Anthropic Q3 2026 Results: Why 47% Revenue Growth Still Led to an Operating Loss

Anthropic’s quarterly revenue reached nearly $17 billion in Q3 2026, up 46.7% from Q2. Yet the company behind Claude returned to a reported operating loss of $0.567 billion, after earning a $0.974 billion operating profit three months earlier. Its adjusted operating profit also increased.

The explanation is that operating expenses grew faster than revenue, while the adjusted result excludes stock-based pay and a large non-cash donation-matching charge. Expensive short-term compute weighed on both measures, but the donation charge was removed from adjusted earnings.

Anthropic therefore earned more adjusted profit in dollars while retaining less adjusted profit from each dollar of revenue.

Anthropic Q3 Results: What Changed From Q2 to Cause an Operating Loss?

Source: Anthropic’s November 2, 2026 S-1/A. All figures are company-reported. Adjusted measures are non-GAAP, meaning they exclude specified expenses from the reported accounting result.

Revenue increased by $5.401 billion. Total operating expenses rose from $10.581 billion to $17.523 billion, an increase of $6.942 billion. The extra expenses exceeded the extra revenue by $1.541 billion, exactly matching the deterioration from Q2’s operating profit to Q3’s operating loss.

Compute was the largest expense increase, rising by $4.305 billion. Anthropic identifies $5.0 billion of Q3 compute under short-term commitments and approximately $2.0 billion of non-cash donation-matching charges as major pressures on the quarter. Personnel expense declined, partly offsetting increases in sales, marketing, partnerships and other operating and administrative expenses.

Why Did Anthropic’s Compute Costs Rise So Quickly?

Think of a business booking space well in advance, then needing extra space before its reservation becomes available. Renting at short notice can be more expensive, even when the advance booking was sensible. The gap between when demand arrives and when reserved supply becomes usable is the relevant comparison for Anthropic.

Anthropic says it typically procures long-term compute 12-18 months before availability. The associated expenses are generally fixed in the near term and increase as committed capacity becomes available. That creates a timing risk: customer demand can arrive before the longer-term capacity intended to serve it.

Q3 included $5 billion of compute expense under short-term commitments. Management says this supply currently costs significantly more than comparable compute obtained under multi-year arrangements. The $5.0 billion is the expense for that supply, not a quantified surcharge that can simply be removed to estimate a cheaper quarter; the filing does not disclose the comparable long-term cost.

Total compute and infrastructure expense grew 77.5%, faster than revenue’s 46.7% increase. Compute absorbed approximately 58.2% of revenue in Q3, up from 48.1% in Q2, an increase of about 10.1 percentage points. A larger portion of sales was therefore consumed by this expense before paying the rest of the operating cost base.

Management’s plan is to replace expensive short-term supply with more cost-effective long-term compute as dedicated infrastructure comes online. Longer commitments can improve availability and pricing, but they also require capacity to arrive on time, attract enough use and remain competitive as chips and models change. The broader risks of reserving AI computing capacity extend beyond this quarter; replacing the expensive supply is a plan whose financial benefit still needs to be demonstrated.

There is also an accounting limit to the comparison. Total compute includes model development, research and internal workloads as well as serving customers, so dividing it by revenue produces an operating-cost ratio, not a gross margin. Gross margin uses the expenses classified as cost of revenue, a narrower grouping explained in our Anthropic gross-margin analysis.

How Did Adjusted Profit Rise While Adjusted Operating Margin Fell?

Anthropic’s adjusted operating profit starts with the reported operating result and adds back specified expenses. In Q3, those exclusions were stock-based compensation and non-cash charitable contributions. The reconciliation is straightforward.

Source: Anthropic’s November 2, 2026 S-1/A. These are company-reported non-GAAP adjustments to an unaudited quarterly operating result.

Q2 had no charitable adjustment in this reconciliation. The large Q3 donation addback helps explain how the adjusted result could improve while the reported result deteriorated, even though the stock-compensation addback became smaller. The full compute expense remains deducted in adjusted operating profit.

The adjustments help compare operations before these identified charges, but the excluded costs still deserve attention. Stock-based pay compensates employees with equity interests and can dilute existing shareholders, reducing their percentage ownership. I would read the adjusted result alongside the reported accounts when assessing the cost of growing the business.

The non-cash donation-matching charge sits in other operating and administrative expenses and is linked to employee equity awards. The programme permits matching in cash or shares. Note 17 of the original October 26 S-1 had already disclosed the Q3 charge and described shares issued in October to settle the obligation. Settling an obligation with shares can still dilute ownership, so a non-cash charge should not be treated as economically free.

The margin question is separate from the addbacks. Adjusted operating profit rose by $0.228 billion, or 10.2%, while revenue grew 46.7%. Because profit grew much more slowly than sales, adjusted operating margin fell from 19% to 15%, using the filing’s rounded figures.

Put simply, approximately $19 of adjusted operating profit remained from each $100 of revenue in Q2, compared with about $15 in Q3. The company earned more adjusted profit overall because the business was much larger, while earning a smaller adjusted profit share of its sales. This measure is adjusted operating income, not cash generated by the business.

What Did Price Cuts Do to Anthropic’s Revenue?

Anthropic says higher usage increased quarterly revenue, but price reductions introduced with new model versions and contractual discounts partly offset that increase by approximately $0.600 billion between Q2 and Q3. This is a revenue offset in management’s explanation of growth. It is not a separate cash expense.

For usage-priced work, a customer can ask Claude to process more information while paying less for a given amount of processing. Revenue can still rise if additional paid usage outweighs the pricing reduction. Anthropic’s disclosed quarter shows that overall usage growth more than offset the reported pricing pressure.

It does not show how much of that usage growth was caused by lower prices. Model selection, workloads and contract terms also affect the revenue generated, and the filing does not isolate a constant mix for this comparison. The $0.600 billion cannot establish a company-wide average price-cut percentage or the profit impact of those changes.

I would read the disclosure as evidence that stronger adoption and pressure on realised pricing can coexist. The economic question is whether useful paid work expands quickly enough, and can be delivered efficiently enough, to support earnings as pricing changes. Q3’s revenue growth answers the sales question more favourably than its falling adjusted margin answers the profitability question.

What Would Show That Anthropic’s Next Quarter Is Economically Better?

My assessment is mixed. Anthropic generated substantially more revenue and more adjusted operating profit, while compute took a larger share of sales. The donation charge made the reported result weaker, but removing it still leaves a decline in adjusted operating margin.

First, I would look for evidence, where disclosed, that new long-term capacity actually replaces expensive short-term supply. Bringing more infrastructure online can initially increase expenses, even if its contractual pricing is better. The benefit depends on how much costly supply it displaces and how productively the delivered capacity is used.

Next, I would compare compute growth with revenue growth over matching quarters and on a consistent accounting basis. A decline from Q3’s calculated 58.2% compute-to-revenue ratio would show that this expense is absorbing less of sales. It would still need context about training activity and workload mix before I inferred better economics for an individual customer task.

I would also check the adjusted operating margin against the full reconciliation. A return to reported operating profit caused mainly by a smaller donation charge would be different from a recovery in adjusted margin. An improving adjusted margin accompanied by larger exclusions would require closer attention to the reported result and the economic cost of those exclusions.

Finally, I would examine cash collections and operating payments alongside infrastructure investment, including purchases of equipment and prepayments for leased assets. Operating cash flow can be affected by customer and supplier payment timing, and it does not deduct all investment spending. Anthropic says its infrastructure build-out will use significant cash, so a rising adjusted profit alone cannot show whether the capacity expansion is supporting itself.

My test for the next results is specific: revenue should outgrow compute expense, adjusted operating margin should recover without relying on larger exclusions, and customer collections should support more of the infrastructure being added. Those would be signs that Anthropic is turning its larger business into stronger economics. Another large revenue increase, by itself, would leave the central Q3 cost question unresolved.