When we published our October analysis of Anthropic's pre-IPO finances, we worked from the company's original Form S-1, which disclosed $430 billion in total AI infrastructure commitments and a $51.8 billion payment obligation due within twelve months against an $85 billion cash pile. We called it a sinking ship. We were right about the direction. We dramatically underestimated the scale.

Anthropic's amended Form S-1/A filing, submitted November 2, 2026, ahead of its planned Nasdaq listing under the ticker "ANT," revises those numbers upward across the board. The company has now committed to $895.4 billion in long-term data center and compute contracts—more than double what the original filing disclosed. Of that total, $69.8 billion comes due within the next twelve months, roughly $18 billion more than the first S-1 showed. The amendment updates the company's financial statements through the third quarter and lands just days before its IPO roadshow is expected to begin.

The headline numbers look like a triumph: $16.96 billion in Q3 revenue, a twelvefold increase year-over-year from $1.4 billion in Q3 2025, and $33.24 billion in revenue over the first nine months of 2026. But the same document that shows the commercial scaling of the Claude platform also shows a growth curve bending downward, operating losses returning, and a near-trillion-dollar lease backlog that exceeds what the company can currently earn in a year.

What the Original S-1 Disclosed Versus the Amendment

Our October analysis warned that Anthropic was trapped by a massive financial burden. The amended SEC numbers show the burden is far larger—and the cash cushion is larger too, which is why the company is not yet insolvent even as its liabilities have nearly doubled in the span of a few weeks.

• Total AI infrastructure commitments: $430 billion (original S-1, October) vs. $895.4 billion (amended S-1/A, November 2)
• On-hand cash: $85 billion (original S-1) vs. $132 billion in cash and short-term investments, $146.6 billion total liquidity (amended)
• Next-12-month payments: ~$51.8 billion (original S-1) vs. $69.8 billion (amended)

The $69.8 billion figure matters because it exceeds the company's entire annualized revenue run rate. Based on Q3's $16.96 billion, Anthropic is annualizing at roughly $67.8 billion—about $2 billion short of the baseline data center rent due over the next year. The company's yearly intake from Claude subscriptions, enterprise API contracts, and cloud partner sales cannot cover just the servers before any other operating expense.

The Growth Is Slowing at the Worst Possible Moment

The filing confirms that sequential growth decelerated sharply. Quarter-over-quarter growth fell to 47% in Q3—revenue rose from $11.56 billion in Q2 to $16.96 billion—down from 144% in Q2 and 132% in Q1. That is a significant cooling from the explosive early-year pace, and it matters because the annualized run rate that anchors the bull case is entirely dependent on Q3's record momentum holding.

The company also swung back to a $567 million GAAP operating loss after posting a $974 million operating profit in Q2. The filing attributes the reversal primarily to $5.0 billion of compute under short-term commitments, alongside a $2 billion non-cash charge tied to an employee donation-matching program. Total operating expenses rose to $17.5 billion from $10.6 billion in the prior quarter.

The headline GAAP net loss of $23.29 billion looks catastrophic, but management notes that $21.9 billion of it is a non-cash accounting adjustment from revaluing convertible notes as the company's enterprise valuation climbs. The operational reality is captured by the $567 million operating loss and the $6.6 billion in cash generated from operations over the first nine months—cash that immediately flows out the door to pay for compute.

Where the Extra Bills Came From

The sudden inflation in near-term obligations has two distinct sources. First, Anthropic spent $5 billion on short-term "spot market" compute because demand for Claude outran its pre-booked capacity. Rather than rate-limit users before the IPO, the company leased server space on the fly at premium retail prices—a choice that preserved the growth narrative and erased the operating profit.

Second, the company signed an additional $100 billion in long-term infrastructure commitments in the five weeks after Q3 ended while marketing the offering, pushing lifetime liabilities from the $417 billion disclosed in late October to $895.4 billion on November 2. The company finalized these deals before closing the books, meaning the amendment captures commitments made specifically in the run-up to the listing.

Those figures still exclude the real-world costs that come with building data centers. The S-1/A flags energy price volatility and a five-year power bottleneck, water-intensive cooling as server racks generate unprecedented thermal output, inter-facility networking and bandwidth, and local compliance fees tied to voluntary offset commitments. The filing also excludes an additional $125 billion to $150 billion in unfinalized future TPU compute from Google slated for 2028 through 2031.

The Math Is Not Mathing

To justify an $895 billion infrastructure plan, Anthropic needs returns that vastly exceed the cost of interest, energy, depreciation, and basic margins. Total global enterprise software spending sits around $1 trillion annually. For Anthropic and OpenAI to generate enough revenue to pay off their combined multi-trillion-dollar compute bills, they would need to cannibalize every dollar currently spent on Salesforce, Microsoft Office, Oracle, SAP, and cybersecurity worldwide.

Current real-world studies show AI integration yields a 10% to 30% efficiency bump in specific tasks. A 20% productivity gain does not unlock the trillions of dollars of corporate cash required to fund this backlog. And the problem compounds: unlike factories or bridges that last decades, AI server hardware becomes obsolete every three to four years, forcing Anthropic to re-commit to the next hardware cycle before the current leases even expire. If revenue plateaus, the company will be paying for empty server space—a deleveraging cascade that hits its primary backers, Amazon and Google, who built out data centers explicitly expecting Anthropic to pay rent.

There is also the accounting question. Anthropic records the gross amount of sales through cloud partners—if a client buys $100 of Claude access through AWS, the full $100 counts as top-line revenue even though a chunk stays with Amazon to cover hosting. The annualized run rate is back-weighted to the best quarter in company history, so a modest Q4 plateau collapses the projection. OpenAI, by contrast, uses net accounting, and its real annualized revenue was quietly recalculated downward to around $50 billion. Both companies are running deep operational losses to sustain the pace, but Anthropic's numbers are the ones that will be under a public microscope every quarter after the listing.

The AGI Bet Behind the Balance Sheet

The entire spending plan rests on a technological assumption: that scaling current models eventually produces general intelligence valuable enough to justify the bill. A growing number of physicists, mathematicians, and computer scientists argue that current silicon-based approaches face hard limits. The thermodynamic argument holds that processing the combinations required for true general reasoning would demand more electrical power than the planet generates, against a human brain that runs on roughly 20 watts. The combinatorial argument holds that mapping real-world complexity requires more unique data points than exist in human history, and models are already running out of human-generated text. The computability argument holds that transformer models are statistical prediction engines, and true causal reasoning cannot be derived purely from token probabilities.

None of this means Anthropic's models are worthless. It means the $895 billion commitment is funding an empirical experiment, not a guaranteed evolution. If capability gains hit a wall, the leases remain.

Why an IPO Is All They Have Left

The IPO is not a victory lap. It is a capital injection to keep the machine from running out of fuel. Private venture capital cannot supply the tens of billions per year the infrastructure treadmill demands, so the listing serves three immediate purposes: it lets early backers cash out, it secures liquidity while the growth narrative still holds, and it creates publicly traded stock to compensate researchers without draining remaining cash reserves.

Institutional investors appear to have noticed. Valuation expectations have reportedly slid from the initially hoped-for $2 trillion toward $1.5 trillion. The company maintains robust liquidity at $146.6 billion and generated positive operating cash flow of $6.6 billion over nine months, and bulls point to a 15% adjusted non-GAAP operating margin. But the explosive infrastructure spend and cooling growth rates are forcing a more conservative read.

Our October analysis said Anthropic was trapped. The amended filing shows the trap is nearly twice as large as the original S-1 disclosed, the growth is cooling faster than anyone expected, and the revenue still cannot cover the bill. The company is not insolvent. But it is broker than it appeared when it first filed, it is asking the public to fund the difference, and the math behind the spend is an argument, not a proof.