On a clear night, the data came back beautiful. A major AI chipmaker reported second-quarter revenue of $96.2 billion, up 106% year on year, with data-center revenue of $89 billion — up 117% and equal to 92% of total revenue. The long view is the only honest view, and over the last two years the long view has been: every quarter the skeptic’s argument gets harder to hold. This quarter, it got harder still.
The Numbers That Answer the Capex Question
The recurring question has been whether AI capital expenditure is peaking — whether the hyperscalers would finally stop buying chips and let the bubble deflate. The data this quarter does not settle the debate forever, but it answers the immediate form of it. Revenue up 106%, data center up 117%, and the compute share of the business at 92%. When the segment everyone predicted would slow is growing faster than the company as a whole, the peak-capex thesis has to wait at least another quarter.
Let me think about what is genuinely new here. It is not the growth rate — those have been enormous for several quarters. It is the guidance: a revenue target for the next fiscal year up roughly 70%, plus a commitment to supply two million additional GPUs to one major cloud provider over the next two years. The forward-looking signal is as strong as the backward-looking one.
Deep Time Settles Arguments, Quarter by Quarter
Deep time has a way of settling arguments. A year ago the claim was that AI compute demand was a temporary spike. The measured wonder here is that the spike has lasted long enough to look structural. Two years is not deep time, but for a capital cycle it is long enough to separate a fad from a build-out. The hyperscalers are not buying because they like the product; they are buying because their own customers’ demand is real — and the numbers in this sheet are the echo of that demand.
The Part Worth Watching, Not Cheering
Evidence keeps the awe honest, so let me hold the awe at arm’s length. A 92% concentration in one segment is exactly the kind of number that flatters a bull case and hides a concentration risk. The company is now, for practical purposes, a data-center chip company with a side business. That concentration is a strength until the moment demand shifts — and nothing in the sheet says when that moment comes. The long view cuts both ways: it has validated the build-out, and it will eventually date it.
To be honest, I expected the peak-capex debate to outlast this report. The guidance to keep growing at 70% is the part that genuinely surprised me — it pushed the horizon out, not the cliff closer.
The 106% Line, Weighed
Weigh the 106% revenue line before letting it awe you. A doubling of revenue in a year is, at face value, the kind of number that makes headlines and sets expectations. Held against the long view, it asks to be understood rather than admired: the growth is not evenly distributed — it is concentrated in one segment, data centers, at 92% of the business. The measured wonder here is not the growth itself, which the market has seen for several quarters; it is the consistency. When the same line grows by triple digits quarter after quarter, the word “bubble” has to fight with the word “trend”, and the trend is winning the argument so far.
The long view cuts both ways: it validates the build-out, and it will eventually date it. This quarter, the validation is what the data shows.
The 92% Concentration, Held
Hold the 92% number up to the light, because it is both the strength and the risk of the whole story. A company with 92% of revenue in one segment is, for practical purposes, a data-center chip company with a side business. That concentration means the bull case is simple and powerful: as long as AI data-center demand grows, the company grows with it. It also means the bear case is equally simple: the day that demand shifts, there is no diversified cushion. The concentration is a wager on one segment, and the wager is large.
Evidence keeps the awe honest: concentration is not a flaw this quarter — it is the source of the extraordinary numbers. But it is a risk being accumulated, and the long view has a way of cashing in risks on its own schedule.
The 70% Guidance, Checked Against History
Check the 70% forward guidance against the historical tape, because guidance is a statement with a deadline. A company that guides to roughly 70% growth for the next fiscal year is saying the build-out continues, and it is saying it with numbers rather than adjectives. Historical precedent matters here: companies that guided to triple-digit growth and delivered it for multiple years earned the market’s belief; those that guided high and missed lost it instantly. The guidance does not settle the future; it sets a target that the next four quarters will test.
The measured wonder is that the horizon keeps being pushed out. A year ago the question was whether demand would last one more quarter; now the guidance extends it a full fiscal year ahead. Deep time is not arriving all at once — it is arriving in scheduled installments, and each installment resets the debate.
The 2 Million GPU Commitment
Look at the commitment to supply two million additional GPUs to one major cloud provider over the next two years, because it is the most concrete number in the sheet. Two million units is not a forecast; it is a production commitment, with silicon, packaging, and power behind it. That is the kind of number that anchors the whole capex debate: a supplier does not commit to that volume for a customer whose demand is fictional. The commitment is the strongest evidence in the report that the build-out is being scheduled, not guessed.
The long view reads the commitment as an echo: the cloud provider’s customers are demanding compute, and the demand is working its way up the chain into the supplier’s order book. The sheet is the trace of that echo, and it is loud.
The Echo of Demand
Trace the echo and the numbers come alive. The cloud provider that committed to two million GPUs does not buy compute for itself; it buys for its customers — the enterprises and developers building AI products. Those customers are the first link in the chain, and their demand is what makes the supplier’s guidance defensible. When the echo reaches the chipmaker’s order book at a scale of two million units, it means the first link is real. The sheet is the supplier’s view of a demand curve that starts far away and is measured in generations of products.
The long view reads the echo as structure, not hype: the demand is not a bet on any single product, it is a bet on computation itself, and computation has been compounding for decades. The numbers in this sheet are the most recent installment of that compound.
What the Skeptics Still Hold
Give the skeptics their due, because the long view is only honest when both sides are weighed. They hold that capex cycles always overshoot, that the concentration at 92% is a single point of failure, that forward guidance is a persuasion device as much as a forecast, and that a demand echo can be amplified into a bubble by the very commitments meant to prove it real. Each of these has historical precedent, and none can be dismissed by one earnings sheet. The peak-capex thesis is not dead; it is temporarily short of evidence.
The measured wonder is that the evidence keeps arriving. Quarter after quarter, the skeptic’s argument has to wait one more quarter. That is not a refutation — it is a postponement, and postponements are how deep time works.
The Capex Question, Re-Asked
Re-ask the question the way the long view demands: is this a capex cycle or a capital build? A capex cycle is spending on assets that depreciate and must be replaced; a capital build is the construction of infrastructure that compounds. Data centers, power, chips, and the software layer on top of them — built once, amortized over decades, then extended. If the current spending is the latter, the peak-capex thesis misreads the phenomenon entirely: the spending is not a spike, it is the foundation of a new capital stock. The sheet cannot settle which reading is right, but it can show that the spending is being scheduled, committed, and extended — the behavior of a build, not a binge.
That is the honest distinction at the center of the debate, and this quarter’s data points one way. The long view will have the final word, but it is accumulating evidence.
The Long View, Final Pass
Make the final pass with the sky still clear. Revenue up 106%, data centers at 92% of the business, forward guidance up 70%, two million GPUs committed to a single cloud partner, and the demand echo running from the cloud provider’s customers back to the fab. The peak-capex argument has lost another quarter of evidence, the concentration risk is real and noted, and the build-out is being scheduled years ahead. The long view is the only honest view: the numbers say this is a capital build in progress, and the honest reading is that the build is not slowing. Deep time will settle the rest — but not this quarter, and not the next.
The Long View, Adjusted
Adjust the long view with the discipline the data demands. Revenue up 106%, data center at 92%, forward guidance up 70%, and a two-million-unit commitment to a single cloud partner. The peak-capex thesis does not die here — but it just lost another quarter of evidence, and the forward commitments push the horizon further out. The concentration is real and should be watched; the growth is real and should be believed until the evidence says otherwise. The long view is the only honest view, and the honest reading is that the build-out is not slowing — it is being scheduled years ahead. Deep time will settle the rest — but not this quarter.
The Long View Is the Only Honest View
So where does this land? Revenue up 106%, data center at 92% of the business, forward guidance up 70%, and two million more GPUs committed to a single cloud partner. The ‘AI capex peaked’ argument does not die here — but it just lost another quarter of evidence. The long view is the only honest view, and the honest reading of this sheet is that the build-out is not slowing; it is being scheduled years ahead. Deep time will settle the rest — but not this quarter.