HomeArticle

Anthropic Goes Public: Three Predictions

大G看科技2026-11-18 11:41
The article forecasts three phases of Anthropic's post-IPO trend: rise, fall and refinancing.

On November 19, Anthropic will be listed on the Nasdaq. It will issue 825 million shares at a price range of $70 to $78. Investment banks are already discussing raising the pricing above $80, with the fully diluted valuation standing at $1.5 trillion to $1.6 trillion. Combined with the private placement, the total capital raised will reach $90.4 billion, the largest in history.

01

What Cards Does Anthropic Have in Hand

It issues 825 million shares at a price range of $70 to $78, with a fully diluted valuation of $1.4 trillion. The subscription is nearly 7 times oversubscribed, with multiple orders worth tens of billions of dollars, among which Altimeter alone subscribed for $5 billion. The underwriters also hold an over-allotment option for 124 million shares.

The $26 billion concurrent private placement was purchased by seven suppliers, accounting for 40% of the offering size. Google and Nvidia each invested $10 billion, AMD $2.5 billion, Broadcom, Micron and Microsoft each invested $1 billion, and TSMC invested $0.5 billion. The equity structure is also clearly defined: Amazon holds a 14.7% stake, which will rise to 23.8% after all convertible bonds are converted; Google holds a 13.7% stake, and the two together hold about 28% of the shares. The founders hold 50.1% of the voting rights through Class F shares.

This is the IPO with the most sufficient ammunition in history, and also the one with the smallest floating share capital. The combination of these two characteristics basically determines the initial trend after listing.

02

Prediction 1: The Price Will Rise on the First Trading Day

The reason lies not in the fundamentals, but in the structure. The 825 million shares to be issued only account for 6.3% of the total share capital of 13.1 billion shares. After deducting the part locked by the $26 billion private placement, even fewer shares can be traded on the first day. The trading pool is so small that any slight buying order will push the price up.

Underwriters also have the incentive to support the price. With the green shoe option of 124 million shares in hand, if the price falls below the offering price on the first day, they will enter the market to buy to support the price, so that the subsequent over-allotment shares can be sold. The full exercise of the green shoe option itself is a bullish signal.

Suppliers will not dump their shares. The $26 billion they invested was used to purchase non-voting Class N shares, and their motivation is to lock in order supply rather than earn price spreads. Another detail is that investment banks give priority to allocating shares to existing shareholders, so retail investors get very limited shares, and they can only increase their positions by chasing prices in the secondary market.

Another signal has already emerged. The pricing is expected to exceed $80, which is higher than the upper limit of the price range. The fact that underwriters dare to raise the price indicates that the order book is thick enough for them to select clients.

The only counter factor is that the AI sector is falling, and the Nasdaq once dropped by more than 5.5% in a single day. However, the 7 times oversubscription is a solid signal. Given such strong demand, I believe the probability of breaking the offering price is very low. Three indicators can be used to verify this prediction: the first-day increase, the turnover rate, and whether the green shoe option is exercised.

03

Prediction 2: The Price May Fall Below the Offering Price Within Half a Year

The first trading day is driven by sentiment-based pricing, while the half-year period is driven by performance-based pricing. Let's look at the precedent first: SpaceX listed in June this year, with its offering price corresponding to a valuation of $1.77 trillion, and its intraday market value once hit $2 trillion, but it fell back to $1.4 trillion after the release of its first financial report. The first-day premium was completely erased within one month.

Anthropic is facing the same situation. Calculated based on a valuation of $1.5 trillion and an annualized revenue of $81 billion, its price-to-sales ratio reaches 18.5x. This multiple is based on the premise that its revenue will continue to double.

But its revenue growth is decelerating. The quarter-on-quarter revenue growth reached 144% in the second quarter, but dropped to 47% in the third quarter. Its single-quarter performance turned from a profit of $974 million to a loss of $567 million. The unit price is also collapsing: the price of tokens with the same level of intelligence has dropped by 97% in two years. With the unit price dropping by 97%, the usage volume needs to increase 33 times to keep the revenue flat.

However, its costs are locked in. In the third quarter, its computing power and infrastructure expenditure reached $9.86 billion, accounting for 58% of its revenue. Of its $900.9 billion in committed payments, $70.7 billion will mature within one year, and the rest will be paid off by 2045. In addition, there is $6.5 billion in equity incentive expenses that will be amortized into the income statement over years. Restored according to the notes, its gross margin of full computing power in the first half of the year is 45.1%, while that of software companies is around 80%.

The time node is also clear. According to the 180-day convention, the share lock-up will be lifted around May 2027, and two financial reports will be released before that, with the first one coming in February or March next year.

If in that financial report, the proportion of computing power expenditure in revenue still remains at 58%, and the adjusted operating profit margin still stays at 15%, the 18.5x price-to-sales ratio will be difficult to sustain. The verification indicators include: operating profit margin, the proportion of computing power expenditure in revenue, and the trading volume before and after the lock-up expiration.

04

Prediction 3: Another Round of Financing Is Highly Likely Within a Year and a Half

This prediction is the easiest to verify and the easiest to be ignored, and it is the most certain one among my three judgments. This time, the net capital raised from the IPO and private placement at the median value is $84.6 billion, while $70.7 billion of payments will mature within one year. To break it down: the IPO part raises about $58.6 billion, and the private placement part raises $26 billion, which will not enter the public market but go directly to the company's account. The raised capital is only enough to cover the company's "rent" for one year.

What will cover the bills for the second year? Revenue growth. But revenue has to deduct the computing power cost, and the net contribution that can actually be used to pay the bills is limited.

The prospectus states that the company needs to pay about $450 billion by the end of 2030, which averages $90 billion per year for five years. However, its current annualized revenue is only $81 billion. In the past month, its procurement commitments to Microsoft and Google have been raised together, adding $11.4 billion more -- the suppliers are not only buying shares, but also receiving larger orders.

Look at another detail. In May this year, it just completed its Series H financing of $65 billion at a valuation of $965 billion, and launched the IPO six months later. There was no other public financing in between, indicating that its capital consumption speed is faster than its capital raising speed. With this $84.6 billion raised, it cannot last until 2028 at the same pace.

With an annual bill of $90 billion, how can it be paid with an annual revenue of only $81 billion? The most likely answer is refinancing, through bond issuance, additional share issuance, or another round of share purchase by suppliers. This time, the $26 billion private placement has proved that suppliers are willing to buy its shares, and the next round of financing will most likely come from them. The verification indicators include the financing announcement in 2027, bond issuance, and whether suppliers continue to raise their procurement commitments.

05

The Three Predictions Share the Same Logical Foundation

The three predictions seem independent, but they are actually connected as one thing. The capital source of the AI industry is shifting from venture capital to the balance sheets of suppliers. This statement can deduce the above three predictions.

With suppliers locking up their shares and the small floating share capital, the price will rise on the first trading day. The valuation relies on revenue growth, which is dragged down by the falling unit price, so the price will fall within half a year. The bills are rigid while the revenue is flexible, so another round of financing is needed one year later.

This logic does not only apply to Anthropic. Amazon is its largest shareholder and also its largest computing power supplier; Google is its second largest shareholder and also its TPU supplier. The list of investors is exactly the list of its procurement partners.

The same situation applies to Nvidia. Its accounts receivable reached $81 billion, more than doubling since the beginning of the year; its days sales outstanding rose from 45 days to 64 days; and its off-balance sheet supply commitments reached $390 billion.

Nvidia's interim CEO himself said that the growth rate of some emerging cloud vendors and cutting-edge AI labs has exceeded the level that their balance sheets can support. These words from the seller carry more weight than any short-selling report.

The IMF also wrote the same point in its October Global Financial Stability Report: if AI investment or profitability weakens, the prices of related assets may correct, and the concentration of leverage and risk exposure will amplify this process.

If venture capital loses money, the loss is borne by LPs, which is a private loss invisible to the market. If suppliers lose money, the loss falls on accounts receivable and inventory, which needs to be disclosed, audited and accrued every quarter.

The same risk is moved to a position that must be recorded publicly. This is the common underlying logic of the three predictions.

Conclusion

What if the predictions are wrong? If the price breaks the offering price on the first day, it means the 7 times oversubscription is fake. If the price does not fall within half a year, it means the computing power cost can really be reduced. If no financing is needed for two years, it means the revenue has caught up with the bills.

Among the three predictions, the third one is the most certain. The first two depend on market sentiment and performance delivery, with elements of luck; the third one is pure arithmetic. The bills have been signed, but the revenue has not been generated. The signed contracts will not disappear, while the ungenerated revenue may never come.

Let's wait and see.

This article is from the WeChat Official Account "Big G Tech Watch", written by Big G Tech Watch, and authorized for release by 36Kr.