ALMU and APLD: Two Bets on AI's Physical Layer — Photonics and Data Centers

Every AI headline you read is downstream of two invisible bets — where the electrons come from, and how the light moves between chips.

Applied Digital (APLD) grew revenue 139% year over year in fiscal Q3 2026 to $126.6M, has $36 billion in contracted future AI data center revenue, and just signed a 15-year lease with an investment-grade hyperscaler worth roughly $5 billion. Aeluma (ALMU) is a $340M-market-cap photonics microcap that produced $1.2M of revenue last quarter — all R&D contracts — and just received a NASA award to accelerate integrated quantum-dot lasers for silicon photonics.

Same story — AI's physical buildout — different positions on the risk curve. APLD is the operator with signed contracts. ALMU is the R&D shop hoping to sell into it.

I'm buying both. $50/week into APLD, $25/week into ALMU — sized to reflect that these aren't the same kind of bet. This post is the case for the pairing.

APLD is the picks-and-shovels play the AI trade has been missing

Applied Digital owns and operates high-performance-compute data centers, sited near cheap power in North Dakota, that hyperscalers lease for AI training and inference workloads. It's a real-estate-plus-power business dressed up as a tech stock — and that's the whole point.

The Q3 fiscal 2026 print is what changed the story.

Revenue came in at $126.6 million, up 139% year over year — a triple-digit growth rate at a size that actually matters. HPC hosting alone contributed $71M, driven by base rent, tenant fit-outs, and power pass-through at the Polaris Forge 1 facility, which just brought its first 100 megawatts of AI-optimized capacity online. Full-year 2026 revenue is now tracking above $450M — up from guidance in the low $400s just a quarter earlier.

Then the backlog number. $36 billion in contracted future AI data center revenue. That's not a marketing figure — it's disclosed contracted revenue tied to actual signed leases, most of it locked in for 15-year terms. The most recent one is a 200-megawatt lease at Polaris Forge 2 with an investment-grade U.S. hyperscaler worth approximately $5 billion over the term. Polaris Forge 2 is under construction with a second 150 MW HPC data center expected online in calendar 2026, and Polaris Forge 3 (another 150 MW) planned for 2027.

The stock has moved. APLD trades around $25.80, well off its 52-week high of $50.72 but a long way above the $9.79 low. The average 12-month analyst price target is $76.70 — implying nearly a triple from current levels if the buildout executes.

The bear case is straightforward. Data center construction is capex-heavy and behind schedule industry-wide. Every megawatt of power APLD contracts for depends on grid upgrades that may not arrive on time. The company has to fund construction ahead of revenue and will raise capital — dilution or debt. Concentration is real: one hyperscaler represents an enormous share of that $36B backlog, and a single tenant walking away would gut the thesis. And AI training demand could plateau if model performance stops improving with scale, which would leave APLD holding empty megawatts.

The bull case is that hyperscalers can't build fast enough themselves, are signing 15-year contracts because they know it, and APLD is one of the few pure-play publicly traded operators with sited land, power secured, and shovels in the ground.

ALMU is a $340M microcap making the light AI chips will eventually need

Aeluma sits at the opposite end of the risk curve — a research-stage photonics company betting that the next generation of chip-to-chip communication is optical, that the optical piece has to be integrated at wafer scale, and that quantum-dot lasers grown directly on silicon are the winning path.

The financials are microscopic. Q3 fiscal 2026 revenue was $1.2 million, essentially flat year over year. GAAP net loss was $1.8 million; adjusted EBITDA loss $0.9 million. Cash on hand as of March 31, 2026 was $37.8 million — enough runway to fund the next 12-18 months of engineering without a raise, but not indefinitely.

What matters is what's happening around the revenue line, not the revenue line itself.

Aeluma has six new government contracts totaling $5 million in fiscal 2026 alone — including $4+ million from U.S. government agencies to accelerate its heterogeneous semiconductor integration platform for lasers, high-speed datacom, and quantum applications. On April 21, 2026 the company received a NASA award to accelerate commercialization of on-chip quantum-dot lasers for silicon photonics — non-dilutive funding that also formalizes manufacturing partnerships with NASA and its suppliers. The company also announced manufacturing alliances with Tower Semiconductor and Sumitomo Chemical Advanced Technologies — real foundry partners, not lab benches.

The stock has whipsawed accordingly. ALMU trades around $18.56 as of early July, down from an all-time high of $31.49 on May 13. The 52-week range is $10.24 to $31.79. This is what a microcap looks like when narrative and reality argue in public.

The bear case for ALMU is what it always is for microcap science stocks: revenue is not scaling, the science may not work, the cash will run out, dilution is inevitable, and the market for on-chip quantum-dot lasers today is roughly zero.

The bull case is that if silicon photonics for chip-to-chip AI communication becomes a real category — as POET and others are also betting — quantum-dot laser sources at wafer scale are one of maybe three or four technical approaches that could win. Aeluma has government validation, real foundry partners, and a share price that assumes it's still years away from commercial scale. If any of that shortens meaningfully, the stock re-rates hard.

Why pairing them — and why the sizing is different

If APLD is the electric utility company for AI, ALMU is the light bulb company hoping to sell into it a decade earlier.

Owning both is a way to be present at two different points on the same buildout. If AI infrastructure demand continues to compound, APLD gets paid on contracts already signed. That's a call on the current cycle. Meanwhile, if silicon photonics eventually replaces copper interconnects between accelerators — and quantum-dot lasers are one of the three-or-four approaches that get to production — ALMU becomes a scarce piece of the picks-and-shovels stack. That's a call on the next cycle.

The sizing has to reflect the risk asymmetry.

APLD gets $50/week for the same reason I DCA into Netflix, Nike, and SoFi at that size — real revenue, real contracts, real balance sheet, and a growth rate that says the story is more likely working than not. ALMU gets $25/week for the same reason POET and INFQ do — a company whose upside is asymmetric but whose downside is real. Combined, that's $75/week into this pair, roughly $3,900 over a year.

Here's what $50/week into APLD would have looked like across the recent four-week window:

After four weeks: 7.12 shares at an average cost of $28.09 — meaningfully below the $32 you'd have paid lump-summing at the top of the window and only slightly above the current price. The stock will move 20% either direction on a single lease announcement or hyperscaler earnings print. DCA is how you stop letting that determine your entry.

For ALMU, the moves are larger and the reasoning is the same. Buying $25/week across a stock that has swung from $10 to $31 and back to $18 in a year averages your cost into the range where the science-project economics still price in a lot of doubt.

What would change my mind

For APLD: Polaris Forge 2 slips past its calendar 2026 online target by more than one quarter, OR the hyperscaler behind the 200MW lease renegotiates terms or delays takedown, OR a single quarter of HPC hosting revenue misses guidance by more than 15%. Any of those and I cut the position by half. All three and I exit.

For ALMU: the cash burn accelerates and a dilutive raise becomes necessary in the next 12 months without a new government contract catalyst, OR the Tower Semiconductor and Sumitomo Chemical manufacturing partnerships fail to produce a first prototype wafer by end of calendar 2026, OR a competing silicon photonics approach (thin-film lithium niobate, integrated III-V lasers, silicon-germanium) gets to production first at a hyperscaler.

If none of those happen, I keep buying. If they do, I stop fast.

The pattern in these posts

I've now written about eight speculative-adjacent stocks across four pairs — OKLO/IonQ, POET/INFQ, Figma/BMNR, and now ALMU/APLD — plus the more conservative pairs like Netflix/Nike, PayPal/SoFi, Circle/CoreWeave. The pattern I keep landing on is that the interesting bets are almost never single stocks. They're small combinations that let one side of the bet be right for reasons the other side wouldn't have caught.

APLD is going to pay you if AI training demand keeps growing at the rate it has been. ALMU is going to pay you if AI training demand keeps growing at the rate it has been and the physical infrastructure of accelerator interconnects moves off copper and quantum-dot lasers win the race to production. Those are stacking probabilities, not the same probability twice. That's why you own the second one small.

The other pattern is that speculative positions belong in a portfolio — they just shouldn't be the portfolio. $75 a week between these two, out of whatever you can afford to allocate to investing total, sits in the right neighborhood of risk-to-reward for me. Yours might be different. The point is to size deliberately and admit it's speculation.

---

This is not financial advice. I own both ALMU and APLD. I'm sharing my personal research and strategy. Small-cap and pre-revenue companies can and do go to zero. Always do your own due diligence before investing.