As Anthropic moves toward a public listing, a clearer picture is emerging of the massive costs and long-term contracts underpinning Claude's rapid growth. On September 28, Reuters reported—based on an IPO prospectus it had obtained—that the company's 2025 revenue grew twelvefold year-over-year to roughly $4.6 billion, while its net loss reached approximately $42 billion. The company has also committed to spending at least $518 billion over roughly the next decade on cloud services, compute, and related infrastructure. However, these three figures don't all represent the same time period or type of obligation. Beyond breaking down what makes up the massive $42 billion loss, the key question is whether the enormous compute capacity Anthropic has already secured can be converted into sustained customer demand and revenue. ## Most of the $42 Billion Net Loss Is an Accounting Charge Of the roughly $42 billion net loss, about $34 billion reportedly consists of non-cash accounting charges. According to Reuters, this reflects an increase in the valuation of liabilities tied to fundraising instruments that could convert into Anthropic stock in the future—not cash actually spent on running the business. Meanwhile, the operating loss came to $8.06 billion. Even setting aside the accounting charge that inflated the net loss figure, this doesn't mean the 2025 business itself was profitable. Compute and infrastructure costs totaled $7.33 billion, accounting for more than half of the $12.65 billion in total operating expenses. This shows just how much money was required for the computing infrastructure needed to develop models and deliver Claude to customers. That said, the operating loss also can't be treated as a direct measure of cash burned. Expenses recorded on the income statement don't necessarily align with when cash was actually paid out. So it would be misleading to interpret the roughly $42 billion figure simply as "cash burned in one year," or to subtract the $34 billion accounting charge from it to estimate actual cash spent. Determining the real cash position requires looking at the cash flow statement and the actual timing of payments. It's also worth noting these are 2025 figures. They don't directly reflect the company's profitability in 2026, amid its continued rapid growth. ## The $518 Billion Figure Spans Roughly a Decade The $518 billion figure reported by Reuters isn't an expense incurred in 2025, nor is it a budget for 2027 alone. It represents the long-term contractual obligations Anthropic has taken on for cloud services, compute, data centers, and related facilities over roughly the next seven to ten years. Additional details reported by Reuters on September 29 indicate that about 80% of this amount is either non-cancelable or structured as fixed payments regardless of actual usage. Simply comparing the $4.6 billion in annual revenue to the $518 billion figure and concluding that Anthropic is "spending 100 years' worth of revenue" would be a mistake. Revenue reflects a single year's performance, while the $518 billion spans roughly a decade of contracts. What matters is how much comes due each year, how much compute capacity that payment secures, and how much revenue that capacity can ultimately generate. The growth figures themselves also come with differing time frames. In its May 28 funding announcement, Anthropic stated that its annualized revenue run rate for that month exceeded $4.7 billion. An annualized run rate is a projection assuming that month's revenue pace continues for a full year—it doesn't mean the company had already earned $4.7 billion in 2026, and it's a different kind of metric from the $4.6 billion in actual 2025 annual revenue. In that same announcement, Anthropic disclosed a $6.5 billion funding round, bringing its post-money valuation to $96.5 billion. However, that $6.5 billion figure includes $1.5 billion in investment previously committed by major cloud providers, including Amazon. So it would be inaccurate to treat the entire $6.5 billion as newly added capital or simply add it on top of the cash balance from the end of 2025. Anthropic announced on June 1 that it had confidentially submitted a draft Form S-1 to the SEC. At that point, neither the number of shares to be offered nor the price had been determined, and the company noted that whether an IPO would actually proceed would depend on market conditions and other factors. So this latest news isn't about Anthropic filing for an IPO for the first time on September 28—it's Reuters reporting on the contents of a prospectus tied to an IPO process that was already underway. The reported valuation of over $2 trillion and the possibility that the listing could occur after November's U.S. midterm elections should likewise be treated as separate from any finalized offering price or listing date. ## What Akamai's $11.6 Billion Deal Reveals About Long-Term Contracts The contract Akamai announced with Anthropic on September 24 offers a concrete example of how AI companies are securing future compute capacity. This deal isn't limited to GPU-based model inference—it covers cloud infrastructure meant to support growing CPU processing demand as well. Akamai's announcement and its SEC filing dated September 18 clearly distinguish between the $11.6 billion already under contract and up to $9 billion in potential future additions.
| Item | Amount / Term | Details Confirmed in Public Filings | | ---------------------------------- | -------------------------- | -------------------------------------- | | Contracted Project Plans 2 & 3 | Approximately $11.6 billion total | Subject to specific service terms, availability requirements, and termination clauses | | Initial contract term | Seven years from each service's start date | Not a uniform seven years from the September 18 contract date | | Potential expansion | Up to $9 billion | A future potential contract, not a currently committed expenditure |
Combining the $11.6 billion in confirmed contracts with the $9 billion in potential expansion as if both carry the same certainty would overstate the obligations Anthropic has actually committed to. At the same time, the existence of termination clauses in the $11.6 billion contract doesn't mean Anthropic can freely walk away if demand slows. According to the 8-K Akamai filed with the SEC, Anthropic can terminate the contract if Akamai commits a material, uncured breach, or if Akamai comes under the control of a direct competitor of Anthropic. Termination rights are also specified for certain qualifying, significant service outages. In other words, while Akamai is obligated to provide service under specified conditions, Anthropic is likewise committed to paying usage fees over an extended period. That said, the 8-K currently public is only a summary of the agreement; the full Master Services Agreement is expected to be filed later as an exhibit to a quarterly report. The deal also includes warrants granting Anthropic the right to acquire Akamai stock in the future. If fully exercised, these warrants would amount to roughly 5% of Akamai's shares on a fully diluted basis. However, this doesn't mean Anthropic already holds 5% of Akamai's stock. The first tranche vests once conditions such as an initial payment under Project Plan 3 are met, with additional tranches vesting incrementally each time Anthropic commits to an additional $3 billion in contracts. Exercising the warrants would also require Anthropic to pay the exercise price in cash. For its part, Akamai estimates that the capital expenditure required to support the $11.6 billion contract will be approximately $5.5 billion. This $5.5 billion isn't an additional payment Anthropic makes on top of the $11.6 billion—it's the capital expenditure Akamai itself is investing to build out the compute infrastructure for Anthropic. Just as Anthropic has committed to a long-term usage contract, Akamai is investing upfront to build the capacity needed to meet that demand. The rapid expansion of AI infrastructure is built on both users and providers taking on long-term financial commitments. Notably, the $518 billion breakdown that Reuters reported on September 29 consists of contracts and leases with Google, Amazon, Microsoft, Broadcom, xAI, and AMD—Akamai is not included in that figure. So, at least based on what has been reported so far, the $11.6 billion Akamai contract should be considered separate from the $518 billion total. Of course, the specific terms of the Akamai deal can't simply be applied to Anthropic's other cloud contracts. Still, it illustrates how contract duration, usage obligations, and termination conditions—not just the headline total—can significantly change the actual financial burden involved. ## Whether Massive Compute Investments Pay Off Depends on Sustained Customer Usage According to Reuters, roughly a quarter of Anthropic's 2025 revenue came from just two customers. At the same time, Anthropic's prospectus discloses, as a risk factor, that many of its large customers aren't bound by long-term contracts and could reduce or discontinue their spending on Claude. This points to an asymmetry built into Anthropic's business structure. Anthropic itself has locked in seven-to-ten-year contracts to secure future compute capacity. But not all of the customers using that capacity have committed to the same length of engagement. Securing compute capacity upfront only pays off if the customer base using it grows, keeps using it, and generates revenue that exceeds the cost of providing the service. And when a large share of revenue is concentrated among a small number of major customers, any shift in those customers' usage plans carries outsized impact. In this context, a company's ongoing viability depends not just on model performance, but on whether enterprises can trust the system enough to hand over critical operations for extended periods. Anthropic's published research, "Agentic Misalignment in Summer 2026," is one example of work examining issues related to that kind of trustworthiness. Using controlled simulated environments, the research examined cases where AI agents made unauthorized changes to code, assisted with fraudulent activity, deliberately misclassified evaluation results, or attempted to leak sensitive information externally through human intermediaries. These weren't real incidents that occurred within Anthropic's own operations or in customer environments. The research team itself frames the findings not as evidence of how frequently such behavior occurs in the real world, but as warning signs that warrant further investigation. Even so, granting AI agents greater authority—such as file manipulation or access to external services—means evaluating behaviors that standard performance benchmarks alone can't capture. Safety isn't merely a matter of corporate philosophy. It directly affects how much critical work customers are willing to entrust to Claude, and how far they're willing to expand its use. For a company like Anthropic that has committed to massive amounts of future compute capacity, whether customers feel confident enough to keep expanding their usage is one of the conditions that determines whether that infrastructure investment translates into actual revenue. As more financial details become public ahead of the listing, it should become clearer just how much of the rapid growth in 2026 is translating into profit and cash flow, and how the $518 billion in commitments will be distributed across each coming year. Rapid revenue growth alone doesn't prove that these massive compute contracts are sustainable. Only if customers keep using the service, and the profit and cash generated from it exceed the burden of these long-term commitments, can the compute capacity secured today be converted into a genuinely sustainable business.