifically, the fund's June filings showed Micron and SanDisk accounting for more than 56% of the portfolio. Micron subsequently fell roughly 24% in July.
So this is one situation where we can reasonably say:
At least part of the selloff was related to portfolio mechanics and leverage rather than Micron's operating performance.
Micron's latest reported quarter was extraordinary: revenue of $41.46 billion, non-GAAP EPS of $25.11 and operating cash flow of $25.39 billion.
MU closed September 1 at $933.44, down 2.64% for the day.
My price map
| MU | Action |
|---|---|
| $900–$950 | First tranche |
| $850–$890 | Strong add |
| $800–$840 | Aggressive add |
| >$1,025 | Wait for consolidation |
I would not buy a full MU position at once because memory is notoriously cyclical and expectations have risen enormously.
Nevertheless, this may be the stock with the best chance of a violent rebound if investors conclude the July forced-selling episode has completely cleared.
12-month working range: $1,150–$1,350, with materially greater upside possible if HBM pricing remains unusually strong.
A correction to our earlier hedge-fund discussion
After checking the primary/current reporting more carefully, I would refine what I told you earlier.
There is strong evidence for forced selling affecting MU, and there is evidence that Bloom Energy was also a major Situational Awareness holding. The fund's other disclosed major holdings included SanDisk, Taiwan Semiconductor and Nebius.
I do not have equivalent evidence tying COHR or CLS directly to that forced liquidation.
So I would no longer characterize their corrections as proven consequences of the Situational Awareness unwind.
That distinction matters.
4. Marvell — MRVL
The stock that almost breaks into the top three
MRVL is becoming increasingly difficult to ignore.
It reported record fiscal Q2 2027 revenue of $2.739 billion, up 37% year over year, while data-center revenue growth accelerated to 46%. Management said AI-related bookings remain exceptionally strong and expects growth to accelerate through the remainder of fiscal 2027.
Yet the shares suffered a severe post-earnings correction.
MRVL closed at $241.45 on August 27, fell more than 10% on August 28, and finished September 1 at $210.40.
That is exactly the type of earnings-related dislocation I look for.
Entry strategy
$200–$215 is attractive.
Below $195, I would become significantly more aggressive provided no customer-loss information emerges.
MRVL is particularly interesting because it touches both sides of the AI infrastructure equation:
custom accelerators and high-speed connectivity.
12-month working range: $270–$325.
If you asked me which stock could displace MU from the top three, MRVL would be the one.
5. Qualcomm — QCOM
The sleeper
QCOM closed September 1 at only $166.61.
That valuation reflects the market's continuing tendency to treat Qualcomm primarily as a smartphone-chip company.
But Qualcomm is increasingly positioning itself across:
edge AI,
automotive,
robotics,
industrial systems,
PC computing,
and now data-center AI.
Its investor materials explicitly highlight a comprehensive data-center roadmap, while the company continues expanding its AI/high-performance-compute positioning.
This is therefore the stock in the ten where the AI optionality is least fully priced in.
That doesn't make QCOM the stock most likely to surge tomorrow. It makes it one of the best asymmetric opportunities if the market begins assigning value to its data-center effort.
My entry range
$160–$170: Buy.
$150–$158: Strong buy.
Below $150 without a fundamental deterioration would be especially attractive.
12-month working range: $205–$235.
For a multi-year Physical-AI portfolio, I would actually rank QCOM above several stocks that rank higher here on short-term momentum.
6. Vertiv — VRT
The picks-and-shovels power winner
AI accelerators are useless without power conversion, thermal management and cooling.
Vertiv sits almost perfectly in this bottleneck.
Q2 was strong enough for management to raise full-year guidance. It now expects approximately $14 billion in 2026 sales, around 31% organic growth, and adjusted EPS of approximately $6.65–$6.75, roughly 60% above 2025 at the midpoint.
At $255.97, however, that implies roughly:
$255.97 ÷ $6.70 ≈ 38× adjusted 2026 EPS.
That is why VRT ranks #6 rather than #2.
The company is excellent.
The valuation is still demanding.
Shares have nevertheless fallen about 20% over the preceding three months.
Entry range
$240–$255: acceptable starter.
$220–$235: very attractive.
Below $215: strong buy, assuming guidance holds.
12-month working range: $310–$350.
7. Broadcom — AVGO
Possibly the best AI company here—but not necessarily the best entry today
Broadcom's fundamentals are astonishing.
Fiscal Q2 AI semiconductor revenue reached $10.8 billion, increasing 143% year over year, driven by custom AI accelerators and networking.
That is precisely where the AI capex cycle is going.
But AVGO reports September 2, meaning buying aggressively tonight would effectively be making an earnings-event bet. Broadcom closed September 1 around $368.55 and remains well below its $495 52-week high.
I therefore prefer:
wait for tomorrow's earnings reaction.
A good report followed by a sell-the-news decline could give us a much better entry.
12-month working range: $440–$500, subject to tomorrow's numbers.
8. Credo — CRDO
Tremendous growth, dangerous expectations
Credo is one of the purest AI-connectivity companies available publicly.
Its just-reported fiscal Q1 2027 revenue was $479 million, up 114.7% year over year, while non-GAAP net income increased approximately 140%.
Yet CRDO fell 8.65% today, closing at $206.63, after already declining sharply from the $280 area in mid-August.
This is interesting—but different from CLS.
CRDO carries much more valuation risk because the market already recognizes the growth story.
Entry range
$195–$210: starter.
$175–$190: meaningful buy.
Below $170: potentially very attractive.
I would not build the entire position today.
12-month working range: $260–$310.
9. Applied Optoelectronics — AAOI
High-risk optical torque
AAOI is the smallest and most speculative optical-networking idea on this list.
The shares closed September 1 at $103.39, down about 4%, versus a 52-week range of roughly $18.50–$233.67.
That enormous range tells you almost everything you need to know about the risk.
If AI optical transceiver demand remains strong and AAOI executes manufacturing expansion correctly, the upside can be enormous.
But this is not COHR.
The balance sheet, customer concentration, margins and execution risk justify a substantially smaller position.
Entry range
$95–$105: speculative starter.
$80–$90: attractive speculation.
I would limit AAOI to perhaps 3–5% of an AI infrastructure basket.
12-month working range: $140–$180, but with unusually wide error bars.
10. Bloom Energy — BE
Fantastic theme; difficult stock price
Bloom's underlying growth is extraordinary.
Q2 revenue reached $1.065 billion, increasing 166% year over year, and management raised full-year revenue guidance to $3.9–$4.2 billion, approximately 100% growth at the midpoint.
And there is a forced-selling angle here too: Bloom was among Situational Awareness's significant holdings and dropped roughly 31% during the July AI unwind.
BE subsequently recovered and closed September 1 at $213.63.
So why only #10?
Because the valuation now assumes a tremendous amount of future success.
The business opportunity may be one of the largest in the entire portfolio, but price discipline is essential.
Entry range
$180–$195: interesting.
$160–$175: buy.
Below $155: strong speculative buy.
At $213+, I would wait.
The portfolio I am planning to construct
If the objective is maximum reasonable alpha rather than simply owning all ten equally, I would not use a 10%/10%/10% structure.
For a hypothetical CAD $100,000 AI-infrastructure portfolio, mine would look approximately like this:
| Company | Weight | CAD amount | Role |
|---|---|---|---|
| COHR | 17% | $17,000 | Optical connectivity |
| CLS | 17% | $17,000 | AI hardware infrastructure |
| MU | 14% | $14,000 | Memory/HBM |
| MRVL | 12% | $12,000 | ASIC/network silicon |
| QCOM | 11% | $11,000 | AI optionality / Physical AI |
| VRT | 9% | $9,000 | Power/cooling |
| AVGO | 8% | $8,000 | Custom compute/networking |
| CRDO | 5% | $5,000 | High-speed connectivity |
| AAOI | 3% | $3,000 | High-beta optical |
| BE | 4% | $4,000 | Distributed power |
That puts 71% of the portfolio in my top five.
I prefer that concentration because the purpose of this portfolio isn't index-like diversification. It is targeted participation in the physical infrastructure bottlenecks created by AI.
But I would not invest the $100,000 today
This is perhaps the most important part of the report.
With Treasury yields rising, equities weak on September 1, and September historically prone to volatility, I would deploy approximately:
Stage 1 — Now: 40%
Establish COHR, CLS, MU, MRVL and QCOM.
Small VRT/CRDO positions are a maybe.
Stage 2 — Another 5–8% sector correction: 30%
Add primarily to whichever of COHR / CLS / MRVL / QCOM falls without a corresponding earnings deterioration.
Stage 3 — Capitulation or catalyst confirmation: 30%
Use this after either:
a genuine September washout,
or evidence that the AI-infrastructure group has established a higher low and resumed leadership.
This avoids the two mistakes investors commonly make during a correction:
waiting forever for the absolute bottom, or
spending all available cash during the first decline.
The three I am looking to buy by next week
If you wanted to keep this exceptionally simple and buy only three,
🥇 COHR
Best rebound setup
$272 is already deep in my preferred first-entry zone.
Its business is growing while its stock has fallen roughly 38% from the high. The connectivity bottleneck is real, and management's FY2027 opening-quarter guidance remains strong.
🥈 CLS
Best business-quality setup
At approximately 26× management's $11.30 EPS guidance, with projected EPS growth approaching 90%, I think the risk/reward is exceptional if Celestica executes.
🥉 MU
Best forced-liquidation/recovery setup
The hedge-fund liquidation is real—not merely speculation—and MU's financial performance is dramatically stronger than its July price action suggested.
Investing a moderate $25,000 as an example
Given today's prices, and although MRVL has become much more attractive after its post-earnings drop.
I like this allocation as entry positions:
| Position | Ultimate allocation | Initial purchase now |
|---|---|---|
| COHR | $10,000 | $5,000 |
| CLS | $8,750 | $4,500 |
| MU | $6,250 | $2,500 |
| Cash held back | — | $13,000 |
So I would put about $12,000 to work now and retain roughly $13,000.
If September gives us another 7–10% AI-infrastructure flush, that cash becomes extremely valuable.
And if the market simply turns higher?
A retail investor would already own meaningful starter positions.
Which one is most likely to “pop” first?
My ranking over the next 30–90 days is slightly different from my long-term ranking:
| Rank | Stock | Why |
|---|---|---|
| 1 | MU | Forced-selling overhang + enormous HBM earnings + potential technical snapback |
| 2 | COHR | Large drawdown + optical-networking rerating |
| 3 | MRVL | Fresh post-earnings dislocation |
| 4 | CLS | Fundamentals increasingly difficult for market to ignore (Canadian gem) |
| 5 | CRDO | Highest beta if connectivity trade resumes |
Over 12–24 months, however, I would put CLS and COHR ahead of MU because their growth is less dependent on memory-cycle economics.
What could invalidate this entire thesis?
There are five things that would negatively impact this scenario!
- Hyperscaler capex cuts. If Microsoft, Meta, Amazon, Google, Oracle or xAI start materially reducing AI infrastructure budgets, almost everything in this portfolio gets hit. (moderate risk)
- Long-duration rates. Today's bond-market selloff matters. Higher yields compress the multiples investors will pay for fast-growth technology. (watch this)
- Networking oversupply. A sudden inventory build would hit COHR, CRDO and AAOI disproportionately.(Minimal risk)
- Memory pricing reversal. MU's earnings are spectacular precisely because memory economics have become spectacular. That cuts both ways.(Moderate)
- AI architecture becoming materially less infrastructure-intensive. This is the existential risk to the entire “trillions of dollars of physical AI infrastructure” thesis.(Moderate risk)
None of those is my base case today.
My September hierarchy
Starter positions by next week
COHR — CLS — MU — MRVL — QCOM
Buy on weakness
Riskier at present
CRDO — AAOI
Wait for a better price
BE
And the most important conclusion from this exercise is that we do not need NVIDIA to keep winning at the same rate for this portfolio to work.
In fact, its strongest scenario may be one in which AI spending keeps climbing but the economic value begins migrating away from the GPU itself toward...
memory, optics, networking, custom silicon, electrical infrastructure and cooling.
COHR, CLS, MU and MRVL now give us four particularly compelling ways to own it at materially reset prices.
I am treating $265–280 COHR, $285–305 CLS and $900–950 MU as actionable first-entry zones right now, while keeping roughly half the intended capital available for September volatility.
