Anthropic amends its S-1: $17B in Q3 revenue, $895B in compute commitments, and a return to loss

Anthropic amends its S-1: $17B in Q3 revenue, $895B in compute commitments, and a return to loss

🔎 The last financial snapshot before Wall Street

On November 2, 2026, Anthropic filed Amendment No. 1 to its S-1 with the SEC, one week after the initial document that made its Nasdaq IPO official under the ticker ANT. It's the last major financial statement before the roadshow — and it looks like nothing Wall Street has read in the past ten years.

On the agenda: roughly $16.96B in third-quarter revenue (+1,127% year-over-year), a return to operating loss (−$567M after +$974M in profit in Q2), a net loss of $23.29B, and above all $895.4B in compute purchase and leasing commitments. No company, public or private, has ever signed on to a liability of this size.

But the most meaningful figure is neither the +1,127% nor the $23B. It's the deceleration: +47% sequential growth in Q3, versus +144% in Q2. That's what investors are going to price in.

The Essentials

• Q3 2026 revenue: ≈ $16.96 B, up +1,127% year over year — but only +47% from Q2 to Q3, versus +144% the previous quarter.
• Return to operating loss: −$567 M in Q3, after +$974 M in profit in Q2. Main cause: $5 B in short-term compute that is "significantly more expensive", plus ~$2 B in non-cash charges.
• Net loss: $23.29 B, driven by the revaluation of convertible notes — an accounting effect, not a cash outflow.
• $895.4 B in purchase and lease commitments, "substantially all" tied to compute: $447.5 B before 2030, $69.8 B within 12 months, with contracts stretching out to 2045.
• Liquidity: $146.6 B, against liabilities representing nearly 27 times the revenue of the first nine months of the year.
• 5 to 7% of revenue for the nine months comes from customers in violation of the terms of use.
• Unprecedented risk factor: future models could "imminently" reach the RSP threshold for "dramatically accelerated progress attributable to AI."

Recommended Tools

A word of advice: don't rely on any summary, including this one. The S-1/A is the reference document, and its risk factors contain phrases that most news reports don't mention.

Q3 figures: +1,127% year-over-year, +47% quarter-over-quarter

Anthropic's Q3 2026 revenue comes to roughly $16.96B. Year-over-year, that's more than a twelvefold increase. Quarter-over-quarter, "only" +47%.

The +1,127% is first and foremost a base effect. In Q3 2025, the company was bringing in about $1.4B per quarter. That percentage measures the success of the past year, not today's momentum.

Today's momentum is the right-hand column: +47% from Q2 to Q3, following +144% the previous quarter. Over nine months, Anthropic reports more than $33B in revenue — $16.29B in the first half (+1,300% year-over-year) according to the initial filing, plus Q3 on top of that.

*Rounded figure, derived from the growth rates published in the S-1/A.

Sources: CNBC, Forbes, TrendingTopics and AICoder.

This slowdown is nothing unusual mathematically — no company doubles every quarter indefinitely. But it comes at a bad time for a company whose valuation hit $900B just as it posted its first quarterly profit, as we covered in our analysis Anthropic: first quarterly profit, $10.9 billion in revenue, and a $900 billion valuation. At that level of valuation, the market is no longer buying growth: it's buying its persistence.

The engine, however, remains unchanged: the Claude models. Claude Opus 4.7 (Adaptive) dominates the agentic leaderboards, Claude Sonnet 4.6 carries the volume, and the bulk of revenue comes through the API and subscriptions. Anthropic has, moreover, already taken the lead over OpenAI in the revenue race, as we detailed in Anthropic overtakes OpenAI in revenue: $30 billion run rate.

$567M Operating Loss: The Price of Flexibility

Yes, Anthropic is back in operating loss in Q3: −$567M, versus +$974M in operating profit in Q2. The document is crystal clear about the cause.

Operating expenses jump to $17.5B — nearly the level of revenue. Operationally, then, Q3 is a draw, as TrendingTopics sums up. Two line items explain the swing.

First item: $5B of compute under short-term commitments, "significantly more expensive," according to Business Insider and Euronews. Translation: Anthropic rented capacity on the spot market, at rates above those of long-term contracts, to absorb demand it didn't want to turn away. A premium paid in exchange for flexibility.

Second item: roughly $2B in non-cash charges, notably tied to philanthropy. They weigh on the bottom line without a single dollar of cash going out.

My read: this isn't a margin collapse, it's an allocation choice. And one detail is worth noting: year over year, operating expenses are up +292% while revenue is up +1,127%. Operating leverage is starting to kick in. Q2 simply benefited from an exceptionally favorable cost structure — that was a peak, not a trend.

$23.29B Net Loss: The Accounting Gap to Understand Before the IPO

No, Anthropic did not "burn" $23.3B in Q3. Between the operating loss (−$567M) and the net loss (−$23.29B), there is roughly $22.7B in charges, for the most part non-cash.

The distinction is fundamental, and it will be massively ignored:

• The operating loss measures the core business: revenue minus operating expenses (compute, salaries, R&D, marketing). Here: −$567M.
• The net loss adds the financial and one-off items: interest, taxes, and above all the fair-value remeasurement of financial instruments. Here: −$23.29B.

The culprit behind the $22.7B gap: convertible bonds. These debt securities raised before the IPO get remeasured at fair value every quarter. When Anthropic's valuation climbs, the company "loses" billions on paper — without a single dollar leaving its cash reserves. This is the same mechanism that produced spectacular accounting losses at other tech companies in their pre-IPO phase.

The headlines will read: "Anthropic loses $23B in a single quarter." It is also the least informative figure in the entire story. The serious investor starts with the operating line — and ends with the risk factors.

$895.4B in compute commitments: a liability without precedent in history

Anthropic has committed, through purchase and leasing contracts, to spend $895.4B — with "substantially all" of it earmarked for compute and AI infrastructure, according to its own S-1/A, as reported by CNBC and tao.media.

The breakdown from the document:

The counterparties are known: Amazon and Google, among others, according to the New York Post. On the other side, Anthropic shows $146.6B in cash — about six times the twelve-month maturity, but six times less than the total liabilities. La Presse notes that the contracts run through 2045, with $132B in reserve.

To put this in perspective: $895.4B is nearly 27 times the revenue for the first nine months of the year. And $447.5B by 2030 implies more than $100B in compute spending per year on average — an unprecedented pace for a company that is not yet publicly listed.

Anthropic has already shown it knows how to seize opportunities when the market offers them: the dissolution of xAI allowed it to pick up Colossus 1 and 300 MW of computing capacity, as we covered in Musk dissolves xAI and founds SpaceXAI: Anthropic picks up Colossus 1 and 300 MW of compute. But opportunistically picking up assets and committing to $895B are two very different exercises.

My take: this is the biggest infrastructure bet ever mounted. If inference demand keeps climbing, these contracts are prepaid options on capacity at locked-in prices — a huge competitive advantage. If demand weakens, it's the most beautiful fixed-cost machine ever built. The entire ANT IPO will hinge on this single variable: demand.

Concentrated revenue — and 5 to 7% from non-compliant customers

The S-1/A lifts the veil on revenue structure for the first time: 79% usage-based consumption, 44% flowing through cloud marketplaces, and 5 to 7% from non-compliant customers. Each figure tells a story of dependency.

First, the 79% usage-based: revenue is indexed to inference consumption. That's volatile, correlated with developer spending cycles, and hard to project even a quarter out.

Next, the 44% from marketplaces: nearly half of revenue flows through Amazon's and Google's cloud marketplaces. In other words, the two companies that are also the major counterparties to the compute commitments. Anthropic depends on two partners for nearly half of its revenue — partners who are simultaneously its largest capacity providers.

Finally, the 5 to 7% from violations — the most surprising passage in the document. According to the S-1/A, as reported by the Wall Street Journal and Crypto Briefing, customers in violation of the usage policy or commercial terms — circumventing restrictions, distilling models, reselling access — accounted for 5 to 7% of revenue for the nine months ended September 30, 2026.

Anthropic stepped up detection and enforcement in August and September 2026, and estimates this share will decline. But it warns in black and white that this decline "may reduce our overall revenue and growth rates in future periods".

It's an acknowledged paradox: enforcing your own terms costs growth points. Few companies in the IPO process would write that enforcing their rules will reduce their revenue. It's rare transparency — and an admission that part of the industry's growth feeds on gray-area usage.

"May imminently meet": the RSP warning tucked into a financial filing

The S-1/A contains a risk factor that no IPO prospectus had ever formulated: Anthropic's future models could "imminently" reach the threshold of its Responsible Scaling Policy for "dramatically accelerated progress attributable to AI".

What are we talking about? RSP v3.4, in effect since July 8, 2026, defines a precise threshold: as much progress in one year as in two years at the baseline rate, plausibly attributable to automated research. In other words: AI that significantly accelerates its own R&D. No current model has been designated as exceeding this threshold.

Two clarifications from CellCog's analysis: Anthropic is studying RSP updates that would impose human review and approval on certain R&D decisions, and the amendment does not designate any existing model as beyond the threshold.

Read this paragraph twice. It is both a safety warning and, implicitly, an investment argument: if future models accelerate research, Anthropic's competitive moat mechanically deepens. The document holds the bullish thesis and the bearish thesis in the same risk-factor paragraph. That has never been seen in an S-1 — and it's exactly the kind of sentence regulators will reread in ten years.

What This Changes for the ANT IPO

These numbers don't threaten the IPO — they redefine its narrative: real but decelerating growth, losses that are explained but massive, and an unprecedented compute liability.

For context: the initial S-1 was filed on October 26, 2026, with the ANT ticker reserved on Nasdaq, as we detailed in Anthropic files its S-1: ANT ticker on Nasdaq and over $1.5 trillion at stake. The November 2 amendment is its major financial update.

The market climate doesn't help matters: OpenAI has scaled back its fundraising round from $30 to $20 billion, a sign that the AI shakeout is hitting even the biggest players — we discuss this in OpenAI scales back its fundraising from $30 to $20 billion: the AI shakeout spreads. Against this backdrop, a prospectus showing both +1,127% year-over-year and a return to losses will be scrutinized line by line.

My take: this filing has a rare virtue — it doesn't sell a dream. The risk factors tell it like it is — reliance on marketplaces, non-compliant customers, AI potentially accelerating its own disruption. It's the opposite of hype. And paradoxically, that may be the strongest selling point for long-term investors.

❌ Common Mistakes

Mistake 1: Confusing net loss with operating loss

Headlines will say "$23B in losses in Q3." The operational reality is a $567M hole. The solution: always read the "operating" line before the "net" line, and check what's hiding in the gap — here, debt revaluation, not real spending.

Mistake 2: Taking the +1,127% as current growth

That's a base effect: it compares Q3 2026 to a Q3 2025 when the company was ten times smaller. Current growth is read sequentially: +47%, versus +144% in Q2. It's this deceleration that deserves comment, not the annual percentage.

Mistake 3: Believing the $895.4B leaves the treasury tomorrow

Only $69.8B falls due within 12 months, against $146.6B in liquidity. The bulk is spread out through 2045. That said, $447.5B before 2030 implies more than $100B per year on average in compute spending: the liability isn't immediate, it's colossal.

❓ Frequently Asked Questions

When will Anthropic go public?

The initial S-1 was filed on October 26, 2026, with the amendment filed on November 2, 2026. The ticker ANT is reserved on Nasdaq, but no listing date has been confirmed. An amendment of this type typically precedes the roadshow by a few weeks. Subsequent filings will be published on SEC EDGAR.

Why does the net loss ($23.29B) so greatly exceed the operating loss ($567M)?

Because the net loss includes the fair value remeasurement of convertible notes: when Anthropic's valuation rises, this debt is revalued and generates massive accounting charges without any cash outflow. Operations themselves lost "only" $567M in Q3.

Are the $895.4B in commitments already due?

No. $69.8B falls due within 12 months, $447.5B before 2030, and the remainder stretches out to 2045. Against this schedule, Anthropic shows $146.6B in liquidity. The liability is real, but its spread leaves a decade to cover it — provided demand holds up.

What does "may imminently meet" the RSP threshold mean?

That Anthropic's upcoming models could soon reach the threshold of its Responsible Scaling Policy: as much progress in one year as in two years at the baseline pace, plausibly attributable to automated research. No current model has been designated as exceeding this threshold.

What share of revenue comes from non-compliant customers?

Between 5% and 7% of revenue for the nine months ended September 30, 2026, according to the S-1/A. Anthropic strengthened detection and enforcement in August–September 2026 and anticipates a decline — while acknowledging that this decline could reduce its future revenue and growth rates.

Is Anthropic profitable?

At the operating level: yes in Q2 (+$974M), no in Q3 (−$567M). At the net level: no, with a $23.29B accounting loss in Q3, essentially non-cash. Over nine months, operations are close to break-even, but the company remains structurally loss-making as things stand.

✅ Conclusion

Anthropic's S-1/A tells the story of a company that chose to show everything: decelerating growth, costly compute, customers who cheat, and an AI that could soon accelerate its own research. Before making up your mind about the ANT stock on the big day, re-read the source document on SEC EDGAR — and check out our analysis of the initial filing for the full context.

Want to cover the next tech IPOs yourself? A site hosted on Hostinger is all you need to get started.