Anthropic's revenue surged by 46% but cannot hide the anxiety over infrastructure: the brutal game between the AI bubble and high macro interest rates

The latest revenue data from Anthropic, which has garnered global attention, has finally been released. In the third quarter, revenue reached $16.8 billion, a nearly 46% increase quarter-over-quarter, achieving adjusted profitability for the first time, with free cash flow turning positive for the quarter.

In the same period in 2025, this figure was only $787 million. By the first quarter of 2026, revenue rose to $4.73 billion, reaching $11.5 billion in the second quarter.

In just over half a year, revenue surged from less than $5 billion to $16.8 billion, exceeding 3.6 times the total for the entire year of 2025 in just one quarter, and crossing the breakeven point. In the report card of any tech company, these are textbook-level numbers.

That same evening, however, the U.S. stock market presented a starkly different picture.

The Nasdaq Composite Index fell 1.88%, closing at 26,692.87 points, marking the largest single-day drop since July. The S&P 500 Index dropped 1.11%, closing at 7,718.94 points. The Philadelphia Semiconductor Index fell over 4%, with a cumulative decline of about 10% over six trading days, resulting in a market capitalization loss of over $1 trillion for its constituents. Computing assets plummeted sharply, with CoreWeave experiencing a drop of 16% at one point, Dell falling nearly 12%, Oracle down 9%, Nebius down 11%, and the optical communication and storage chip sectors collectively under pressure.

The next day, the sell-off spread to Asia. The South Korean KOSPI triggered a circuit breaker.

Market sentiment was extremely poor.

Anthropic has proven that real demand exists, products have buyers, and the business model is crossing the breakeven point. However, at the same time, the macro environment and financial pipelines supporting this massive technological narrative are simultaneously revealing vulnerabilities across multiple dimensions.

Inflation Rebound and 5.2% Treasury Yield

The direct sell-off faced by global tech stocks that day primarily stemmed from the resurgence of inflation data.

The U.S. Bureau of Labor Statistics reported signs of reversal in the September CPI. The September CPI rose 3.7% year-over-year, higher than August's 3.4% and the market expectation of 3.6%, with a month-over-month increase of 0.6%, reaching a five-month high. Core CPI rose 0.3% month-over-month and 2.6% year-over-year. The main drivers were energy, with gasoline prices rising 9.1% month-over-month and 34.2% year-over-year, and heating oil up 74.2% year-over-year. Brent crude oil reached $106 per barrel.

Inflation not only accelerated again but also began to spread to a broader range of goods and services.

Market pricing quickly shifted. The CME FedWatch Tool indicated that the probability of a rate hike at the Federal Reserve's meeting at the end of October jumped from about 22% to 44%. The yield on 10-year U.S. Treasuries rose to 5.188%, while the yield on 30-year U.S. Treasuries hit 5.522%. The money market began to fully price in two rate hikes before the end of January next year, with the U.S. dollar index hitting a new high since May 2025.

Massive computing power expansion has driven up electricity loads, pushing up prices for transformers, copper cables, high-bandwidth storage chips, and data center land rents, entering the price basket. Rising prices have suppressed expectations for monetary easing, pushing up nominal interest rates and consequently raising the discount rate for future cash flows.

Many long-term capital commitments signed under the assumption of low interest rates in 2024 now must seek funding in an environment with the highest capital costs since 2007. Revenue addresses the numerator, while interest rates determine the denominator. When the speed of the denominator's increase exceeds the speed of the numerator's expansion, asset valuations at the level of the assets will face overall downward adjustments.

Financing Channels are Clogging at the Physical Level

In addition to the rise in macro interest rates, a series of signals indicating a standstill have also emerged in the financing channels of the AI infrastructure chain.

In the past week, three data center IPO plans in the Asia-Pacific region have faced obstacles.

The Australian data center operator Firmus Grid, supported by Nvidia, withdrew its IPO on October 9. Originally planned to raise up to $5.5 billion, this was expected to be one of the largest IPOs in Australia in thirty years. The company's current revenue is about $51 million, but the intended issuance valuation was as high as $30.6 billion, corresponding to nearly 600 times the sales ratio. The company showcased about $72.8 billion in future intent contracts, but only 5% of the actual capacity has been built, with remaining projects requiring hundreds of billions of dollars in construction funding. Even with a 25% reduction in the issuance price during the book-building period, institutional buying interest remained low, leading to the final withdrawal.

The delayed IPO of AirTrunk, a subsidiary of Blackstone Group, is an even heavier signal.

AirTrunk is a large-scale computing infrastructure platform jointly acquired by Blackstone Group and the Canada Pension Plan Investment Board for AUD 24 billion in 2024, with assets covering Australia, Japan, Singapore, Malaysia, Hong Kong, and India. It was originally planned to raise about $1.5 billion in Singapore through a REIT structure, seen as a defensive AI infrastructure target with stable leases and dividend returns.

This IPO plan has now been postponed until 2027. Faced with a benchmark 10-year U.S. Treasury yield of around 5.2%, even if the issuer considers raising the dividend yield to 6%, the approximately 78 basis points of credit and liquidity compensation still fails to attract institutional investors seeking certainty. In the same week, the Philippine telecom operator's data center Vitro postponed its IPO plan to 2027, and SoftBank's energy company SB Energy also suspended its IPO roadshow.

More indicative of the shift in funding direction than the cooling of equity financing is the credit market's cutting attitude.

On October 12, AI cloud infrastructure platform Volta Infrastructure launched a $5.012 billion six-year syndicated loan. Of this, $1.3 billion in term loans was underwritten by JPMorgan and Goldman Sachs, while $3.712 billion in special loans for procuring about 36,000 Nvidia GPUs was refused by the two major banks for underwriting and was directly distributed to the public market. This unrated hardware mortgage loan was priced at SOFR plus 625 to 650 basis points, with an overall yield approaching 11.5%.

The two top investment banks were willing to arrange the deal for fees but refused to keep the exposure of rapidly depreciating hardware assets on their balance sheets.

CoreWeave's CDS Fluctuations and Cloud Computing Revaluation

If traditional banks are tightening underwriting risk control, the secondary market and derivatives market have begun to signal red lights for highly leveraged computing lessors.

On October 12, market data showed that CoreWeave's five-year credit default swap (CDS) quotes soared to nearly 881 basis points, while at the end of June, this indicator was around 547 basis points.

CDS are essentially insurance contracts purchased by investors against debt default. When a CDS quote jumps hundreds of basis points in a short time, it means that bondholders must pay a much higher insurance premium to guard against or hedge default risk.

Trace pricing data shows that CoreWeave's unsecured high-yield bonds maturing in 2032 have fallen below 80% of par value, with secondary market yields soaring to nearly 15%, and the yield on bonds maturing in 2031 with a coupon of 9.75% also rising to 14.9%.

When a company's borrowing costs are pushed to 15%, the capital market is no longer offering a premium for growth assets but pushing it into the pricing range of distressed assets. Based on the current CDS levels, the market-implied five-year default probability has exceeded 50%.

Elon Musk described this data as "scary" on social media. Analysts have referred to CoreWeave as the "subprime loan tier" in the AI boom, warning that if refinancing costs remain high, the valuations of heavy asset lessors will face significant pressure. Although CoreWeave's management insists that backlogged orders are sufficient to support debt repayment, the stock market has also experienced severe fluctuations, with prices dropping 16% at one point.

The destructive power of this signal lies in its penetration beyond a single company, affecting the financing conditions of the entire NeoCloud group.

As the most aggressively expanding group in this round of computing infrastructure, NeoCloud largely relies on balance sheet arbitrage. By mortgaging Nvidia GPUs to Wall Street investment banks and private credit for billions in loans, they purchase chips and then lease them to cutting-edge large model startups, cycling to increase debt leverage.

When the leading company's CDS soars to 881 basis points and bond yields reach 15%, the temperature of the entire blood supply pipeline has completely changed. Creditors are beginning to reassess the credit risks of the entire NeoCloud sector, significantly raising the risk compensation for new loans, and tightening asset collateral ratios and guarantee terms. For computing cloud vendors that need to continuously borrow to repay old debts and bear the capital expenditures for the next generation of chips, the significant rise in funding costs directly narrows the original arbitrage space.

On one side is the grand blueprint presented by cutting-edge laboratories to the outside world, and on the other side are intermediary institutions tasked with buying cards and leveraging, facing distressed-level default premiums charged by the capital market. These two scenes placed at the same trading desk directly change the risk appetite of buyers.

Public Statements from Broadcom and Intel Leaders

The tension at the capital level was laid bare before global investors at the DBS Bank Asian Future Frontiers Conference held in Singapore on October 12.

Broadcom CEO Hock Tan provided a frequently cited judgment in his speech. He believes that the biggest constraint on the AI industry in the future may no longer be production capacity, but capital.

Hock Tan reviewed the industry supply chain over the past two years, noting shortages in high-bandwidth memory, insufficient wafer production capacity, and slow grid access, all of which are gradually being overcome. The wafer shortage is expected to ease in the next two to three years, and new power generation equipment is also gradually being connected to the grid. However, the patience of capital and the cost of funds are becoming hard constraints. He candidly stated that many investors entered the market with extreme expectations, a state closer to a casino, while the actual efficiency of businesses applying AI still needs to provide more convincing evidence.

Intel CEO Pat Gelsinger expressed similar views at the same event. He pointed out that the most pressing challenge right now is the rising cost of capital and the stringent scrutiny of investors regarding return on investment (ROI).

Both semiconductor leaders do not deny the real demand for computing power, but they jointly pointed out a premise: the sustainability of growth in the coming years ultimately depends on whether downstream customers can prove that AI brings real monetary commercial returns. If the industrial penetration of cutting-edge models falls short of expectations, the procurement pace of upstream computing infrastructure may be adjusted.

DBS Bank CEO Piyush Gupta provided a qualitative financial perspective on Bloomberg Television, stating that the risks associated with AI-related investments are being repriced, with the funding side demanding higher return compensation and stricter transaction structures. DBS Bank has already led about $20 billion in computing-related syndicated loans this year, but it generally retains only a very small proportion of exposure in transactions.

The capital faucet has not been completely turned off; it has merely shifted in direction and risk tolerance.

Lowered Expectations for Large Model Revenues

Meanwhile, within the narrative of cutting-edge large models, some easily overlooked detail adjustments have begun to emerge.

First is the transparency of revenue metrics. Updated data from OpenAI to institutional investors shows that as of the end of September, its annualized revenue scale is about $50 billion, lower than the previously widely circulated $70 billion. The reason for the downward adjustment is that the new metric excludes revenue-sharing portions from cloud partners.

This action has triggered a horizontal comparison in the market. If OpenAI adopts a net basis that excludes channel revenue sharing, the gross revenue from Amazon and Google Cloud channel flows for Anthropic loses its comparable benchmark. When the expected direction of core industry metrics shifts from upward adjustments to downward squeezes, the valuation system faces realignment.

In terms of the pace of technological development, the frontier laboratories themselves are also signaling a slowdown. At the beginning of September, OpenAI CEO Sam Altman clearly stated that the company would not pursue an IPO within 2026, citing that model safety issues require more time to address. Anthropic CEO Dario Amodei also published a statement around the same time, calling for the entire industry to proactively slow down the pace of capability advancement to allow for a window for safety alignment research.

The restrained attitudes of both companies are directly related to the abnormal events exposed in recent safety tests. When leading forces in the industry begin to call for a slowdown, the capital market's linear extrapolation of the endless acceleration of computing power demand loses a layer of emotional support.

The Balance Sheet Behind the Impressive Numbers

Returning to Anthropic's $16.8 billion report card.

Based on the disclosed annualized run rate of $65 billion at the end of July, the $16.8 billion in the third quarter has met the passing line, with an adjusted gross margin maintained at around 85%, and an adjusted operating profit margin exceeding 11%, with both operating cash flow and free cash flow turning positive.

However, the market's focus is on the growth slope.

From the first to the second quarter of 2026, the company's revenue grew approximately 143% quarter-over-quarter. From the second to the third quarter, the quarter-over-quarter growth rate slowed to 46%. The transition from doubling acceleration to a change in growth rate occurred in just two quarters.

The annualized run rate for the third quarter is approximately $67 billion, which is a slight increase compared to the publicly disclosed $65 billion at the end of July, indicating that the monthly net increase from August to September may have already shown a flattening trend. The core assumption that gives rise to the extremely high valuation multiples often relies on a very high slope of exponential expansion. Once the slope changes, the benchmark for the valuation model will face recalibration.

Based on the current annualized revenue of $67 billion, the IPO target valuation of $2 trillion corresponds to an approximate 30 times sales multiple. Five months ago, Anthropic completed a post-investment valuation of $965 billion, just surpassing OpenAI in valuation. Months later, planning to push towards the public market with a doubled valuation of $2 trillion, this span has caused potential cornerstone investors to show hesitation.

Underwriters are using models based on forward revenues for 2028 to anchor valuations, with Anthropic presenting a revenue target for 2028 between $190 billion and $200 billion to institutions. If calculated based on $200 billion in forward revenue, a $2 trillion market cap corresponds to about 10 times the forward sales multiple.

Whether the public market is willing to fully trust the forward assumptions for two years from now constitutes the biggest point of divergence. The founder of the independent research firm New Constructs publicly raised questions, suggesting that to support a $2 trillion valuation, its steady-state net profit needs to approach that of Nvidia. The firm believes that the excessively high listing valuation is more a liquidity pricing sought by primary market capital for exit rather than a reflection of the true enterprise value.

Greater financial hard constraints are hidden in the undisclosed prospectus. Content disclosed by Reuters shows that Anthropic has committed to investing at least $518 billion in AI infrastructure over the next decade, with partners including six computing power and cloud platforms, of which about 80% of the contracts are irrevocable or include minimum payment agreements that must be paid in full regardless of actual usage.

Contract breakdowns show that the company has committed approximately $161.2 billion for equipment leasing related to Broadcom, with an agreement with Google amounting to $111.1 billion, covering a performance period from April 2026 to July 2033, and clearly stipulating that if actual computing power consumption is insufficient, the difference must be made up. The cooperation amount with Amazon is approximately $110 billion, with an execution period until April 2036, also including strict minimum payment clauses, and the computing power agreement with Microsoft is about $31.4 billion.

Compared to OpenAI's shared infrastructure allocation mechanism built with SoftBank, Oracle, and others, Anthropic bears an extremely high individual performance responsibility. Using a revenue scale still in its cultivation phase to bear over $500 billion in rigid payment flows essentially shifts the risk of computing power consumption over the next few years entirely onto its own balance sheet. If the procurement budget for downstream enterprise software does not expand as expected, these irrevocable minimum payments will directly translate into continuous rigid cash flow expenditures.

The details of revenue quality are also under scrutiny. The company's top two clients contribute about 24% of revenue, and most have flexible usage contracts, lacking long-term locking mechanisms. By 2025, approximately 47% of the company's revenue will come from Amazon and Google's cloud channels, with a platform revenue share of around 20%. This means that gross revenue cannot be fully converted into retained funds on the books, and the two major cloud channels are both investors and suppliers, as well as direct competitors to self-developed models.

On the product pricing side, pressure from the global open-source ecosystem and closed-source competitors is gradually becoming apparent. Data from Artificial Analysis shows that the usage price of Anthropic's flagship model exceeds that of OpenAI's comparable products by more than double. Chinese open-source and open-weight models, with extreme cost control and usability, are prompting overseas companies to restructure their AI budgets. Research from software spending analysis firm Ramp indicates that enterprise clients are reducing unchecked model usage and turning to more cost-effective alternatives. As the mixed token prices across the industry continue to decline, maintaining an 85% gross margin will heavily rely on the speed of improvement in inference efficiency outpacing the speed of price declines.

In the enterprise deployment sector, data from Counterpoint Research shows that Anthropic achieved a 31.4% revenue share in large language models in the first quarter of 2026, ranking first in the industry, with leading institutions such as BlackRock and Barclays Bank integrating the Claude system. However, on the consumer traffic side, statistics from Sensor Tower show that ChatGPT holds a 46.4% global share, Gemini rises to 27.7%, while Claude only accounts for 10.3%. The competition among underlying large models is extremely dynamic, and the window of technological advantage may not cover the depreciation cycle of computing power contracts lasting over a decade.

The over-the-counter trading market has already provided immediate feedback. The off-exchange trading code ANTH set up by Hyperliquid has seen its price drop from above the $2 trillion mark to a range of $1.89 trillion to $1.94 trillion.

The $16.8 billion disclosed by Anthropic mirrors the 1.88% decline in Nasdaq.

On one side is the astonishing commercialization expansion of frontier large model companies, with revenue soaring from $4.7 billion to $16.8 billion within six months, achieving underlying profitability and crossing the cash flow breakeven point. This proves that the willingness of enterprises to pay for advanced intelligent tools is real, and the direction of technological advancement has not changed.

On the other side is the rational reconstruction of the capital market, with $518 billion in irrevocable procurement minimums, soaring debt default insurance nearing 881 basis points, approaching 15% borrowing costs for computing power intermediaries, continuous delays in the listing of data center assets, and pressure towards a 5.2% benchmark treasury yield.

Revenue proves that demand is indeed exploding. However, the capital system is voting with its feet on the real costs required to support this future.