Five More Earnings Updates!

🌾 Welcome to StableBread's Newsletter!

The next time you find yourself on EDGAR, consider using my SECSift reader instead!

It turns SEC filings into clean, readable documents. You can search every word, export anything, merge tables, highlight what matters and gray out the rest, and compare against previous filings.

Here’s a 3-minute interactive demo (viewable on your computer) and video I made walking through the entire tool.

As a reminder, annual plans (for more AI usage, extra storage, merging tables, exporting) are 25% off with code “SB25” through August 31.

It’s been another week of packed earnings! This post breaks down my quarterly analysis for five names from the write-up archive: Quantum (Monday), Elmet and QXO (Thursday), plus Boost Run and DigiPowerX (Friday).

Reviva reported on Wednesday too, but it’s getting its own write-up later this week. There’s more to the company’s story than just its quarterly earnings!

If you missed it, check out last Sunday’s recap of Toast, Root, Fluence, Energy Fuels, Celsius, and Doximity.

This is another very dense post so I hope you find it helpful!

Quantum (Nasdaq: QMCO) reported its first non-GAAP profitable quarter since 2023 on Monday, and the stock closed the week at $26.11/share.

I published my write-up on May 6 at $7.36/share and called it a watch-list name. The stock is up 254.8% since! Here's what I learned following their earnings on August 10.

What I Got Right

What I Missed

What Beat Even the Bull Case

Dilution Question

The write-up's core question was what fraction of the upside existing shareholders capture through the dilution stack.

So far, more than I modeled. The share count nearly tripled, and the stock more than tripled anyway.

At $26.11/share, that leaves Quantum at 3.3x trailing EV/Sales (~$1.03B market cap less $54.4M cash, on $296M TTM revenue) against 0.85x when I passed.

The operating story keeps outrunning the share count. From here, the multiple has to hold while growth is capped by how many tape drives IBM can deliver.

The Elmet Group (Nasdaq: ELMT) reported its second quarter as a public company on Thursday. I covered the company in a full deep dive in early July.

The Quarter

First, the headline numbers:

The margin number is lower quality than it reads.

CFO Mike Lee said about half of CMC's growth in the quarter, or slightly more, came from capturing the tungsten spike, either by leveraging existing supply agreements or by selling material at market against what was already on the shelf.

That's an inventory windfall on a price move, not operating leverage, and it doesn't repeat if tungsten prices flatten or fall.

Though the same tungsten spike worked against the smaller EMP division. Its gross margin fell to 18.1% from 27.5%, because material costs ran ahead of the fixed long-term contracts on the big circulators it builds for CERN and Fermi. CEO Peter Anania said it "caught us off guard."

So one division sold into the spike and the other, albeit smaller division, got caught by it.

The working capital cost is also worth highlighting:

GAAP was a $(4.5)M loss, but $14.2M of SBC ran through the quarter and $12.9M of that was one-time IPO vesting. Adjusted net income was $5.2M.

Backlog

The backlog is central to the Elmet thesis, and two items stand out:

That's the second quarter in a row where the record was set before the catalyst arrived. Last quarter it was the molybdenum import restrictions, which Lee said hadn't shown up in orders yet. This quarter it's the appropriation cycle.

In other words, the appropriation cycle is still ahead of the order book, and the backlog is setting records without it.

That mix also carries the margin story. Management still targets 30% gross margins in 4-5 years and says AD&G runs above the company average, so the segment doubling its backlog is the one pulling margins toward the target.

What hasn't improved is the control side. Disclosure controls are still not effective as of July 3, with the same two material weaknesses in segregation of duties and IT controls. However, Elmet says it hasn't identified a material misstatement resulting from them.

Paper Gains

The headline reports that adjusted EBITDA was up 57.9% to $8.9M. But the sequential comparison is misleading. $3.7M of the $9.2M Elmet reported in Q1 came from paper gains, not operations, and Q2 had none of that.

Adjusted EBITDA was $8.9M in Q2 against $9.2M in Q1, so it reads as a step down even though revenue rose from $56.0M to $66.4M. The difference is what each number includes.

Elmet's adjusted EBITDA adds back interest, taxes, D&A, and stock comp. It doesn't add back gains and losses on investments, so those stay in the number.

Elmet holds a stake in EQ Resources (ASX: EQR), the Australian tungsten miner, which began as options granted under a five-year supply agreement.

Q1 carried a $3.1M gain on those options. It also carried a $636K gain on marketable securities. Together that's $3.7M of paper gains inside the $9.2M (~40% of the number).

Q2 ran the other way, an $881K loss on the same derivative against a $445K securities gain, for a $436K net drag.

Take those gains and losses out of both quarters and the operating business looks like this:

So the operating number didn't stall. It rose ~70% sequentially and the margin gained 420bps.

In my write-up I said true earning power was somewhere between the GAAP and adjusted figures, because the adjusted number carried that EQR gain. This quarter settles it. The clean operating result is better than the headline suggests, not worse.

Elmet exercised those options in late June and now holds ~23.7M EQR shares outright, so the position moves into marketable securities from here. That changes the line it reports on, not the effect: every move in the EQR share price will keep running through reported and adjusted EBITDA.

A rally flatters the number, a selloff cuts it, and neither says anything about how Elmet is operating.

Valuation

At Friday's $18.82/share close, Elmet is a $573M market cap and ~$513M enterprise value net of cash. That's 16.1x TTM adjusted EBITDA of $31.8M and 2.2x TTM revenue of $228.5M.

When I wrote it up at $19.25/share it was 17.5x and 2.4x. The stock slipped ~2% while TTM adjusted EBITDA grew 11.3% and backlog went from $113.3M to $131.5M, so it's cheaper on both sides of the ratio.

QXO (NYSE: QXO) reported Q2 on Thursday, its first report since my June write-up.

The headline was gross margin of 24.7%, up from 21.1% a year ago. But last year's number was held down by a one-time accounting charge, and on a clean comparison the margin actually fell.

Q2 2025 carried an $80M inventory step-up from the Beacon close. Purchase accounting writes acquired inventory up to fair value, and the markup runs through cost of products sold when that inventory is sold.

QXO's own reconciliation strips it out, and doing that lifts last year's margin from 21.1% to 25.3%. This quarter has no step-up in it, so the clean comparison is 25.3% against 24.7%, down 60bps.

Bought Growth

Adjusted EBITDA grew at less than half the rate of revenue:

Growth is running well ahead of profitability, which is what happens when the growth is bought rather than earned.

But the decline comes from what QXO bought, not from how the business is running.

Kodiak closed April 1 and added $595M of revenue in its first quarter. It sells lumber, trusses, windows and doors, which carry thinner margins than roofing.

Roofing dropped from 76.8% of sales to 61.7% y/y:

Since QXO bought $595M of revenue at a lower margin than the base business, the blended margin fell.

Q3 should reverse that.

TopBuild closed July 1, the first day of the quarter. It did $6.2B of 2025 revenue at an 18% adjusted EBITDA margin, more than double the 8.4% QXO ran in Q2.

TopBuild would be the first acquisition that lifts the blended margin rather than pulling it down.

So the Q2 margin says what QXO bought. Q3, with TopBuild in for the full quarter, starts to indicate whether the roll-up works.

Preferred Stack

The other side of the quarter is the preferred stack. QXO reported $130M of adjusted net income, and common shareholders were entitled to $73M.

Here's where the other $57M went:

That's 44% of adjusted net income gone before common sees anything. A year ago the preferred took $26M a quarter. Now it takes $54M, because the $2.0B Series C issued to fund Kodiak started paying in Q2.

Over the first half it's worse. Adjusted net income of $73M for the six months becomes $(11)M attributable to common. The preferred took all of it and more.

So QXO can report a profitable half while the piece belonging to common is negative.

And it doesn't stop there. The last $1.0B of the Series C commitment was drawn on July 1 to help fund TopBuild, so Q3's dividends accrue on the full $3.0B.

The share count grew too. Adjusted diluted shares are 911.8M against 702.0M a year ago, and none of TopBuild's 312M new shares are in that number yet, since the deal closed July 1. Post-close shares outstanding are 1.037B, almost exactly the 1.04B I modeled in June, so this is expected dilution, not anything surprising.

Valuation

At Friday's $14.51/share close that's a $15.0B market cap. Add the $7-9B of post-close net debt and ~$3B of preferred and pro forma enterprise value is $25-27B.

Against $2.1B of combined EBITDA that's 11.9-12.9x, or 10.4-11.3x with the full $300M of synergies. At $16/share in June those figures were 13.6x and 11.9x.

So the setup is unchanged and slightly cheaper. But that multiple is on a combined company at a 12% margin, and QXO standalone ran 8.4% in Q2 and 5.5% for the first half.

I invested in Boost Run (Nasdaq: BRUN) pre-merger and covered the company in a deep dive when it was still Willow Lane (WLAC).

Plan vs. Actual

Boost Run grew revenue 270% in Q2 and is still running behind the $170-180M its merger proxy projected for 2026.

As a reminder, the projections Willow Lane filed to get the deal approved, modeled a quarterly ramp to the top of that range: $16.5M in Q1, $35.8M in Q2, $54.7M in Q3, and $73.0M in Q4.

Q1 came in at $10.96M against $16.5M, a 34% miss. Q2 came in at $31.1M against $35.8M, a 13% miss.

Regardless, the gap is closing. Revenue almost tripled sequentially and the shortfall to plan more than halved.

First-half revenue is $42.1M against the $52.3M modeled, 20% short. To reach even $170M for the year, the back half needs $127.9M against the $127.7M the proxy laid out.

The company itself has stopped talking about the range. Neither the release, the deck, nor the call mentions $170-180M, and everything now runs through ARR.

On that measure Boost Run looks better:

So the back half asks less of them than the first half already delivered.

$1.9B TCV

The other headline number needs a date and a definition. The $1.9B of contracted TCV is a 7/31 figure in a Q2 release, and the deck's own backlog chart puts contracted revenue at $440M on June 30.

The appendix defines TCV as remaining performance obligations (RPO) plus executed LOIs, and says it includes renewal options and LOI volume that aren't performance obligations yet.

The deck's deployment schedule puts $0.4B of the $1.9B in production as of Q2:

So $1.9B is not $1.9B of signed, non-cancellable revenue. And there's no RPO to check it against, because Boost Run books GPU rentals as operating leases under ASC 842 rather than as ASC 606 revenue.

Lastly on the quarter itself, the $(75.0)M net loss reads worse than it was. $55.7M of it is deferred tax from the LLC-to-C-corp conversion at closing, which is non-cash and doesn't recur.

Nebius Comparison

Some investors like to call Boost Run the next Nebius (NBIS). It's a lazy comparison, and CEO Andrew Karos explained on the Q2 call why he builds capacity the opposite way on purpose.

Given the choice between one 100MW site and four sites at 25MW each, Karos said he takes the four every time:

Set that against Nebius. It owns more than 75% of its contracted power, guides to $20-25B of 2026 capex, and is building 1.2GW in Missouri, which comes online in 2027.

Boost Run owns no data centers at all. It guides to $1.0-1.4B of 2026 capex and goes from 6 sites to 9 over the next 6 months.

Nebius is buying land, power, buildings and GPUs. Boost Run is buying GPUs.

But Boost Run still pays for the building. The cost just shows up as rent instead of capex, and colocation lease cost was $15.2M in the first half against $42.1M of revenue.

So the trade isn't cheaper capacity. Boost Run gets speed, parallel deployment, and no multi-year construction risk, and gives up control of its own ceiling.

And Boost Run says the ceiling part itself, in the S-1 risk factors, more bluntly than any bear would:

Management frames colocation as an advantage on the call. The S-1 calls the same dependency "significant limitations." Both are true, and the Q1 10-Q shows the cost.

In that quarter alone Boost Run signed 2 colocation leases with fixed undiscounted payments of $6.3M over 3 years and $113.8M over 7 years, plus $7.2M of prepayments and a $6.4M standby letter of credit against the 7-year one.

The same filing's lessor note shows what customers owe Boost Run—almost nothing. As of March 31, they had no obligation to pay anything beyond the $11.5M already on the balance sheet. Customers prepay, so the money isn't missing, but nothing forces them to stay past what they've already paid for.

Q2 moved that in Boost Run's favor. Customer deposits went from $41.1M at March 31 to $128.4M at June 30, so a lot more prepaid revenue backs those leases now.

But the mismatch is still there. The rent is contracted for 7 years, and the customer revenue is contracted only as far as the cash already collected. The Q2 10-Q's lessor note is where the $1.9B of TCV either becomes future minimum lease payments or doesn't.

That's what I'd push back on with the mini Nebius framing. Nebius owns >75% of its power and rents the rest. Boost Run rents all of it and owns only the GPUs.

NVIDIA Credential

The NVIDIA relationship also finally got specifics on the call. NVIDIA's Exemplar Cloud status is a validation that a provider's GPU clusters perform to spec, and Karos said that when Boost Run obtained it for the Blackwell B300, one other cloud had the same status:

In my June write-up I called the NVIDIA relationship "a credibility marker on the deployment stack," with nothing concrete underneath the phrase.

Karos finally gave the specifics on what Exemplar requires:

Karos says the founders he talks to have "some serious frustration about the quality of the compute that's being handed off." A cluster that benchmarks well on paper and underdelivers in production is the industry's quiet problem, and a validated handoff is what a customer is actually buying.

Boost Run's June customer slide carries the strongest evidence. Nebius appears on the logo list, a company with a $75B enterprise value, building 1.2GW in Missouri, buying compute from one roughly a fortieth its size.

Competitors don't rent from you unless you can deliver something they can't, or can't deliver fast enough.

Two caveats though:

Valuation

At Friday's $22.93/share close, Boost Run is a ~$2.0B enterprise value, on ~85.8M shares assuming the last ~5M warrants exercise before the August 20 redemption, against ~$192M of pro forma cash and $238.1M of finance leases.

That's 5.0x the $400M target and 13.9x the $145M ARR they have today. In June at $40/share I had it at 7.3x the target, and I said the mid-$20s was the downside if deployment stalled.

But deployment didn't stall. And the gap to plan narrowed rather than widened, from a 34% miss in Q1 to 13% in Q2.

I've covered DigiPowerX (Nasdaq: DGXX) since the stock was trading at $3.50/share back in April (before the Cerebras (CBRS) deal).

Financing Answer

Two weeks ago I mentioned that you should be watching which financing route DigiPowerX takes next, debt or more ATM.

The Q2 filings answer it—shareholders have already paid for Phase 1.

DigiPowerX has deployed ~$110M of capex into the Columbiana build with no long-term debt on the balance sheet. All of that came from equity. The ATM has issued 52,028,450 shares for $254,758,203, at an average of $4.90/share.

The 10-Q also discloses that after June 30, DigiPowerX issued another 2,850,000 shares for $11,256,693. That's $3.95/share, against Friday's $3.97 close.

Here's what the share count has done:

On the call, Amar was asked what he's doing to limit dilution. He pointed at the Q2 raise, saying most of the ATM capital was taken early in Q2 at prices well above where the stock trades now:

DigiPowerX closed between $7 and $8.50 from mid-May into early June, and the H1 draws averaged $5.84 across 27,950,000 shares.

But the draws since June 30 went out at $3.95/share, with the stock under $5.50 the entire time, and they're disclosed in the 10-Q filed the same morning as the call.

So the debt is not what gets Phase 1 built. Amar says the cash on hand does: "we accumulated enough cash to... self-fund most of the data center of Alabama, so we are not at risk of execution from a financial perspective."

What the debt would do is return the money already spent: "We already CapExed over $110 million. So we are going to get cash back."

DigiPowerX has engaged Goldman to arrange the loan, and nothing is signed yet.

If it closes, this year's equity was a bridge and DigiPowerX ends up holding both the data center and the cash. If it doesn't, shareholders paid for Phase 1 outright and Phase 2 needs another round of equity.

Currency Gain

The other headline was positive adjusted EBITDA of $3.3M. But it's only positive because of a currency gain.

The reconciliation walks a $(14.4)M net loss up to positive $3.3M by adding back depreciation, $5.8M of stock comp, a $2.8M crypto revaluation loss, and a $5.0M warrant fair-value loss (all non-cash):

The $7.7M foreign exchange gain is also non-cash, and it never gets added back. Take the gain out and adjusted EBITDA is negative $4.4M.

The cash flow statement does take it out. In the operating section, DigiPowerX reverses $10.7M of currency gain across the first 6 months to reach operating cash.

The 10-Q also says where the gain came from: "currency exchange fluctuations on the intercompany balances." Money owed between the company and its own subsidiaries, revalued as the exchange rate moved.

Phase 1 comes online in December, and adjusted EBITDA is how investors will track the ramp from there. This quarter it turns positive on an intercompany balance rather than on operations.

What's in the $1.1B

Then there's the number DigiPowerX has repeated since May: $1.1B of contracted AI data center revenue. The Q2 10-Q gives the contract terms, and the terms show how much of that is actually contracted.

The Cerebras agreement runs 10 years at a fixed, take-or-pay colocation fee of $195.00 per kW per month, rising 3% a year. It covers Phase 1 at 15MW and, in the filing's words, "a conditional Phase 2 (25 MW)."

Price the phases separately, with the 3% annual increase applied:

The $1.1B is the sum of the two. Phase 1 is 37% of the headline. The conditional phase is the other 63%.

The condition is financing. From the MD&A:

Phase 2 carries $58.5M in its first year, take-or-pay, against a Columbiana build that has absorbed ~$110M so far. Those are the economics a lender is underwriting.

The 2027 target works the same way. DigiPowerX targets $250-300M of ARR by Q3 2027, of which ~$140M is contracted. The rest comes from an additional 40MW of colocation and 10MW of GPU capacity, which the 8-K puts "subject to customer contracting, deployment schedules and utilization."

Deployed AI compute capacity today is just 0.6MW.

Phase 1 is still targeted at December 15, 2026, Phase 2 at the end of Q1 2027, and management says the long-lead equipment is ordered and starting to arrive.

But the version of DigiPowerX that collects $1.1B is the version that closes project debt.

The Scorecard

Lastly, a scorecard on my own call. I wrote DigiPowerX up on April 29 at $3.50/share and named two catalysts. One landed in a week. The other has not landed at all.

The first was a second and third customer contract inside 60-90 days. Five days later DigiPowerX signed Cerebras, a 10-year colocation agreement at a scale I was not modeling.

It goes without saying that catalyst hit!

The second was operational ramp toward ~30MW by Q3 2026, following management's 5/15/30/55 quarterly deployment plan.

That roadmap is still in the Q2 10-Q, unchanged: 5MW in Q1 2026, 15MW in Q2, 30MW in Q3, and 55MW by Q4 with 40MW of critical load.

But the same filing says the Alabama facility "remained under construction and not ready for use" as of June 30, and it targets Phase 1's 15MW for December 15 with the full 40MW by the end of Q1 2027.

So there was nothing to verify each quarter. It all lands in one window: December to March.

Dilution ran ahead of my model too. I forecasted 84.7M shares by the end of FY2027 and 89.8M by FY2028. DigiPowerX closed June at 98,543,358 and reports 101,393,355 outstanding as of August 14.

Here's what that did to the return:

The business got 67% more valuable. But the stock only returned ~13%, partly because that value is spread across 45% more shares.

So two inputs in my April model are stale, how fast the megawatts come on and how many shares are outstanding.

Disclaimer: This write-up is for informational and educational purposes only and is not investment advice. I may hold positions in the securities discussed. Do your own research before buying or selling any security.