OpenAI has no city walls? The era of AI improvisation software is here, and the U.S. faces the "Browser Curse" again!

If you trade stocks, just rely on the Golden Qilin analyst research reports—authoritative, professional, timely, and comprehensive. Help you uncover high-potential theme opportunities!

(Source: Xinzhi Yuan)

Xinzhi Yuan Coverage

Edited by: Qingqing

【Xinzhi Yuan Insight】OpenAI doesn’t even have a moat! Top analyst Benedict Evans’ judgment: Large models are essentially “commodities.” OpenAI is highly likely to repeat the tragedy of Netscape. 80% of users don’t interact with AI more than a thousand times in a year; Meta is spending 50% of its revenue to buy chips. The “financial gravity” moment in the AI industry is already here.

Don’t keep fantasizing that OpenAI will become the next Google or Microsoft!

In the latest episode of The MAD Podcast, top tech analyst Benedict Evans went head-to-head with Matt Turck in a hard-hitting, no-nonsense debate that lasted a full hour.

Evans drew the conclusion: OpenAI is facing a destructive moat crisis.

Unlike Windows and Google, the characteristics currently shown by foundation models are extremely expensive yet also extremely homogeneous.

Now, OpenAI has 900 million weekly active users, but 80% of users press Enter fewer than 1000 times in a year.

Clearly, the AI industry’s finances are already stretched to the limit.

The vanishing moat: Are large models just “commodities”?

In the history of technology, all monopolies are built on networks.

You use Windows because developers are writing software for Windows; you use Google because its search results are continuously optimized based on what users click.

But in the field of large models, that logic fails.

Evans points out that, at present, there is no evidence showing that a leading model can prevent competitors from building equally good models.

Claude, Gemini, and Llama are frantically taking over the lead positions—every few weeks, the throne is handed over again.

If you were Sam Altman, what you’re holding is a kind of commodity technology,

Evans said. OpenAI has no infrastructure of its own, no differentiated network effects—what it has is only a huge amount of “mindshare.”

It’s exactly like Netscape in 1995.

Back then, Netscape had the best browser in the world and the highest level of attention.

But once the browser itself becomes a piece of basic infrastructure, and when the giants with traffic and systems react, the early pioneers without a moat will be quickly swallowed up.

Now, OpenAI is desperately trying to use this mindshare to exchange for hard assets.

One mile wide, one foot deep: ChatGPT’s real usage dilemma

By scraping and analyzing annual summary data shared by thousands of users on Reddit, Evans found a shocking pattern: the vast majority of users, on average, interact with AI fewer than 3 times per day!

If you entered 1000 instructions into ChatGPT last year, then you’re already among the top 20% of hardcore users globally.

For ordinary people, ChatGPT is only an occasionally useful search enhancer or a copywriting rewriter—not a productivity hub.

The consequence of this low-frequency usage is that fewer than 5% of users are willing to pay for it.

Evans offers a profound insight: simply making the model “better” can’t solve the problem.

If you ask AI 50 questions and it got 10 wrong last month, you still have to check; if it gets 8 wrong this month, you still have to check.

Improving accuracy from 90% to 95% still requires humans to step in and verify.

Unless AI evolves to be absolutely correct, it can’t truly free up value from human workflows.

The “financial gravity” moment: Meta’s gamble to buy chips with 50% of its revenue

The AI race has entered the deep-water zone of “big effort brings miraculous results,” and along with it comes runaway financial data.

Meta’s latest earnings report shows that its capital expenditures are expected to account for more than 50% of total revenue.

Evans was astonished, saying:

This isn’t 50% of profit—it’s 50% of total revenue!

Even as strong as Google and Microsoft, they’re increasing capital expenditures at a double-digit percentage rate.

This scale of investment has already gone beyond building factories—it’s more like building a country’s infrastructure.

But will this investment bring back output in the same proportion?

In the current revenue structure of the AI industry, there is a large amount of recurring income and vendor financing.

NVIDIA sells chips to cloud service providers; cloud service providers then rent compute power to AI startups; and the financing for these startups often comes from these same giants.

This left-hand-to-right-hand game conceals the weakness of real market demand.

A software “improvisation” revolution: Will SAP be replaced?

When discussing AI’s impact on the software industry, Evans proposed a new way of categorizing it: improvisational software vs. institutionalized software.

In his view, the idea that everyone will write their own custom ERP is pure illusion.

Large companies need stable, compliant, standardized systems like SAP and Oracle; they can’t hand everything over to a casual AI agent to randomly run wild.

In the past, many mundane business processes were too small in scale to justify hiring programmers to develop dedicated tools, so people had no choice but to fight it out with Excel and email.

Now, AI has driven code costs down to nearly zero, and AI is turning a large amount of non-standard work into software.

This doesn’t mean the software industry will shrink. On the contrary, according to Jevons’ paradox, when the utilization efficiency of a resource increases, the total demand for that resource actually increases.

We will end up with more software, not less. Evans believes. But the premium power of software companies will shift from “writing code” to “a deep understanding of business pain points.”

If you’re just a simple database wrapper, you’ll quickly drown in the AI code flood.

1997 vs 2026: AI’s “Netscape moment”

Looking back, the internet in 1997 was right on the eve of its craziest period.

At the time, people predicted e-commerce, video conferencing, and mobile internet—but they absolutely never predicted Uber.

We know AI is important, and all the giants are throwing money into it, but we’re still using the mindset of moving PDFs online to do AI social.

Truly AI-native applications have not appeared yet.

OpenAI is trying to find new growth points through agent tools like OpenClaw, but that puts it directly in the line of fire of Google and Apple.

When AI tries to take over users’ desktops and inboxes, privacy, permissions, and control of the system will become a gap harder to cross than algorithms.

In this wave, there will definitely be companies that make a lot of money, but it may not be because their algorithm benchmarks are high.

Reference materials:

https://x.com/mattturck/status/2034661487445258382

https://www.youtube.com/watch?v=jH2ZIUKvazU

A massive amount of information and precise analysis—exclusively on Sina Finance APP