AI leader Anthropic reveals the harshest truth of the entire AI industry!

A few months ago, Anthropic was still benchmarking itself against SpaceX, touting a mythical $2 trillion valuation. But as their roadshow approaches, not only did trading prices on secondary platforms fall below $1.7 trillion early on, but many institutional buyers privately stated: if the final listing price isn't cut below $1.5 trillion, nobody is willing to step in at such high valuations.

Strangely, Anthropic had just revealed an impressive card: $16.8 billion in revenue for Q3 alone, with its first adjusted profit and positive free cash flow.

With such impeccable results, why did the capital markets still throw cold water on them? Because beneath the dazzling numbers, the market smelled a hint of financial embellishment:

• So-called gross profit calculated without deducting the costs of new model training;
• Layered, inflated revenue figures due to gross accounting methods;
• The harsh truth of growth deceleration—quarterly growth rate slashed to one-third.

This article is hardcore. It's not one of those three-minute overviews or clickbait news headline recaps that rush to a conclusion. It's a deep dive, dissecting layer by layer the commercial and financial truths after thoroughly reviewing the IPO prospectus and financial details.

Today, we'll focus on three main questions:

• First, why is an AI company that just turned profitable actually seeing its valuation cut?
• Second, the AI industry’s “Profit Impossible Triangle”: model companies, compute power giants, enterprise clients—all want a bigger share of the pie. But profits are finite; who earns the most, and who gets squeezed out?
• Third, what financial maneuvers are Amazon and Google using in their reporting to cover for Anthropic at all costs?

Bookmark and like this now. By the end, you'll know whether you should get involved in this $2 trillion IPO and why your AI stocks could be dramatically impacted by the Anthropic saga.

What does a $2 Trillion Valuation Require?

Let’s talk valuation.

First, let's be clear: We’re not judging whether Anthropic is a great company. We're assessing whether the $2 trillion IPO price has sufficient fundamental support.

Here’s the valuation trajectory:

• In February this year, Series G funding valued the company at $380 billion;
• By May, Series H saw it jump to $965 billion;
• In less than four months, the valuation rose to roughly 2.5 times the previous number;
• Now, IPO expectations are set at $2 trillion—over double the Series H figure.

The valuation logic: They forecast 2028 annual revenues of $190 billion to $200 billion.

A $2 trillion valuation divided by $200 billion revenue gives a forward price-to-sales ratio of about 10x. For a tech giant maintaining ultra-high growth, a 10x forward P/S isn't unreasonable.

But there's one core premise for this math to hold: Ultra-high growth must be sustained!

This is exactly where problems arise.

Let's look at the latest quarterly data:

• 2025: full-year revenue of $4.59 billion;
• Q1 2026: $4.7 billion revenue, one quarter almost matches the previous year;
• Q2: $11.6 billion revenue, quarter-on-quarter growth of 147%;
• Q3: $16.8 billion revenue, sequential growth slows to 45%.

Objectively, 45% quarterly growth is still good. But for capital markets, high valuations are priced for acceleration.

From 147% to 45%, growth slowed by two-thirds in a single quarter. With a $2 trillion target, the market’s main focus is this downward momentum.

Here is another key detail:

Q3’s $16.8 billion, annualized, is about $67.2 billion. Back in late July, management’s public annualized guidance was already $65 billion—a mere 3% difference.

What does this indicate? Either there was almost no growth July to September, or that $65 billion figure was inflated. Either way, the much-anticipated $100 billion annual run rate isn’t happening short term.

The Scale of Revenue: First, Look at the Accounting Basis

Besides the slowdown in growth, a deeper issue is the lack of transparency in the financial definitions used by leading AI unicorns.

Mentioned frequently is ARR (Annualized Recurring Revenue)—essentially, multiplying the latest monthly revenue by 12 and assuming customers will keep renewing. Yet, the way different companies confirm this varies widely.

For example, Anthropic sells models via Amazon AWS and Google Cloud, using “gross method” accounting; while OpenAI uses the “net method,” deducting channel commissions first, then counting the remainder as revenue.

It’s the same business on paper, but accounting differences alone can make a 30%–40% difference in reported revenue.

OpenAI previously published several different annualized revenue figures, causing volatility in the secondary market. Now, the market is more skeptical of this type of company-reported operational data.

High Gross Margin Masks Persistent R&D Investment

With potentially inflated revenues, profit considerations become even more realistic.

Anthropic externally touted an 11%+ adjusted operating profit margin in Q3 and an 85% adjusted gross margin.

But this 85% gross margin does not account for the cost of new model training. In the AI space, pre-training and constant iteration of models are not one-off capital expenditures, but must be sustained R&D investments.

Under strict GAAP requirements, Anthropic’s operating loss in 2025 is $8.06 billion, with a net loss of about $42 billion.

Even after excluding $34 billion in non-cash expenses, the business is still facing significant operating losses.

Finally, one structural risk that's easiest to overlook: its deep binding with compute power giants.

Amazon and Google are not just shareholders and channels—they’re the fundamental compute suppliers and, in some ways, potential competitors in the ecosystem.

Over the next decade, Anthropic is contractually committed to purchase at least $518 billion in cloud and compute power from its main partners, with about 80% of that being irrevocable.

This creates a financial mismatch:

Outflows are rigid—contracts must be fulfilled; inflows (revenues) are flexible, with customers paying as they go and not obligated to renew.

During an industry boom, this model allows rapid land grab; but if downstream monetization slows, these fixed spending commitments create persistent cash flow pressure.

So why is the market re-evaluating the $2 trillion target?

Because supporting such a scale requires extremely harsh conditions: either demonstrating solid GAAP-positive profits in the formal prospectus, or a quarter-on-quarter rebound of 50%+ in Q4 growth.

If neither happens, a more realistic scenario is to adjust the valuation down from $2 trillion to between $1.4 and $1.7 trillion—a 15%–30% discount on prior expectations.

This price range leaves about 50% upside for Series H investors, makes it easier for underwriters to get approval, and matches strategic shareholders’ expectations.

Ultimately, business judgement boils down to this:

Whether Anthropic is a highly promising tech company is distinct from whether $2 trillion is a rational price.

No matter how good an asset is, if you overpay, it’s a high-risk investment.

The AI “Profit Impossible Triangle”

Zooming out from just Anthropic and looking at the entire AI supply chain, you’ll notice a more counterintuitive phenomenon:

Over the past three years, Wall Street has priced nearly every key link in the value chain as a “super winner.”

• Model developers, like Anthropic and OpenAI: Want software-level 80%+ gross margins to support trillion-dollar valuations.
• Chip suppliers like Nvidia and Broadcom: Rely on technical barriers and compute scarcity to sustain very high margins and capital returns.
• Cloud infrastructure giants such as Microsoft, Amazon, Google: Spend hundreds of billions building data centers, expecting steady rent from long-term compute leasing.

And what about enterprise clients—the true end users?

They want smarter AI at cheaper prices, keeping productivity gains in their own profit statements.

It sounds like everybody wins. But here’s the question: Who ultimately pays for these excess profits?

This is the AI “Profit Impossible Triangle.”

Model firms want software-like high profit margins, compute players want sustained high returns, end users want low costs and high yields. The same dollar of economic value cannot be infinitely split among all parties.

Same $100 of Value Creation, Profits Can Vary Widely

Why? Let’s run a simple calculation.

Suppose an AI service generates $100 in value for a company. Subtract $40 in essential costs (model R&D, compute depreciation, infrastructure), leaving $60 for stakeholders to split.

But who pockets that $60 depends on who holds pricing power.

Case One: Compute Suppliers are Strongest

If advanced GPUs and electricity remain scarce, Nvidia and the cloud vendors hold stronger bargaining power.

Say the client pays $65 for the service, with $50 going upstream. After costs:

• Upstream earns $20;
• Client keeps $35 as value;
• The model company in the middle gets just $5.

Case Two: Model Companies Dominate

This is the scenario Anthropic wants the market to believe in for long-term profitability at a $2 trillion valuation.

If Claude maintains a lead in reasoning, coding, and agent tasks, becomes deeply embedded in enterprise workflows, and pushes down compute costs, profits tilt toward the model company.

Clients are willing to pay $70, compute costs drop to $35. After deducting other costs:

• Model company earns $25;
• Upstream receives $5;
• Client keeps $30 as value.

Case Three: Model Commoditization and Price Wars

With open-source models and rival competition, companies realize that most tasks can be accomplished similarly by different models—so why pay more?

Now, clients only pay $40. Even if the model company drives compute down to $33, after all costs:

• Model company loses $3;
• Upstream only makes $3;
• Client retains $60 in value.

Of course, these are hypothetical cases—not real financial numbers.

But it reveals the problem:

The $100 in value generated by AI isn’t shrinking—the internal profit division varies dramatically.

Why did everyone get along for the last three years?

Because exploding AI demand massively enlarged the pie, and for a while, growth masked conflicting interests. Now, a new risk is emerging: upstream compute scarcity and downstream model price competition may hit simultaneously.

On one hand, high-end compute remains expensive; on the other, model API prices keep dropping. Model companies, stuck in the middle, might see profit margins squeezed from both ends.

This highlights another critical trend—Intelligence Deflation.

In short: AI keeps getting stronger, but the price per unit of intelligence keeps falling.

• If API prices fall 50%, usage must double just to maintain revenue;
• If prices fall by 70%, usage would need to rise more than 3.3x to offset.

As tech advances, some tasks require fewer tokens, while complex agents may consume more.

So, even if AI handles more tasks in the future, that doesn’t mean model companies’ revenues and profits will keep pace.

That's why Anthropic keeps developing Claude Code to go deeper, striving to become core to enterprise workflows.

In the future, a model company’s pricing power depends not just on leaderboard scores, but on how deeply enterprises rely on its products.

If a company has integrated Claude into its codebase, permissions, and automation, even cheaper alternatives may not be tempting—stickiness and switching costs become Anthropic’s true long-term moat.

But if model capabilities converge and clients can switch to cheaper suppliers anytime, maintaining high model-layer margins gets far harder.

More importantly, model company profit pressure might ripple through the entire AI supply chain. Recall Anthropic’s massive long-term compute contracts: if revenue growth disappoints, the company may cut future procurement, delay expansion, or renegotiate deals.

Cloud vendors would have to rethink data center utilization and ROI; Nvidia, Broadcom, and chip suppliers might face slower order growth ahead.

The entire AI hardware and infrastructure sector could face a valuation reset.

Thus, Anthropic's IPO isn't just about whether it sustains a $2 trillion valuation. It could trigger re-pricing across the whole AI supply chain.

Why Are Amazon and Google Still Investing So Heavily?

At this point, many may ask:

If the market is now skeptical of Anthropic’s valuation, why are giants like Amazon and Google still pouring massive resources in?

Is it just a bet on their technology? Of course not. Anthropic’s valuation moves now affect not only its own investors but directly ripple through to these tech giants' financial reports.

This involves an intriguing phenomenon—self-reinforcing cycles of valuation.

Many assume Amazon and Google only realize gains on Anthropic once the IPO or a sale happens. In fact, under accounting rules, some unlisted equity investments can be periodically re-marked to changing valuations from new financing, with the resulting gains directly boosting current profits.

Do you realize what this means?

In Q1 this year, Amazon recorded a $16.8 billion pre-tax gain in its financials thanks to Anthropic investment.

By Q2, the book value of Amazon’s Anthropic preferred shares was marked up by $50.5 billion. Of the $53.4 billion in Q2 non-operating pre-tax "other income," Anthropic’s investments accounted for the bulk.

Note: this isn’t money Amazon made from sales or AWS cloud—it’s paper gains from Anthropic’s rising valuation. Yet, these still boost Amazon’s net profits and EPS.

Crucially, these impacts have flowed through to the S&P 500’s overall profit data.

According to Reuters’ LSEG estimates:

• In Q2 2026, aggregate S&P 500 constituent profits grew 52% year-on-year;
• But, excluding the revaluation gains from Amazon and Alphabet's investments, that figure drops to 33%.

In other words, a big slice of the S&P 500’s headline earnings growth isn’t from underlying business operations. Investors focusing only on net profits or EPS could easily overestimate operating performance.

There’s also a mutually reinforcing dynamic here.

The higher Anthropic’s valuation, the higher the paper value for shareholders; these investment gains improve reports, in turn boosting market confidence in tech giants’ AI strategies. At the same time, Amazon and Google, as Anthropic’s main financial, compute, and commercial partners, keep deploying resources to expand its business.

This is the self-reinforcing feedback of valuations: the market's optimism about the future ends up impacting today’s earnings and investment actions.

This doesn’t mean the giants are manipulating valuations, but it shows that Anthropic maintaining a high valuation offers them direct financial benefits. Conversely, should the valuation drop, investments could face mark-to-market losses or impairments.

Previously realized paper gains are not permanent.

From Valuation Debate to Credit Risk

Meanwhile, capital markets are reassessing AI investments from another angle—credit risk.

On Thursday, Nvidia-backed GPU cloud company CoreWeave saw its stock drop nearly 15% in a single day; by Friday, it fell another 11.9%, making a weekly loss of nearly 25%. Some long-term bonds now yield nearly 15% in secondary markets.

What does this mean?

Bond investors now demand much higher risk premiums to take on the credit risks of such companies.

While this doesn’t mean Anthropic has funding issues, it signals that the market is scrutinizing the financing capacity of capital-intensive AI companies much more seriously.

The US stock market may pay high prices for growth stories, but bond markets care more about: “Will there be cash to repay at maturity?”

This risk preference shift is also reflected in Anthropic’s IPO valuation debates.

Though $1.5 trillion isn’t the final IPO valuation, the market’s attitude toward high valuations has clearly become more cautious.

So, Anthropic’s IPO is noteworthy not just for its listing price, but because the logic underpinning all AI investment pricing may be changing.

In previous years, Wall Street would look at how much capital cloud vendors spent and how many compute orders they secured, to project future revenue generation.

Now, the market is asking another question:

How much real free cash flow are these expenditures actually generating?

Especially since some tech giants' reported profit growth includes huge unrealized investment gains, investors need to distinguish:

• What profits stem from core operations;
• What are merely paper gains from asset revaluation.

Because valuations may surge with market optimism, but long-term investor returns ultimately must be realized via real cash flows.

Wall Street may not have misread the direction of the AI revolution, but in this technological upheaval, the number of companies that can actually generate sustained high returns for shareholders may be fewer than the market imagines today.

This $2 trillion Anthropic IPO may mark a key juncture where Wall Street begins to rethink AI valuations.

Alright, that’s it for today’s article.

Regarding Anthropic, what valuation do you think it will eventually go public at—$1.5 trillion or $2 trillion? Have your AI stocks been affected by this round of valuation resets? Share your thoughts in the comments.