Internet Queen Mary Meeker's 340-page "AI Trends Report" PPT

48 hours ago, the release of the Internet Queen's 340-page "AI Trends Report" immediately caused a stir in the industry. After six years, Mary Meeker still proves her prowess. Silicon Valley magnates have been studying it sleeplessly, all night long!

Internet Queen, legendary investor Mary Meeker, makes a comeback!

In the past, when the Queen's "Internet Trends Report" was released, the entire tech world would pay close attention. Silicon Valley titans would lose sleep, studying this top-headline report all night.

After several years of quietude, she is back with a weighty 340-page report.

Report PPT: https://www.bondcap.com/reports/tai

Image

This time, she has set her sights on OpenAI, the rising star of the AI world.

Image

Among founders and CEOs, this report has gone viral.

In this 340-page report, the term 'unprecedented' appears 51 times. The core message is that the AI-driven revolution is pervasive and irreversible, representing both a golden age of abundant opportunities and a 'critical moment' for the singularity!

ImageImage

The Queen Herself Returns After Five Years

Mary Meeker, the renowned Internet Queen.

Previously, she was a member of Morgan Stanley's TMT team. This team held significant weight, having led Netscape's IPO, which directly ushered in the 1996 internet boom!

Image

In 1996, she released the first edition of the "Internet Trends Report" – a 322-page in-depth analysis exploring the potential of the web.

Since then, she has annually published internet trends reports, and early on predicted the following trends.

a. 1996 → Online population surge (10x growth by 2000)

b. 2000 → Online advertising > print advertising (occurred in 2004)

c. 2008 → Mobile dominance (occurred in 2014)

In 2010, she joined Kleiner Perkins, leading their growth fund, and subsequently invested in Facebook, Spotify, Square, Twitter, and Snap, witnessing their brilliance.

In 2018, KP Growth Fund spun off and was rebranded as BOND Capital.

In 2019, she paused writing the "Internet Trends Report" to focus on founding BOND.

Now, after a six-year hiatus, she has returned with the weighty 340-page "AI Trends Report 2025"!

In this report, quotable golden phrases are abundant, such as the examples below.

Like electricity and other general-purpose technologies of the past, AI and cloud computing data centers represent the next phase of industrialization.

——Brad Smith, Microsoft Vice Chairman and President

This time is different. We can make up for it through scale, and in the future, we will find ways to monetize users.

——The three most dangerous sentences in business

Now, let's look at the core content of this report.

Image

AI Inference Costs Have Plummeted to Rock Bottom

Currently, the speed at which AI usage costs are declining is unprecedented.

While the cost of training a top-tier model has soared to $1 billion, a 2400-fold increase in the past 8 years, inference costs have plummeted by 99.7% within two years.

ImageImage

(The above figures are calculated based on the cost per million tokens)

Consequently, the rules of the game across the industry are being reshaped.

The per-token power consumption of the 2024 Blackwell GPU is already 105,000 times lower than its predecessor, Nvidia's Kepler GPU launched in 2014.

The costs of top AI models have since fallen rapidly.

Therefore, lightweight models focused on customized scenarios will directly outperform giants like OpenAI, who burn through massive amounts of capital.

Image

Meanwhile, Google's TPUs and Amazon's Trainium are also being developed at scale for cloud services, and progress is rapid.

In Mary Meeker's view, these are not fringe projects, but foundational bets.

Image

OpenAI's Valuation Myth: Supported by Indian Users?

This report also reveals an unexpected fact—

India has become the second-largest global market for ChatGPT applications!

Yes, the largest monthly active user base for ChatGPT now comes from India, accounting for 13.5%, directly surpassing the US domestic 8.9%.

Image

Even more surprisingly, India is also the largest market for DeepSeek.

Among DeepSeek's global mobile app users, 7% are from India.

One reason is the low cost of internet data per GB, which enables 67% of the population to go online.

Image

Another highlight is that Indian users are far more price-sensitive than American users, so any competitor that can break through with low costs has a very strong entry point.

Image

Chinese Models Swallowing Market Share, Is the Cash-Burning Model Failing?

Meanwhile, Chinese models are rapidly capturing market share at extremely low costs, at an unprecedented speed.

Currently, OpenAI's valuation has reached $300 billion.

However, Mary Meeker sharply points out: Compared to its revenue, OpenAI's valuation seems 'too expensive'!

Image

After all, these three top AI companies—OpenAI, xAI, and Anthropic—have raised a staggering $95 billion, reaching astronomical levels.

However, their combined projected annualized revenue is merely $12 billion.

Specifically, OpenAI's annualized revenue is approximately $9.2 billion, yet its valuation has surged to $300 billion, making its valuation 33 times its revenue.

Meanwhile, Perplexity's valuation/revenue multiple has reached approximately 75 times.

Image

Compared to Chinese models, especially the low-cost open-source alternatives provided by DeepSeek, this bubble is on the verge of bursting.

It can be said that their fundamental business model is rapidly being uprooted by Chinese players.

Mary Meeker's comment on this is illuminating: 'Investors, please only invest the amount you are willing to lose.'

Putting all your eggs in one basket is risky because everything is rising now and seems foolproof—until one day the situation suddenly reverses.

Currently, Chinese AI models are rapidly catching up, with performance close to the top US AI levels, but at incredibly low costs.

For example, Baidu's ERNIE 4.5 Turbo costs only 40% of DeepSeek V3 and 0.2% of GPT-4.5.

This cost comparison is truly stark.

Meanwhile, DeepSeek-R1 has scored a high 93% on the MATH Level 5 math benchmark, very close to OpenAI's o3-mini model's 95%.

Image

At the same time, the installed base of industrial robots in China now accounts for 75% of the global total.

Image

According to Meta CTO Andrew Bosworth, the global AI landscape has now evolved into a full-blown new 'space race'.

It is now widely recognized that AI reflects the comprehensive strength of nations, directly determining future geopolitical influence.

Image

It can be said that the current AI model supremacy battle is essentially a battlefield between China and the United States.

Image

Statistical data from 2017 to 2024 shows that the number of large-scale AI systems released by China is now comparable to that of the United States.

Image

ChatGPT's Unprecedented Surge of 800 Million Users in 17 Months

Of course, no one can deny ChatGPT's previous brilliance.

After all, it reached 800 million users within 17 months, a growth rate that surpasses any technology in human history.

Image

It can be seen that ChatGPT's in-app usage time has doubled within two years.

Among them, the desktop version is where users engage in heavy work, with daily engagement time at 18 minutes, longer than Duolingo's 15.5 minutes.

Image

Furthermore, Google's search market share is being relentlessly swallowed by OpenAI.

Currently, OpenAI's annual search volume is breaking through at a rate 5.5 times faster than Google's. The former took 2 years, while the latter took 11 years.

It can be said that we are gradually witnessing a major shift in the search market.

Image

Currently, AI user growth, usage, and capital expenditure are all experiencing unprecedented explosive growth, achieving the fastest technological penetration rate in human history.

Over the past 1000 years, technological advancements have driven exponential growth in global GDP.

Image

However, will ChatGPT always dominate the market share? The answer is not necessarily.

Perhaps it will become like Google, or be completely forgotten like AltaVista.

Charts show that DeepSeek and Grok have clearly caught up and surpassed.

ImageImage

Where Will the Next Billion-User Market Be?

Will the next billion users be AI-native users?

Ten years ago, Google launched the NBU (Next Billion Users) project, focusing on technology language and low-connectivity accessibility, to enable technology to embrace the next billion internet users.

Today, 'NBU' is shifting from being language-centric to AI-centric.

Low connectivity → Satellite internet connectivity

Browser/app interface → Voice/language interface

Currently, 32% of the world's population—2.6 billion people—are still not connected to the internet.

However, with the growth of satellite-driven internet access, the potential for this group to get online will greatly increase.

Image

Just like the billion internet users from India who bridged the desktop/PC and broadband eras.

The next billion internet users will bypass the application ecosystem and directly enter the agent ecosystem.

At that time, they will bypass browsers and search bars, directly using AI.

Thus, they will completely skip the traditional application layer, and the agent-centric experience will overturn all existing internet technology hierarchies, with the meaning of platforms being dissolved and reallocated.

Image

In summary, today's AI winners are not eternal winners. Infrastructure is undergoing massive changes, and applications will change accordingly.

The only thing known for sure is that we are at the dawn of another AI-driven technological supercycle.

Image

Main Tag:Artificial Intelligence

Sub Tags:AI TrendsGeopolitics of AITechnology AdoptionMarket Analysis


Previous:Can AI "Admit Its Own Mistakes"? Solving the "Rashomon" of Multi-Agent Collaboration, Earning ICML 2025 Spotlight

Next:First Explanation of How LLMs Reason and Reflect: Northwestern University & Google's New Framework Introduces Bayesian Adaptive Reinforcement Learning to Comprehensively Enhance Mathematical Reasoning

Share Short URL