In the same period of 2025, this figure stood at only $787 million. By Q1 2026, revenue rose to $4.73 billion, and reached $11.5 billion in Q2.

Within half a year, revenue surged from under $5 billion to $16.8 billion—a quarterly increase exceeding 3.6 times the full-year 2025 total—and crossed the breakeven threshold. In any tech company’s financial report, these figures would be textbook-perfect.

That same evening, however, the U.S. equity market presented an entirely different picture.

The Nasdaq Composite Index fell 1.88%, closing at 26,692.87 points, marking its largest single-day drop since July. The S&P 500 declined 1.11%, ending at 7,718.94 points. The Philadelphia Semiconductor Index dropped over 4%, with cumulative losses of nearly 10% across six trading days, erasing more than $1 trillion in component stock market value. Compute assets plunged sharply—CoreWeave briefly fell 16% intraday, Dell dropped nearly 12%, Oracle shed 9%, Nebius slid 11%, while optical communications and memory chip sectors came under broad pressure.

The next day, selling spread to Asia. The Korean KOSPI index triggered circuit breaker mechanisms.

Market sentiment was extremely poor.

Anthropic has demonstrated real demand exists, products are being purchased, and the business model is crossing the breakeven point. Yet at the same moment, the macro environment and financial pipelines sustaining this massive technological narrative are simultaneously revealing vulnerabilities across multiple dimensions.

Inflation Resurgence and a 5.2% Risk-Free Yield

The immediate downward pressure on global tech stocks that day stemmed first from resurgent inflation data.

The U.S. Bureau of Labor Statistics’ September CPI showed reversal signs. September CPI rose 3.7% year-on-year, surpassing August’s 3.4% and the market consensus of 3.6%, with a month-on-month increase of 0.6%, the highest in five months. Core CPI rose 0.3% month-on-month and 2.6% year-on-year. The primary drivers were energy: gasoline jumped 9.1% month-on-month and 34.2% year-on-year; heating oil surged 74.2% year-on-year. Brent crude oil climbed past $106 per barrel.

Inflation not only resumed acceleration but also spread to broader goods and services.

Market pricing rapidly shifted. CME FedWatch Tool indicated the probability of a rate hike at the Fed’s October 31 meeting jumped from around 22% to 44%. The 10-year U.S. Treasury yield rose to 5.188%, and the 30-year yield hit 5.522%. Money markets began fully pricing in two rate hikes before early January 2025. The U.S. Dollar Index reached its highest level since May 2025.

Massive compute expansion drove up electricity load, pushing up transformer and copper cable prices, high-bandwidth storage chip costs, and data center land rents—all entering the consumer price basket. Rising prices suppressed expectations for monetary easing, elevated nominal interest rates, and consequently increased the discount rate applied to future cash flows.

Many long-term capital commitments signed under low-interest assumptions in 2024 now face funding challenges in an environment of capital costs at their highest levels since 2007. Revenue addresses the numerator; interest rates determine the denominator. When the denominator rises faster than the numerator expands, asset valuations face systematic downward revisions.

Funding Channels Are Physically Clogging

Beyond rising macro interest rates, financing channels along the AI infrastructure chain are also showing signs of stagnation.

Over the past week, three data center IPO plans in the Asia-Pacific region were successively stalled.

Australian data center operator Firmus Grid, backed by NVIDIA, withdrew its IPO on October 9. Originally planned to raise up to $5.5 billion, it would have been one of Australia’s largest-ever listings in three decades. The company currently generates around $51 million in revenue, yet its proposed valuation reached $30.6 billion—equivalent to nearly 600 times its sales multiple. The firm publicly highlighted approximately $72.8 billion in future contract intentions, but only 5% of actual capacity had been built, with the remainder requiring hundreds of billions in additional investment. Even after a 25% price reduction during bookbuilding, institutional demand remained tepid, leading to the cancellation.

The delay of Blackstone’s AirTrunk IPO was an even more significant signal.

AirTrunk is a hyperscale compute infrastructure platform acquired jointly by Blackstone Group and Canada Pension Plan Investment Board (CPPIB) in 2024 for AUD 24 billion. Its assets span Australia, Japan, Singapore, Malaysia, Hong Kong, and India. Originally slated for a REIT listing in Singapore aiming to raise ~$1.5 billion, it was viewed as a defensive AI infrastructure play with stable leases and dividend yields.

The IPO was ultimately postponed until 2027. Facing a 10-year U.S. Treasury benchmark of around 5.2%, even if the issuer considered raising its dividend yield to 6%, the mere 78 basis points of credit and liquidity compensation could not attract institutions seeking certainty. Within the same week, Philippines telecom operator’s data center subsidiary Vitro delayed its IPO until 2027, and SoftBank’s energy arm SB Energy halted its IPO roadshow.

More telling than the cooling of equity financing is the cold stance of the credit markets.

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 were underwritten by JPMorgan and Goldman Sachs. However, the $3.712 billion facility designated for purchasing approximately 36,000 NVIDIA GPUs was rejected by both banks for underwriting and instead distributed through open markets. This unrated hardware-secured loan priced at SOFR plus 625 to 650 basis points, yielding close to 11.5% upon maturity.

Two top-tier investment banks were willing to act as arrangers for fee income but refused to retain the exposure to rapid depreciation of hardware assets on their balance sheets.

CoreWeave’s CDS Anomaly and Compute Cloud Revaluation

If traditional banks are tightening underwriting risk controls, the secondary market and derivatives market have begun flashing red lights at highly leveraged compute leasing firms.

On October 12, market data showed CoreWeave’s five-year Credit Default Swap (CDS) quote surged to near 881 basis points—up from around 547 basis points in mid-June.

CDS is essentially an insurance contract investors purchase to hedge against debt default. A sudden jump of several hundred basis points in a CDS quote indicates bondholders must pay significantly higher premiums to hedge or mitigate default risk.

Trace pricing data revealed CoreWeave’s unsecured high-yield bonds due in 2032 traded below 80% of face value, with secondary market yields soaring to nearly 15%. Similarly, the 2031 bond with a 9.75% coupon saw yields rise to 14.9%.

When a company’s borrowing cost reaches 15%, the capital market no longer assigns a growth premium—it has effectively moved into the distressed asset pricing range. At current CDS levels, implied five-year default probability exceeds 50%.

Musk described the data as “scary” on social media. Analysts labeled CoreWeave as the “subprime tranche” within the AI boom, warning that persistently high refinancing costs would place heavy pressure on asset-heavy leasing firms’ valuations. While CoreWeave management insisted backlog orders were sufficient to service debt, the stock experienced severe volatility, with intraday declines of up to 16%.

The true impact lies in how this signal transcends a single company to affect the entire NeoCloud cohort’s financing conditions.

As the most aggressive expanders in the current compute infrastructure wave, NeoCloud firms rely heavily on balance sheet arbitrage. They collateralize NVIDIA GPUs with Wall Street banks and private credit providers to secure multi-billion-dollar loans, purchase chips, then lease them to frontier AI startups—repeating this cycle to amplify debt leverage.

When the lead firm’s CDS spikes to 881 bps and bond yields reach 15%, the temperature of the entire pipeline has changed. Lenders are re-evaluating credit risk across the entire NeoCloud segment, dramatically increasing risk compensation for new loans and tightening collateral ratios and covenants. For compute cloud vendors dependent on continuous rollover financing and bearing capital expenditures for next-generation chips, a sharp rise in funding costs directly narrows previously available arbitrage margins.

On one side, frontier labs showcase grand visions to the outside world; on the other, intermediaries responsible for buying chips and leveraging capital are charged distress-level default premiums by capital markets. Placing these two images side-by-side on the same trading desk fundamentally alters buyer risk preferences.

Public Statements from Broadcom and Intel Leadership

The blunt truth about capital tightness was laid bare before global investors at the DBS Bank Asia Future Frontiers Conference held in Singapore on October 12.

Chen Fu-Yang, CEO of Broadcom, delivered a widely quoted assessment: he believes the biggest constraint for the AI industry in the coming years may not be capacity, but capital.

Chen reviewed the supply chain over the past two years—shortages of high-bandwidth memory, limited wafer capacity, slow grid access—issues that have largely been resolved. Wafer shortages are expected to ease within two to three years, and new power generation facilities are coming online. But investor patience and funding costs are becoming hard constraints. He stated outright that many investors entered with extreme expectations, resembling gambling rather than rational investing. Meanwhile, actual business efficiency in applying AI remains insufficiently proven.

Intel CEO Pekka E. K. Liu expressed similar views at the same event. He emphasized that the most urgent challenge today is the rising cost of capital and increasingly stringent scrutiny of return on investment (ROI).

Both semiconductor leaders acknowledged the genuine demand for compute power but jointly pointed to a critical precondition: sustained growth in the coming years hinges on downstream customers proving tangible commercial returns from AI. If AI penetration in industries falls short of expectations, upstream compute infrastructure procurement rhythms may adjust.

DBS CEO Chen Shushan offered a financial perspective on Bloomberg Television, stating that risks associated with AI-related investments are being repriced. Capital providers now demand higher return compensation and stricter transaction structures. DBS has led approximately $20 billion in compute-related syndicated loans this year but retains only minimal exposure in each deal.

The capital faucet hasn’t shut completely—it’s just shifted direction and risk appetite.

Revised Expectations for Large Model Revenue

Meanwhile, subtle adjustments are emerging within the frontier large model narrative, easily overlooked.

First, revenue transparency. OpenAI’s updated data for institutional investors shows annualized revenue of approximately $50 billion as of the end of September—below the widely circulated $70 billion figure. The downward revision stems from the new accounting method excluding revenue share portions from cloud partners.

This move triggered cross-market comparisons. If OpenAI adopts a net revenue model excluding channel splits, Anthropic’s gross revenue—including Amazon and Google Cloud’s distribution—loses comparability. When core industry metrics shift from upward revisions to downward compression, valuation frameworks face recalibration.

On technical development pace, frontier labs themselves are signaling slowdowns. Early September, Sam Altman, CEO of OpenAI, explicitly stated the company will not pursue an IPO within 2026, citing the need for more time to address model safety. Dario Amodei, CEO of Anthropic, issued a concurrent call urging the entire industry to voluntarily slow capability advancement to allow space for safety alignment research.

These cautious stances follow recent abnormal incidents exposed in safety testing. As industry leaders begin calling for moderation, the capital market’s linear assumption of endless accelerated compute demand loses emotional support.

The Balance Sheet Behind the Impressive Numbers

Returning to Anthropic’s $16.8 billion performance report.

Based on the $65 billion annualized run rate disclosed by end-July, Q3’s $16.8 billion met the minimum threshold. Adjusted gross margin remained around 85%, adjusted operating profit margin exceeded 11%, and both operating and free cash flow turned positive.

Yet the market’s focus shifted to growth slope.

From Q1 to Q2 2026, revenue grew approximately 143% quarter-over-quarter. From Q2 to Q3, the growth rate slowed to 46%. The transition from exponential acceleration to deceleration occurred within just two quarters.

The annualized run rate for Q3 was approximately $67 billion—slightly above the $65 billion disclosed in late July—indicating that monthly net additions from August to September may already be leveling off. The core assumption behind the sky-high valuation multiples has always been hyper-steep exponential growth. Once the slope changes, the valuation model’s foundation faces re-calibration.

At the current $67 billion annualized revenue base, a $2 trillion IPO target implies roughly 30x revenue multiple. Five months ago, Anthropic completed a post-money valuation of $96.5 billion in fundraising, overtaking OpenAI in valuation. Months later, planning to enter public markets with a doubled $2 trillion valuation, this leap has caused hesitation among potential anchor investors.

Underwriters are using forward-looking revenue models anchored on 2028 projections. Anthropic’s projected 2028 revenue ranges between $190 billion and $200 billion. Assuming $200 billion in forward revenue, a $2 trillion market cap corresponds to about 10x forward revenue multiple.

Whether public markets will grant full trust to a two-year-forward hypothetical scenario represents the central point of contention. Independent research firm New Constructs’ founder publicly questioned whether sustaining a $2 trillion valuation requires steady-state profitability approaching NVIDIA’s levels. The firm argues that such a high listing valuation reflects primarily liquidity pricing for primary market capital seeking exit, not a true reflection of enterprise value.

Larger financial constraints lie hidden in undisclosed prospectus materials. Reuters reported that Anthropic committed to investing at least $51.8 billion in AI infrastructure over the next decade, partnering with six compute and cloud platforms. Approximately 80% of contracts are irrevocable or include minimum payment guarantees regardless of actual usage.

Contract breakdowns show $16.12 billion in commitments to Broadcom-related equipment, $11.11 billion with Google, spanning April 2026 to July 2033, with explicit clauses requiring shortfall payments if actual compute consumption falls short. The agreement with Amazon totals around $11 billion, running through April 2036, also includes strict floor provisions. The agreement with Microsoft amounts to approximately $3.14 billion.

Compared to OpenAI’s shared infrastructure model via softbank, Oracle, and others under Stargate, Anthropic bears an exceptionally high individual execution burden. Using a revenue base still in early development to shoulder over $500 billion in rigid payment obligations effectively transfers all future compute consumption risks onto its balance sheet. Should enterprise software procurement budgets fail to expand as expected, these irrevocable floor payments will directly become persistent cash flow liabilities.

Revenue quality details are also under scrutiny. The company’s top two clients account for approximately 24% of revenue, mostly under flexible, non-lock-in contracts. In 2025, about 47% of revenue came from Amazon and Google Cloud channels, with platform split ratios around 20%. This means gross revenue cannot be fully retained as cash reserves. Moreover, both major cloud channels are simultaneously investors, suppliers, and direct competitors in self-developed models.

On product pricing, pressure from global open-source ecosystems and closed-source rivals is gradually materializing. Artificial Analysis data shows Anthropic’s flagship model’s invocation price exceeds twice that of OpenAI’s comparable product. Chinese open-source and open-weight models, leveraging extreme cost control and usability, are prompting overseas enterprises to restructure AI budgets. Software spend analytics firm Ramp’s survey reveals corporate clients are reducing unconstrained model invocations and shifting toward more cost-effective alternatives. As mixed token prices continue declining across the industry, maintaining an 85% gross margin will heavily depend on inference efficiency improvements outpacing price declines.

In enterprise deployment, Counterpoint Research data shows Anthropic captured 31.4% of large language model revenue share in Q1 2026, ranking first in the industry. Institutions including BlackRock and Barclays Bank have integrated the Claude system. However, in consumer traffic, Sensor Tower statistics indicate ChatGPT holds 46.4% of global share, Gemini rose to 27.7%, while Claude accounts for only 10.3%. The underlying competition among large models is intensely dynamic—the window for technological leadership may not cover the depreciation cycle of decade-long compute contracts.

The OTC market has already provided immediate feedback. Hyperliquid’s OTC ticker code for Anthropic, ANTH, has retreated from above the $2 trillion mark down to the $1.89 trillion to $1.94 trillion range.

Anthropic’s reported $16.8 billion revenue and the Nasdaq’s 1.88% decline mirror each other.

One side shows frontier large model companies achieving astonishing commercial expansion—revenue leaping from $4.7 billion to $16.8 billion in half a year, achieving bottom-line profitability and crossing the cash flow breakeven point. This proves genuine demand for advanced intelligence tools exists at the enterprise level, and the direction of technological progress remains intact.

The other side reflects capital market rationalization: $51.8 billion in irrevocable procurement floor commitments, CDS quotes surging to near 881 basis points, compute intermediary lending costs nearing 15%, consecutive delays in data center asset IPOs, and the 5.2% benchmark U.S. Treasury yield pressing upward.

Revenue confirms demand is indeed exploding. But capital markets are voting with their feet—revealing the true cost required to sustain this future.

Original: BlockBeats