The Earnings Audit

Big tech is about to report. What will this season's earnings will say about the AI trade?

THE EARNINGS AUDIT

The buildout still needs the money. The money wants proof.

The AI infrastructure trade enters earnings season looking considerably less inevitable than it did two months ago. Semiconductors have surrendered double digits from their June peak, while memory has endured an even harder reversal after nearly doubling in the second quarter.

Broad AI and technology baskets remain below their highs. Leadership has narrowed, capital has become more selective, and companies once treated as interchangeable pieces of the same trade no longer move together.

This looks like deterioration, but it does not yet look like the end. The longer advance remains intact across most of the infrastructure complex, while the strongest suppliers still show healthy fundamentals and resilient balance sheets.

Memory has begun responding after its sharp correction. Capital is also moving back toward parts of the power and physical-infrastructure trade before price leadership has fully recovered.

The evidence still fits a correction within an intact Stage 2 advance. This earnings season is where that interpretation gets audited.

Alphabet reports first. Microsoft and Meta follow alongside the Federal Reserve’s rate decision, with Apple and Amazon closing the cluster. Nvidia will not report until August, but the hyperscalers will establish the premise before the market gets the punchline.

The question is no longer whether AI spending is real. The question is whether the spending is working.

What the Market Has Already Decided

The trade is splitting into three groups. The first contains companies capable of funding the buildout largely from their own cash flow.

The second contains businesses tied to real physical bottlenecks—power, cooling, construction, optical connectivity, and eventually copper—but where market confirmation remains uneven.

The third contains companies whose business models depend on debt markets remaining generous.

Their earnings results will not carry the same meaning. Weakness in the first may still be an opportunity, weakness in the second may expose an unproven thesis, and weakness in the third may close the financing window entirely.

Tier One: Self-Funded and Confirmed

This remains the strongest part of the trade: leading semiconductor suppliers, memory producers, interconnect companies, select construction firms, and cloud platforms with enough cash flow to absorb their own ambition. The correction here has been severe in places, but it has not been accompanied by a broad deterioration in operating quality.

Several of the strongest suppliers now trade well below their highs while revenue growth, margins, order books, and balance sheets remain healthy. Memory has given back a substantial portion of its parabolic second-quarter move, but the underlying demand cycle has not yet rolled over.

Some of the most important compute and connectivity companies show the same pattern. Price has weakened faster than the business, which is what a correction looks like, not a degradation. Earnings could help put the pin in the definition.

Expected

The hyperscalers raise or reaffirm capital spending, cloud demand remains strong, and management teams repeat that AI capacity is constrained. None of that would be new.

Another capex increase without evidence of returns would confirm only that the bill is getting larger. The market now needs evidence that the spending is producing something beyond additional capacity. Watch for any guidance or data that supports the play is paying and be wary of exactly how the play is being paid.

Upside

The companies begin showing where the returns are coming from. Cloud growth accelerates, backlog converts into revenue, and AI products produce measurable demand rather than broad claims about engagement.

Margins hold and free cash flow begins stabilizing despite the construction burden. That would support the Stage 2 correction thesis by showing that the market has been repricing the pace of the buildout rather than rejecting its economics.

Downside

A modest cloud miss or temporary margin compression would be disappointing but probably survivable. A capex reduction would be different because it would directly weaken expected demand across chips, memory, networking, cooling, construction, and power equipment. This would be a real check on the thesis of an intact Stage 2 correction play.

Also watch how much of the spending increase is funded through new debt or equity issuance. Rising capex supported by operating cash flow strengthens the thesis. Rising capex increasingly financed through borrowing or shareholder dilution suggests the buildout is becoming less self-sustaining and materially narrows the distinction between Tier One and the weaker parts of the trade.

The downside signal is therefore not limited to a capex cut. It also appears when spending continues, but the balance sheet or share count is absorbing more of the burden.

The Read

Ordinary weakness remains a buying opportunity where operating quality has held and capex is still being funded primarily through internal cash generation.

A capex cut would weaken the demand thesis. A material shift toward debt or equity issuance would weaken the self-funded thesis. Either would argue against treating the selloff as a routine Stage 2 correction and should be points of caution.

Tier Two: Real but Unconfirmed

The physical-bottleneck thesis remains persuasive, but the market has not confirmed it broadly. Price leadership is uneven, capital flows are selective, and many companies tied to power, cooling, construction, and connectivity still lack the combination of strength and durability seen in Tier One.

Power availability is already restricting where and how quickly data centers can be built, but the stocks positioned to benefit have not moved together. Some leading resource providers remain clear leaders in their fields, while other leaders in other essential fields have weakened despite carrying the same demand narrative.

Transformer, turbine, cooling, and electrical-equipment lead times remain extended. Optical connectivity must scale with compute density, construction capacity is finite, and copper demand should eventually rise with all of it.

But eventually is doing too much work in parts of this trade. Power and physical-infrastructure stocks have not moved together, with some remaining clear leaders while others have lost price strength despite retaining strong demand narratives.

Copper continues to trade primarily as a macroeconomic industrial input rather than as a direct measure of AI construction. The story may be correct, but the market has not confirmed all of it.

The strongest evidence beneath the surface is that capital has begun returning to the AI-and-power infrastructure complex before broad price leadership has recovered. That is constructive, but it is not conclusive.

Expected

The hyperscalers continue discussing power procurement, cooling requirements, construction schedules, and equipment lead times. More importantly, announced projects continue moving through permitting, utility interconnection, equipment ordering, and construction without a material rise in cancellations or delays.

A non-material change in planned data centers would preserve the thesis but not strengthen it. The market already understands that bottlenecks exist and it’s baked into the prices already. The baseline result needs to be evidence that planned capacity is still becoming buildable.

Upside

Announced projects are converting into physical demand faster than expected. Permits are approved, utility interconnections are secured, power contracts are signed, equipment orders expand, and more sites move from planning into active construction.

These announcements would strengthen the thesis by showing that the bottlenecks are only delaying the buildout, not preventing it. They are institutionalizing immediate demand for power equipment, cooling systems, construction capacity, and connectivity infrastructure that will translate to future earnings.

Downside

Any announcements that projects remain planned but are failing to become buildable due to things like community resistance, zoning restrictions, litigation, utility delays, financing problems, or grid constraints indicate a push of construction further out or possible forced cancellations.

Any flags that delays construction weaken the near-term thesis, even if long-term demand remains intact, and will deepen the current correction. Scarcity only supports Tier 2 when planned capacity continues converting into orders, groundbreakings, and revenue.

The Read

Tier two investments remain a conditional opportunity. The thesis strengthens only when announced capacity converts into permits, power commitments, equipment orders, and active construction.

Routine progress preserves the setup. Faster conversion supports a rerating. Rising delays, resistance, or cancellations argue that the bottleneck is postponing demand rather than creating near-term revenue.

Tier Three: Debt-Funded and Breaking

The weakest part of the trade is not defined by technology. It is defined by financing.

Oracle is committing substantially more to capital spending than its operating cash flow can support, financing the difference through an extraordinary combination of debt and equity issuance. The strategy may still work, but its recent credit downgrade shows that lenders are beginning to question how much additional strain the balance sheet can absorb.

The neo-clouds face the same problem from a weaker starting position. Their growth depends on continuously financing GPUs, data centers, and long-term supply commitments before the associated customer revenue fully arrives.

The financing window remains open, but the terms are becoming part of the verdict. Capital increasingly arrives through high-yield debt, asset-backed loans, convertible securities, or equity issuance. Each structure either raises the required return, pledges more of the underlying business, or dilutes existing shareholders.

The market has already punished the equities. The next risk is not simply that financing disappears, but that it remains available only on terms that consume too much of the economics the buildout was supposed to produce.

Expected

Oracle and the neo-clouds are expected to continue reporting rapid revenue growth, large backlogs, high utilization, and strong demand for additional capacity. Those figures should be impressive, but they largely confirm what the market already knows. They need proof of that narrative to hold their side of the thesis bargain intact.

The expected tension is that capex, interest expense, and external financing remain elevated alongside that growth. The important question is whether revenue and operating cash flow are beginning to catch up with the cost of building the capacity, or whether each new contract still requires another round of debt, structured financing, or shareholder dilution.

Upside

An improved outlook would be where revenue and operating cash flow begin catching up with the cost of expansion. Backlog converts faster, utilization stays high, margins improve, and new capacity produces enough cash to reduce dependence on external financing.

The strongest signal would be continued growth alongside lower financing needs, cheaper borrowing, or less dilution than expected. That would suggest Oracle and the neo-clouds are moving toward self-sustaining scale rather than simply financing their way to a larger balance sheet.

Downside

A cautious outlook would be where revenue growth slows, backlog conversion slips, or utilization falls short while capital spending and interest expense remain high. The most important warning would be operating cash flow failing to improve even as capacity comes online, suggesting that the economics are not scaling with the balance sheet.

Also watch whether Oracle and the neo-clouds need more debt, structured financing, or equity issuance than expected to complete existing commitments. Higher borrowing costs, tighter collateral requirements, or heavier dilution would show that financing is consuming more of the value created by the buildout.

However, the most damaging outcome would not be weak demand alone. It would be strong reported demand that still cannot produce enough cash to fund the next stage without repeatedly returning to the capital markets.

The Read

Tier Three remains a financing trade disguised as a growth trade. Strong revenue and backlog matter only if they begin reducing dependence on external capital.

If growth starts funding expansion, the category can improve. If debt, structured financing, and dilution keep rising faster than operating cash flow, rallies should be treated as temporary and the underlying thesis as deteriorating.

Proof, Not Promises

The market has already shown how it intends to judge these reports. Earlier in the cycle, Alphabet reported accelerating cloud revenue by 63% and rapidly expanding backlog to roughly $462 billion. The stock rose because this is what direct monetization through cloud growth looks like.

Meanwhile, Meta delivered strong headline growth of 41% but increased capex guidance without offering any convincing evidence that its rising AI expenditure was producing an adequate return. Naturally, its stock fell.

The difference was not the size of the capital budget. It was proof. That is the standard entering this earnings cluster. Spending may establish commitment, but returns establish value.

What Would Confirm the Trade

The Stage 2 correction thesis survives if the strongest operators keep spending while funding most of that expansion through internal cash generation. Revenue, margins, and free cash flow do not need to fully catch up this quarter, but they must show that the economics are moving in the right direction.

The physical-infrastructure thesis strengthens if announced data-center capacity continues converting into permits, power commitments, equipment orders, and active construction. Delays are tolerable; widespread cancellations or projects that remain permanently stuck in planning are not.

The leveraged tier improves only if revenue and operating cash flow begin growing faster than financing needs. Strong backlog alone is insufficient if each new stage of expansion still requires more expensive debt, structured financing, or shareholder dilution.

The trade weakens if capex is cut, Tier One becomes materially more dependent on external capital, planned data centers fail to become buildable, or financing costs begin consuming the returns promised by the buildout. One disappointing quarter would not end the secular trade, but evidence that spending, construction, or financing can no longer sustain the next stage would.

The Setup

The AI infrastructure trade has weakened, but it has not yet broken. The current evidence is more consistent with a Stage 2 correction than a secular decline. Price leadership has narrowed, enthusiasm has cooled, and weaker financing models are being separated from stronger ones. Underlying demand, growth, and physical constraints remain intact. Arguably, this puts the trade in a good buying opportunity.

Earnings now need to prove that spending will continue, infrastructure is producing revenue, and the strongest companies can fund the next leg without becoming the weakest ones. Capex alone will not be enough.

This time, the market wants the receipt. The companies that cannot provide it will not be treated as innocent until proven guilty anymore.

Avalanche Markets — The Setup | Issue 002

Jason Bartlett

Jason Bartlett is CEO and President of Veche, Inc, parent company to Avalanche Markets. He works extensively in U.S. energy market finance and economics and is a member of the board of Thinking About Thinking, Inc--a 501c3 research and convening organization dedicated to advancing ideas about intelligence.