Author: Merkle3s Capital; Source: X, @Merkle3sCapital

Google’s market capitalization stands above $4.6 trillion, once surpassing Apple. The reason is not complicated - it fights on six fronts at the same time, and the winning potential, rhythm and return cycle of each are completely different.

Search, the cash cow, is being backfired by AI. The cloud business has turned from a loss-making black hole into a growth engine. Large models have gone from falling behind to catching up. Self-developed chips have been able to challenge NVIDIA. Quantum computing is betting on 2029. Waymo and YouTube represent the future and hold on to the present. If you break down these six lines, you will find that this company is actually six companies.

In this article, we will cut Google into six pieces and evaluate its moat strength, challenges and fulfillment time piece by piece. Finally, let’s return to a core question -Do these six lines add to each other or drag each other down?

Search advertising: The ultimate cash cow, but the cracks are emerging

Google search remains the single most profitable product on the planet. The global search market share is over 90%, and the mobile terminal is as high as 93.89%. This dominance has supported Alphabet’s total revenue of US$402.8 billion in 2025, of which Google’s advertising business contributed approximately US$294.7 billion throughout the year. Search advertising is Alphabet’s absolute ballast.

In Q4 of 2025, Google Search & Other increased by 17% year-on-year, and Google Services’ full-quarter revenue was US$95.9 billion.

Everything looks great—but beneath the surface, the very foundations of search are being rocked by a wave of AI of its own.

AI Overviews, launched in May 2024, has seen coverage climb from 16% to nearly 60% of search results by 2025. Google officials said that queries showing AI Overviews increased by 10%, meaning that AI summaries made users ask more questions. But Seer Interactive’s data is painful – when AI Overviews appears, the natural search click-through rate plummets from 1.76% to 0.61%, a drop of 61%; the paid search CTR also drops from 21.27% to 9.87%.

This is a typical "internal subversion paradox": if Google does not do AI summarization, it will be taken away by ChatGPT and Perplexity; if it does AI summarization, its advertising space and publisher ecosystem will begin to shrink. It is a dilemma!

There is also movement at the antitrust level. In August 2024, federal judge Amit Mehta ruled that Google constituted an illegal monopoly in the general search and search text advertising markets. In its relief ruling in September 2025, the court did several things -Ban exclusive default search agreements, force the sharing of some search data to qualified competitors, and establish a technical committee to monitor compliance.

But what is more noteworthy is what the court did not do - it did not require the sale of Chrome, it did not require the divestment of Android, and it did not prohibit Google from continuing to pay device manufacturers (as long as the agreement is not exclusive). The judge clearly wrote: The rise of generative AI has changed the competitive landscape and weakened the need for structural breakup.

Google lost the lawsuit and won the war.

In terms of competitive threats, Perplexity’s latest valuation has reached US$20 billion, and ChatGPT Search is trying to divert the market by leveraging OpenAI’s 200 million monthly active users. However, between 2024 and 2025, Google’s search market share has barely moved at all. This is the result of triple locking in user habits, network effects and distribution channels.

The search moat is still the deepest of the six lines, but it is no longer inerodible.

AI Overviews is a double-edged sword - the query volume has increased, publishers have clicked down, and the advertising monetization efficiency is "comparable" to traditional results, but it is no longer leading. The increase in CPC to $5.26 is also driven by the compressed advertising space. In the short term, the fundamentals are stable, but in the long term, changes in the AI ​​search paradigm will force Google to revolutionize itself while maintaining a closed loop in the advertising ecosystem.

AI large model: The pursuer’s counterattack is still short of success

When Gemini 1.0 is released at the end of 2023, the industry's evaluation is basically "OpenAI is still far ahead." Two years later, Google has evened the technical gap and even surpassed it on multiple benchmarks.

The release of Gemini 2.5 Pro in March 2025 is a turning point. SWE-Bench Verified got 63.8% (code repair), AIME 2024 mathematics competition 92.0%, GPQA Diamond graduate-level scientific reasoning is ahead of GPT-4o and Claude 3.5 Sonnet.

The context window of 1 million tokens is nearly 8 times larger than GPT-4o (128K). Gemini 3 Pro in November 2025 further pushed GPQA Diamond to 91.9%, Deep Think mode reached 93.8%, and WebDev Arena ranked first (1487 Elo).

Gemini 2.5 Pro has an input price of $1.25/million tokens and an output of $10/million tokens - half the price of GPT-4o, but the context window is 8 times larger. This is a differentiation card played with both price and length.

But a beautiful benchmark does not mean the market accepts it. In the consumer AI market, ChatGPT still dominates everything with a 60% share, while Gemini only has 13.5%, although its monthly active users have reached 750 million.

The enterprise-level LLM market is even more embarrassing - Anthropic leads with 40% share, OpenAI 21%, Google 21%, which has climbed rapidly from 7% last year but has not yet secured the second place.

The commercialization figures for 2025 are pretty solid. Gemini’s annual subscription revenue is approximately $1.2 billion, Gemini Enterprise has 8 million subscribers, and API calls will double to 85 billion by August 2025. The Gemma series on the open source side is also running - Gemma 3 27B can run on a single GPU/TPU, and the API input price is only $0.08/million tokens, which is directly comparable to Llama of the same size.

DeepMind’s scientific research narrative is tougher. AlphaFold 3 allowed Demis Hassabis and John Jumper to win the 2024 Nobel Prize in Chemistry. The paper has been cited more than 15,000 times. Isomorphic Labs directly used this technology for drug discovery. This is a story that OpenAI and Anthropic cannot tell.

Model capabilities have caught up, but distribution advantages have not yet been realized into market share. Android’s 3 billion devices, Chrome browser, Gmail’s 1.8 billion users, and the entire Workspace ecosystem—this is Google’s strongest card on paper, but users’ brand preference for AI assistants has been locked in by ChatGPT. To truly make a comeback, what Google needs is not a higher benchmark, but Gemini's default experience on Android and workflow embedding in Workspace, turning "useful" into "fun to use."

AI chip: an underestimated hidden barrier

If Gemini is Google’s AI weapon on the stage, then TPU is the trump card hidden under the desktop. This is a piece of the puzzle that is severely undervalued by the market.

Ironwood (TPU v7), released in April 2025 and GA in November, has an FP8 computing power of 4614 TFLOPS, which is almost the same as NVIDIA B200’s 4500 TFLOPS. HBM has the same capacity of 192 GB and bandwidth of 7.37 TB/s, which is slightly lower than the B200’s 8 TB/s by 9% but there is no generation difference. In terms of single-chip performance, Google is challenging NVIDIA's flagship head-on for the first time.

What’s more critical is the system-level advantage. The Superpod of the 9216 chip is interconnected through Google's self-developed optical circuit switching (OCS), with a total computing power of 42.5 ExaFLOPS, far exceeding a standard GPU cluster. Trillium (TPU v6e) will be released in May 2024, with a peak computing power of ~918 TFLOPS, which is close to the same as H100, but the pricing is a dimensionality reduction blow.

TPU v6e has a three-year commitment price of US$0.39/chip-hour, while H100 is priced at about US$2.75/GPU-hour on demand on mainstream cloud platforms - TPU is about 85% cheaper. This is not a marketing rhetoric, but the real cost of Broadcom’s foundry + self-developed OCS + end-to-end vertical integration.

It is estimated that TPU’s TCO is about 40% lower than NVIDIA GPUs, and OCS optical interconnect saves Google more than $3 billion in data center network costs annually. This means that running the same AI training task internally at Google is much cheaper than renting it from NVIDIA - this structural cost gap directly translates into the iteration speed of Gemini training and the pricing space of Google Cloud.

Hard-software collaboration is another hidden barrier. JAX + XLA compiler + TPU chip + Gemini model, this is the only full-stack closed loop that Google can run through. OpenAI is inseparable from Microsoft Azure and NVIDIA, and Anthropic also rents computing power from others. Only Google builds its own from silicon wafers to models.

In November 2025, Anthropic signed the largest TPU order in Google’s history—hundreds of thousands of Trillium units in 2026—which in itself is the strongest endorsement of the TPU ecosystem.

The risks are also obvious. CUDA’s developer ecosystem is too mature. Although JAX/XLA is expanding, there is still a gap; TPU can only run on Google Cloud and is highly locked-in; NVIDIA’s Blackwell series is still accelerating iteration, and B300 is already on the way. But taken together,TPU is the moat that is most easily overlooked in the market but most difficult to copy. Among all cloud vendors, only Google holds the three cards of self-developed top AI chips + self-developed top large models + self-developed top search engines.

Google Cloud: The third place’s counterattack

In the cloud business, Google is going through the most difficult and most dramatic path. Cumulative losses from 2008 to 2022 exceeded US$50 billion, and it was only in Q1 of 2023 that it achieved quarterly profits for the first time. Two years later,it has become Alphabet’s fastest-growing engine.

The numbers for 2025 are ridiculously tough. Full-year revenue was $58.7 billion, a year-over-year increase of approximately 36%. Q4 single-quarter revenue was US$17.66 billion, a year-on-year increase of 48%, a record high, and an annualized rate of more than US$70 billion. Operating profit reached $5.31 billion in Q4, up 154% year-on-year, and operating margin climbed to nearly 30%. From a cash-burning machine to a stable profit center, it only took three years.

Horizontal growth is more exciting than the growth rate - AWS will grow by about 19% in Q4 2025, Azure will grow by about 31%, Google Cloud will grow at 48%, leaving the two big brothers behind. In terms of market share, AWS is about 29-30%, Azure is about 20-22%, and Google Cloud is about 12-13%. The absolute gap is still there, but the relative gap is rapidly narrowing.

In August 2025, Meta signed a six-year, more than $10 billion cloud contract with Google—the largest single customer contract in Google Cloud’s history. Q3 backlog reached US$155 billion, and further surged to US$240 billion at the end of the year, a year-on-year increase of more than 55%.

Behind the order backlog figure is Google’s complete turnaround in breaking through major customers. Thomas Kurian made it clear at the earnings conference - The number of billion-dollar deals signed in 2025 will exceed the total number of the previous eight years. ServiceNow $1.2 billion, NATO Sovereign Cloud, Adobe, Cathay Pacific, Kraft-Heinz… the list is looking more and more like a Fortune 500 list.

What is the driving force? 70% of revenue growth comes from AI-related workloads. TPU and GPU consumption time on GCP will increase 3 times year-on-year in 2024, and Vertex AI (now renamed Gemini Enterprise) has become one of the preferred platforms for enterprises to deploy AI. This is the first time that Google's "AI full stack" narrative has truly turned into cash flow.

The risks should also be clearly stated. The absolute share is still less than half of AWS, enterprise-level sales and channel capabilities have historically been shortcomings, and the bundle of Microsoft 365 + Azure still crushes Workspace in the traditional office market. AI workloads have high gross profit margins, but CapEx investment is huge - Alphabet's capital expenditures will be US$52.5 billion in 2024, and will continue to increase significantly in 2025. When and at what speed this money will be paid back is the biggest uncertainty in valuation.

The growth rate is overwhelming, but the absolute share gap remains.

Google Cloud is no longer a drag on Alphabet, but the next growth engine for relay search. However, whether it can reach half the size of AWS in five years depends on the migration speed of new customers in the AI era such as Anthropic and Meta.

Quantum computing: leading technology, water far away cannot quench near thirst

In December 2024, the Willow quantum chip released by Google Quantum AI caused a nuclear bomb-level shock in the quantum computing circle.

With 105 superconducting qubits, on the Random Circuit Sampling (RCS) benchmark, to complete the calculation in 5 minutes, the world’s most powerful supercomputer would need 10 to the power of 25 years—that is, 10 septillion years. This is an order of magnitude completely divorced from daily intuition. But the real historical significance of Willow is not in the computing power itself, but in the sub-threshold error correction achieved on a surface code for the first time - when the number of qubits increases, the logical error rate decreases exponentially, rather than increases. This result was published in "Nature", and the title of the paper is "Quantum error correction below the surface code threshold".

Sub-threshold error correction is a critical step in quantum computing from "noisy intermediate scale (NISQ)" to "fault-tolerant quantum computing (FTQC)". As long as we continue to pile up physical qubits, we can create logical qubits with any low error rate - a theoretical promise that has been verified on a silicon chip for the first time.

Sundar Pichai’s goal for 2021 is to build a practical, error-corrected quantum computer by 2029. The six major milestones of the roadmap are quantum superiority (2019 Sycamore + 2024 Willow has been achieved), sub-threshold error correction (Willow has been achieved), multiple logical qubit systems, long-distance qubit connections, programmable fault-tolerant logical qubits, and large-scale quantum computers. Currently between the second and third steps.

The competitive landscape is crowded. IBM is betting on the superconducting route. Condor in 2023 has already accumulated 1,121 qubits, and the target is the Starling system in 2029 (200 logical qubits, 100 million gate operations). Microsoft is taking the topological qubit route. The Majorana 1 chip released in February 2025 has only 8 topological qubits, but it uses a revolutionary "topological conductor" material, which is theoretically more resistant to errors and is also the most controversial. IonQ uses trapped ions to achieve dual qubit gate fidelity >99.99% in 2025, the highest record in the world, but Q2 revenue is only US$20.7 million.

The problem is that quantum computing will not generate meaningful revenue in the short term. The global quantum computing market is estimated to be worth approximately US$2.7 billion in 2024, is expected to be US$3.52 billion in 2025, and is forecast to be US$5-10 billion in 2030. This scale is basically invisible in Alphabet’s US$400 billion financial report.

So why is Google still investing? Three reasons.

First, long-term optionality - once fault-tolerant quantum computers are commercially available, cryptography, AI training acceleration, drug research and development, and climate simulation will all be subverted, whoever comes first will get it.

Second, scientific research reputation - Willow’s Nature paper, like AlphaFold’s Nobel Prize, is the core proof in Google’s AI full-stack narrative.

Third, talent siphon—the top Ph.D.s in quantum physics go to only a few places, and Google wants to stay on that list.

Leading technology, far away from business. Quantum computing is more like a free lottery ticket in the valuation model than a cash flow contribution. The market won't pay a premium for it, but if it does materialize in 2029, the rules of the game will be rewritten.

Waymo and YouTube: One represents the future, the other represents the present

We talk about these two businesses together because they represent two ends of Google's asset portfolio -One is long-term options that are about to explode, and the other is a cash machine that is already running.

Waymo completed a textbook leap from 0 to 1 in 2024-2025. There were around 10,000 paid rides per week in August 2023, and already over 450,000 per week in December 2025 – a 45x increase in 18 months. In 2025, more than 14 million trips will be taken, and the cumulative driverless mileage will reach 170.7 million miles. The fleet has expanded from 1,500 Jaguar I-PACE vehicles in April 2025 to approximately 2,500 vehicles in November, with a target of 3,500+ vehicles in 2026.

Waymo’s safety data is too solid to question—peer-reviewed research based on 56.7 million miles shows a serious injury crash rate of 0.02/million miles, compared with the human baseline of 0.22, a reduction of approximately 91%. There was an 82% reduction in any injury reported incident rate and a 57% reduction in police reported incident rate.

What is even more interesting is the competitive landscape. After the Cruise San Francisco accident in October 2023, GM officially stopped funding the Cruise Robotaxi business in December 2024, and a total of US$10 billion in investment was wasted. Waymo becomes the only robotaxi company in large-scale commercial operation in the United States. Tesla’s FSD is still L2+ assisted driving, Cybercab is still being tested, and Waymo’s leading position in the field of autonomous driving is already an industry consensus.

In terms of commercialization, Waymo’s revenue in 2024 is approximately US$125 million, 2025 is expected to be US$180 million, and 2027 is forecast to be US$1.3 billion. In February 2026, it completed a US$16 billion Series D financing at a valuation of US$126 billion, bringing the cumulative financing to more than US$27 billion. Grand View Research predicts that the global robotaxi market will grow from $610 million in 2025 to $147.25 billion in 2033 (CAGR 99.1%) - a track where valuation models are still catching up with reality.

YouTube tells another story here - A cash machine that has matured and is still accelerating. Total annual revenue in 2025 will exceed US$60 billion, with advertising revenue of approximately US$40.37 billion. Q4 single-season advertising revenue hit a record of US$11.38 billion. YouTube officially surpassed Netflix ($45.18 billion) and Disney in 2025 to become the world's largest media/streaming company.

The subscription ecosystem is more stable. YouTube Music & Premium reaches 125 million paid subscribers in March 2025, with an average of 2 million new subscribers per month and estimated subscription revenue of approximately $16 billion per year. Shorts has an average of about 200 billion daily views, more than 2 billion monthly active users, and a participation rate of 5.91%, leading TikTok and Instagram Reels - More importantly, the hourly viewing revenue of Shorts in the US market has exceeded that of long videos, which is a qualitative change from "traffic" to "monetization".

AI is feeding back to YouTube, too. Dream Screen allows Shorts creators to generate background videos with one sentence. Auto-Dubbing (based on Gemini) automatically translates and dubs videos into multiple languages, retaining tone and emotion. It will expand to 80 million creators and add 11 new languages ​​by 2025. Veo 3 Fast is integrated into YouTube Studio, upgrading the creator tool stack.

Waymo’s moat is the first-mover advantage of the trinity of technology + regulation + data, but commercialization is still in its early stages and valuations will fluctuate greatly. YouTube's moat is a closed loop of creator ecology + distribution + advertising, which has taken shape. As long as TikTok's fate in the United States remains unchanged, its cash flow will be as stable as an old dog. One is options and the other is fundamentals. The combination of these two pieces of the puzzle just completes the time dimension of Google's asset portfolio - both now and in the future.

Google’s Hexahedron: Who is supporting the valuation and who is holding it back?

Pull the six lines into a table and the whole picture of Google will be clear:

The core judgment is simple - Search + YouTube is the basic market, supporting the current valuation; Cloud + AI is the growth engine, which determines the slope of the curve in the next three years; TPU is the hidden barrier, which determines whether competition is sustainable; Quantum and Waymo are options, which determine whether they will still be on the poker table ten years from now.

These six lines are not independent of each other, but are highly coupled. TPU trains Gemini, Gemini is installed in Search and YouTube, Cloud sells TPU and Gemini to Anthropic and Meta, and DeepMind's scientific research results feed back Cloud's customer stories. This kind of full-stack collaboration cannot be replicated by any single player such as OpenAI, Anthropic, AWS, and Azure.

But coupling is also a risk. If AI search accelerates the diversion of search advertising, more than 50% of Alphabet's overall revenue will be under pressure; although the antitrust relief does not dismantle Chrome, the ban on data sharing and default protocols will erode distribution advantages in the long term; CapEx continues to be at a historical high, and once the return cycle is lengthened, the patience of the capital market will be exhausted.

The biggest tail risk is path dependence - Google is spending money on each line, but how many lines the market is willing to pay a premium for depends on whether the synergy between these lines can actually be realized. If Gemini cannot become the killer of ChatGPT on Android, and TPU cannot break through Anthropic with external customers, once the Cloud growth rate slows down below 25%, the hexahedron will collapse into several disconnected businesses.

Google’s story has never been a straight line, but is woven across six fronts -Some people see the cracks in AI Overviews, others see Ironwood’s computing power, and what they see determines which Google you buy.