Is Google Search a Dead Man Walking? Enter the Dragon, err…Apple Intelligence.

Search Dominance: The Holy Grail of the Internet

For over two decades, Google dominated the way we access the internet. Discovering user intent is the holy grail of internet usage, and Google seems to have found it with its Search. Sometime in 2006, I recall reading the book, “The Search: How Google and Its Rivals Rewrote the Rules of Business and Transformed Our Culture” by John Battelle, published in September 2005. It was a seminal book on understanding Google and Search. In this book, Battelle talks about the “Database of Intentions,” a collection of data reflecting the desires, needs, and interests of users as expressed through their search queries.

Today Google has over 90% market share of all searches on the internet. This is the basis of the US DOJ (Department of Justice) antitrust lawsuit against Google filed in October 2020.

How much is Search worth to Google?

As of last year, $175 billion!

From the time Google started monetizing Search in 2002 (and posted a mere $440 million in revenue that year, all derived from Search), its revenues exploded and grew every single year, except in the last few years. Here is a summary of how much Search brings to Google:

1. Early Years (2002-2010): In the early years, Google's revenue was almost entirely derived from search. For instance, in 2002, the total revenue was $0.44 billion ($440 million)? all of which came from search, making it 100% of the total revenue. In 2010, Google Search’s revenues were $29.32 billion.
2. 2011-2020: During this period, Google's revenue continued to grow, and search remained a dominant source. However, the exact breakdown of search revenue as a percentage of total revenue is not explicitly detailed for each year. It is generally understood that search constituted a significant portion of the revenue, often close to 100%. Starting with $37.91 billion in 2011, the revenues hit $182.53 billion in 2020.
3. 2021-2023: In recent years, more detailed data is available, showing a decline in the percentage of revenue from search as Google expanded into other areas like YouTube ads, Google Cloud, and other services.

2021: Total revenue was $257.64 billion, with search revenue at $162.45 billion, making search revenue approximately 63.05% of the total revenue.

2022: Total revenue was $282.84 billion, with search revenue at $175.00 billion, making search revenue approximately 61.88% of the total revenue.

2023: Total revenue was $307.39 billion, with search revenue at $175.00 billion, making search revenue approximately 56.94% of the total revenue.

Search is so valuable to Google that it paid Apple $20 billion last year for Google Search to be the default one on iOS. I think paying someone 20 bucks when you make 175 is not a bad bargain when that someone has all the mobile users outside your own set of mobile users, and there are only the two of you. Technically, this (“only the two of you”) is not a true comparison as Google just gets licensing fees for the Android OS, whereas Apple gets to reap revenues from the hardware.

Feeding the Database of Intentions: Will it change from a ‘Search Query’ to a ‘Chatbot Prompt’?

Google has steadily expanded its reach into other areas, and for the very first time, there are signs of flattening of revenue growth from Search. Is it a coincidence that ChatGPT from OpenAI was released around the same time that Google’s Search revenues showed flattening? No one can say for sure, but in the future, more and more people will surely turn to Generative AI to express their “desires, needs, and interests,” to use the phrase from my first paragraph above.

Though OpenAI’s GPT and other large language models (LLMs) do not index the web in the same way that search engines like Google do; there is, for the first time in over two decades, a truly viable alternative to how we can interact with the internet. Strictly speaking, “Search” is not the way we access the internet, it’s the browser at the application layer on top of the TCP/IP stack. I view Search as an application on application (AOA), but that’s a moot point, as Chatbots and APIs access LLMs, which can access the internet, all without a browser or the Search box.

For those who would argue that LLMs do not cut it for real-time web searches, I would point to perplexity.ai and the like, which seem to happily marry an LLM with a Web Crawler.

So just when Google had the tiger by the tail (after two-plus decades of incredibly hard work), did the tiger escape? Did the genie return to the bottle? I am speaking of the Database of Intentions—user intent—the holy grail, for which we have to start our search (no pun intended) all over again.

I contend that the rise of Generative AI and the ease with which we can access the LLMs (through nice Chatbots) can wreak havoc on Google’s Search monopoly. 175 billion dollars of revenue will not disappear overnight, but Google I am sure is having sleepless nights about the future of its search. But don’t write Google off so fast. Google has tremendous advantages in AI, and I will write about it in the future. Just to give a little teaser for now: Google has the most integrated tech stack for enterprise AI; their Chips/processors, Cloud platform, LLM, Chatbot, OS, and the Pixel phone.

Enter the Dragon, but is he late to the party?

I am speaking of Apple which threw its hat into the AI ring with yesterday’s WWDC keynote. I watched the entire keynote live. Apple even dared to call its offering, Apple Intelligence. Is AI now Artificial Intelligence, or Apple Intelligence? Get it?

Is Apple late to the game? Some people have opined that Apple is late to the AI party, and it’s already over for them. I think not. Its entry is perfectly timed. Apple’s entry is the beginning of the AI experience for the masses.

At its core, Apple is a hardware company and excels in design and user experience (felt through its products). They can make AI seamless and invisible so that an average iPhone user will reap all the benefits without even realizing that it is Generative AI in the background. Oh, and Apple will make sure those users will hear the phrase, ‘Apple Intelligence,’ repeatedly instead of “Artificial Intelligence”!

You can have the LLMs, Chatbots, Cloud Infrastructure, and Data Centers. But I am the gatekeeper that will let all the users into the playground. I will delight them along the way. I will protect their privacy and will let them play on their device, which has a chip capable of running the LLMs locally, for the most part. When I let them out, I will allow them only into the civilized, safe areas and not the wild jungles of your servers and data centers.

That’s what Apple is saying to its competitors. Note: To be fully transparent, the about quote is mine, and not Apple's. I am expressing its message and tone in my own words in the above quote.

Users don’t want to use LLMs, Chatbots, or APIs; they just want to get things done, and Apple knows this very well. Most users will just want a free, intelligent, executive assistant that will delight them in doing those things. Apple will exactly provide that for them and do it well because of the tight integration between the Operating System (iOS) and Apple Silicon (‘A17 ‘and ‘M’ series of chips).

Is Apple Intelligence inferior to the offerings by others?

Elon Musk couldn’t help himself from negatively opining about Apple Intelligence. He tweeted (or “X” ed) that he will ban all his employees from using Apple AI phones and will require them to check them into a Faraday Cage! Will he require them to remove all Chatbots (except for his xAI-made Grok Chatbot of course) too? This is preposterous! Musk has this ongoing feud with OpenAI and has been critical of Apple for a long time. Is this a case of sour grapes? Does Musk’s visceral reaction tell us something about the future of Apple Intelligence, as in Apple touching a raw nerve here?

Later today Apple is expected to release a paper detailing the architecture of Apple Intelligence. I will surely keep an eye out for that. But from what we know now, Apple Intelligence is the most personalized intelligence we will see. It will be aware of your personal context like no one else would. It will try to do most things on the device (the chip on the device will be capable of that) because it will run the LLM on the chip. The LLM on chip will not be as big as GPT-4, but more similar to the smaller Phi-3, a family of open AI models developed by Microsoft. What Apple Intelligence cannot do on-chip, it will delegate to a secure Private Cloud to run GPT-4 from OpenAI. Apple calls this, ‘Private Cloud Compute’ (PCC). It will ask your permission every time it will have to use this PCC, and frankly, that can get annoying. But Apple is taking the safety and privacy first route, as was expected from it.

So, this is not an integration of GPT-4 into the Operating System as Musk claims; Apple did not do that. This is not Apple being unable to create their own Generative AI, as Musk taunts. On the contrary, this is Apple carefully calibrating their own LLM on-chip and doing it right.

Finally, some commentators have mentioned the fragmentation of Apple’s offerings between the older devices (before iPhone 15 and Intel-based Macs, for instance) and newer Apple silicon-based devices. The non-Apple Intelligence phones and devices, and the Apple Intelligence-infused devices. Again, this is not fragmentation—this is integration at its best. A carefully calibrated and orchestrated integration that Apple has always been good at.

In conclusion, Apple Intelligence is state-of-the-art, first-of-its-kind, personal intelligence. Apple’s big bet can be a game changer in how billions around the world will experience Generative AI.

What does Apple Intelligence have to do with the future of Search?

Everything! When Apple calls its generative AI, Apple Intelligence and positions itself as totally personal intelligence, users’ methods of access to the internet can change dramatically. It can also happen with Chatbots and hybrid search engines (like Perplexity), but Apple alone can influence that change much faster than anyone, and on a scale larger than anyone (with over a billion users).

Google does have advantages in AI. They can do all the things that Apple Intelligence can do, but now the keys (Search) to the kingdom (internet) are no longer exclusive. Moreover, Google does not have the same market share for its own Pixel phone as Apple has because of the Android OS business model. Google licenses its AndroidOS to OEM phone makers and can never be the hardware company that Apple is.

Access to the Internet, Chatbots, LLMs, and hybrids (LLM and Web Crawlers) are now crowding the space, each with its own gates to the Internet. But the biggest threat to Google Search will come through Apple’s influence on how users will do things (which will include internet access and search). Apple might continue to have Google Search as the default search, and Google might continue to pay them. But if how we access the Internet and do things with it is changing, Google Search can be a dead man walking. He will not die in two years, but in ten years, he might.

But we don’t have to write off Alphabet (Google’s parent) as a company. It can do fine without Search dominance, but only if it can replace the $175 billion revenue which can be impacted over time.

Conclusion

Google Search might no longer be the holy grail of the Internet, and the Internet is just a part of the universe. The entire universe now comprises everything digital, and Personal Intelligence is the new holy grail, and Apple seems to be ahead in finding it. Google Search will remain for the time being, but it will no longer be as dominant in our culture as it once was. Alphabet (and Google) does have the advantage of having an integrated AI stack when it comes to Enterprise AI (I will write more on this later), but the massive lead Apple has in the consumer space with the best-integrated stack just got better with the inclusion of Generative AI to its stack.

This might not bode well for Google Search as a product in the future.

Correction (June 13th, 12:07 AM)

In the article above, I mentioned this:

Apple Intelligence is the most personalized intelligence we will see. It will be aware of your personal context like no one else would. It will try to do most things on the device (the chip on the device will be capable of that) because it will run the LLM on the chip. The LLM on chip will not be as big as GPT-4, but more similar to the smaller Phi-3, a family of open AI models developed by Microsoft. What Apple Intelligence cannot do on-chip, it will delegate to a secure Private Cloud to run GPT-4 from OpenAI. What Apple Intelligence cannot do on-chip, it will delegate to a secure Private Cloud to run GPT-4 from OpenAI.Apple calls this, ‘Private Cloud Compute’ (PCC). It will ask your permission every time it will have to use this PCC, and frankly, that can get annoying. But Apple is taking the safety and privacy first route, as was expected from it.

This is not entirely correct. What is italicized is wrong and needs correction. Apple will not invoke GPT for everything that goes to their PCC. My apologies!

From Apple's Machine Learning blog

In the following overview, we will detail how two of these models — a ~3 billion parameter on-device language model, and a larger server-based language model available with Private Cloud Compute and running on Apple silicon servers — have been built and adapted to perform specialized tasks efficiently, accurately, and responsibly. These two foundation models are part of a larger family of generative models created by Apple to support users and developers; this includes a coding model to build intelligence into Xcode, as well as a diffusion model to help users express themselves visually, for example, in the Messages app. We look forward to sharing more information soon on this broader set of models.

So, there are two Large Language Models (LLMs) that Apple has developed.

1. A small on-device LLM that runs on the device chip
2. A larger server-based language model

Apple calls both of them, their "foundation models". Both are ultra-secure. Only when "world knowledge" is required will Apple take the query to GPT (from Open AI), and this is when it will ask for permission explicitly. Apple will try to do most of the work on the device or their Private Cloud Compute (PCC).

So here is the flow perhaps:

Smaller on-device LLM--> (failing which) Larger PCC LLM-->(when World Knowledge is required) GPT,

or

Smaller on-device LLM-->(when World Knowledge is required) GPT.

PS: I don't want to do shoddy work, hence the correction.