AI: The Future—or the Next Economic Bubble? Separate Product, Compute and Valuation

First, separate the two IPO claims

Anthropic officially confirmed on June 1, 2026 that it had confidentially filed a draft S-1 registration document. This only means that the company has entered the preparation process for listing; the number of shares to be issued, price, timetable and whether it will eventually be listed have not been finalized at that time.

The situation is different with DeepSeek. As of August 20, 2026, we did not find an official listing document or official IPO timetable on the company’s official website. There can be rumors in the market, but the rumors cannot be written as "DeepSeek is IPOing." If the G70 meets with a company, it should directly request written information on the round, investment vehicle, shareholder rights, liquidity and next round of dilution, and should not assume an unconfirmed IPO as a return.

The underlying industry has strong momentum, but that doesn’t mean every valuation is reasonable.

The Stanford 2026 AI Index pointed out that global private AI investment will increase by 127.5% year-on-year in 2025, and enterprise AI investment will more than double; 88% of organizations in the survey have used AI in at least one business function. These numbers illustrate that enterprise adoption is no longer just an experiment, but they can only prove that funding and usage are growing, not return on investment alone.

Anthropic self-disclosed that the post-money valuation of Series G financing in February 2026 was US$380 billion; in May, Series H had risen to US$965 billion. This is private placement round pricing, not listing market value, and it is not a guarantee of future earnings. The company also said that its annualized revenue run rate in May exceeded $47 billion; investors have to further ask: How much of the revenue can be converted into sustained gross profit and free cash flow?

The real pressure point lies in compute economics

The unit economics of an LLM product don’t just look at the subscription price. Each query may involve GPU, memory, power, network, model routing, security monitoring and data storage. When model prices fall and user usage surges, revenue and computing costs can rise simultaneously.

Therefore, when a company says "token usage growth," it should also answer unit token revenue, unit token cost, customer retention, net revenue retention after discounts, and computing resource commitments. If revenue comes from large promotional discounts, or computing power requires long-term upfront payment, superficial growth may not necessarily have improved capital efficiency.

"Bubbles" can exist in some prices at the same time without denying the technology itself

Railroads, electricity and the internet have all had their share of overinvestments while transforming economies. AI may also satisfy two propositions at the same time: it is a long-term universal technology; some companies, infrastructure or transaction terms may reflect too much future success in current prices in advance.

This is why "being optimistic about AI" does not mean "accepting any AI valuation." Investment decisions require separate assessments of technical feasibility, customer value, business model and shareholder returns.

G70 perspective: When meeting DeepSeek, don’t just ask about model rankings

If it is a second-round financing meeting, the G70 should narrow down questions to four pieces of evidence: calculated costs and supply commitments, sustainable revenue and retention, rights and dilution provisions, and data and compliance risks. Model performance can attract customers, but the return on investment ultimately comes from a company's ability to translate performance into repeatable cash flow.

Answers you should get before the meeting

1. Is this round common stock, preferred stock, convertible bonds, or other instruments?

2. What are pre- and post-investment valuations, liquidation preferences, anti-dilution and information rights?

3. How do revenue, gross margin, and computing costs change with usage?

4. If there is no IPO within two years, what are the realistic liquidity paths for investors?

5. Who bears the main risks of technology, training materials, chip supply and export restrictions?

Important note

This article is for general information and educational purposes only and does not constitute a recommendation for any company, security or private equity investment. The conclusions about DeepSeek IPO are the search results of public official information and do not mean that the company has no private plans.