Why did Google crash tonight

G
$GOOGL , AMZN, and MSFT, the three giants of cloud computing, plummeted with high volume tonight, indicating that this is not a typical systematic decline of the Nasdaq, but rather funds cutting their exposure to specific AI cloud giants.

After seeing the news, the main reason is that G00G plans to cut enterprise token prices by 80%, and this news is the direct trigger for the flash crash.
Tonight's trading isn't about "long-term debt issues," but about "the potential full-scale price war for large model tokens."

Google officials emphasized in their 1/0 statement that if clients shift 80% of their workload to models like Gemini Flash, they can save over 1 billion USD; Google also disclosed that Gemini API has reached a processing scale of about 19 billion tokens per minute.

Microsoft's side is even clearer. Axios reports that Microsoft is considering using DeepSeek as a low-cost model option for Copilot Cowork and switching Copilot Cowork to usage-based billing; Microsoft's own statement is that testing shows this agent product cannot be supplied infinitely, or the bills will spiral out of control.

So the core logic behind tonight's plunge: the large model industry is shifting from "who has the strongest model" to "who offers the cheapest per million tokens," which is bearish for Google, Microsoft, and Amazon.

Previously, the market valued their AI cloud businesses based on this story: the more AI enterprises use, the more tokens consumed, the higher the cloud vendor's revenue, the higher the model vendor's revenue, and the entire chain expands together.

But now the market suddenly sees another side: if enterprise clients start to find tokens too expensive, model vendors will have to lower prices; if model vendors lower prices, cloud vendors will also need to give discounts; if Microsoft begins to introduce low-cost models like DeepSeek, it indicates even Microsoft is actively reducing reliance on high-cost models from OpenAI.

This will impact three expectations:
First, it undermines Google's AI monetization flexibility.
Google's original advantage was its own TPU, Gemini, cloud, and search entry points, so the market believed it could enjoy thicker profits in the AI era. But if Google actively reduces token costs, the market will question: is this because of efficiency gains allowing for price cuts, or because of intense competition forcing them? This is the key reason for the stock price decline.

Second, it undermines Microsoft's high-margin narrative for Copilot.
Microsoft's previous story was: Office + Copilot could turn AI into a high ARPU subscription. But if products like Copilot Cowork switch to usage-based pricing and introduce DeepSeek to cut costs, it suggests that the "fixed subscription fee with unlimited AI use" model can't hold up. Microsoft isn't unprofitable, but the market will reassess Copilot's profit margins and pricing power.

Third, it undermines Amazon AWS's AI cloud premium.
Amazon lacks the strong proprietary front-end AI applications like Google and Microsoft; it relies more on AWS + Anthropic + Bedrock. Now, if enterprises start routing models, handing complex tasks to Claude/OpenAI, and simpler tasks to DeepSeek, open-source models, or cheap models, AWS's AI revenue may still grow, but the value per token will be suppressed. The WSJ also directly wrote last week that the AI price war has begun, with companies using DeepSeek, Alibaba, and other low-cost models and open-source models to reduce AI costs, with some scenarios reducing costs by up to 95%.

Therefore, tonight's market isn't killing "AI demand," but rather: AI demand is huge, but clients are unwilling to pay current prices.

If the token price war begins,
the winners may be some low-cost model providers like DeepSeek, Alibaba, and parts of the open-source ecosystem.
The losers are companies supported by "high-priced frontier models + high valuation multiples."
OpenAI and Anthropic will be under pressure, but Google, Microsoft, and Amazon will also be affected because their cloud growth expectations include "high token unit prices and high AI profit margins."

So tonight's decline isn't because AI demand has collapsed, but because AI pricing power is starting to collapse.

If the token unit price drops by 80%, it means that what used to sell for 1 yuan now only sells for 0.2 yuan. Revenue won't necessarily decline; token usage must increase fivefold. If gross margins are also compressed, the required usage increase will be even higher.

Thus, the market isn't killing "AI has no demand," but rather: AI demand may continue to explode, but the monetization of each token is declining.
This shifts valuation logic from the previous: token explosion > cloud revenue explosion > high-margin software > high multiples
to: token explosion > price decline > cost competition > cloud vendors becoming
AI water, electricity, and coal
and valuation multiples naturally need to be revised downward.

Google's decline today was the sharpest because it hit three valuation pressures simultaneously:
First, Google's previous AI premium was based on the complete closed loop of Gemini + TPU + Google Cloud + search entry. It was seen as the most likely to regain pricing power in the AI era.

Second, if Google starts actively lowering token prices, the market will doubt: is this due to cost advantage or defensive necessity? If it's a cost advantage, it's good in the long run; if it's defensive, short-term valuation will decline.

Third, Google's core search advertising was already threatened by AI search, agents, and chat entry points. Now, with token prices being cut again, the market worries: future AI search will not only change traffic entry points but also lower the commercial value behind each query.

But Google also has a counterlogic. It might be the most qualified among the three to initiate a price war. Google disclosed in 1/0/2026 that its model API has processed about 19 billion tokens per minute, with over 8.5 million developers using its models monthly; this shows it has scale, self-developed models, TPU, and cloud infrastructure.

Therefore, the impact on Google's valuation is:
Short-term: PE and cloud business valuation multiples decline.
The market will reclassify Google from "AI pricing power winner" back to "AI cost war player."
Mid-term: If it proves that low prices come from TPU cost advantages, valuation can recover.
In the end, price wars favor low-cost producers gaining market share.
Long-term: Google isn't afraid of price wars; it's afraid that simultaneous price wars and search entry point restructuring will occur.

Google's biggest decline today is because it was seen as the first to fire the shot; Microsoft's decline is because signals of DeepSeek entering Copilot suggest major clients are already pushing for cost reduction; Amazon's decline is because AWS/Bedrock/Anthropic also can't escape the same price war.
Google's current dilemma isn't about "whether it has AI capability," but a more troublesome issue:
The stronger Google gets, the easier it is to overthrow itself; the less it lowers prices, the more likely it is to be revolutionized by others.
This is the biggest difference between Google and Microsoft, Amazon.
Microsoft can treat models as procurement items, integrating OpenAI, DeepSeek, Anthropic, and open-source models into Copilot and Azure; Amazon can treat models as cloud shelves, selling computing power and services via AWS. But Google is different; its core profit pool is search advertising. Once A becomes low-cost, real-time, question-answer, agent-style entry, the first to be restructured will be search.
So Google's dilemma is essentially: it must use Gemini to protect search, but the more successful Gemini is, the more traditional search advertising models need rewriting.
Google's past business model was perfect: users input keywords, Google provides search results, advertisers bid, users click, and Google takes a cut. The advantage was low cost, abundant ad inventory, and clear click paths.
But AI Search isn't like that. Products like AI Mode, AI Overviews, and Gemini operate on a different logic: users may no longer click ten blue links but get answers directly. This improves user efficiency, may be more precise for advertisers, but creates a complex new ledger for Google: higher computational costs per AI query, fewer traditional ad placements.
Google is aware of this problem, so it has been testing new ad formats for the AI search era, such as new ad types in Search, Direct Offers, etc., trying to migrate traditional search ads into AI Mode and AI Overviews.
The problem is, AI ads are not just a simple translation of search ads. Traditional search is "users actively look for products," with ads naturally embedded; AI assistants are "users let the machine make judgments," and if ads are inserted too aggressively, trust may be broken. If inserted too lightly, the original commercial efficiency can't be maintained.
This is Google's awkward position: it doesn't lack new entry points, but these new entry points haven't yet proven they can replicate the profit margins of old ones.
Google's recent moves clearly lean toward price wars.
It lowered Google AI Plus monthly fee from $7.99 to $4.99, doubling the included storage from 200GB to 400GB.
TechCrunch directly called this Google's move in the AI subscription price war.
In 1/0/2026, Google also cut the top-tier AI Ultra subscription from $250/month to $200/month, and launched a $100/month Ultra version aimed at developers and professionals.
Strategically, this is very smart. Google has TPU, cloud, models, search, Android, YouTube, and is most qualified to cut prices.
But the capital market will ask another question: if even Google has to cut prices to attract users, is the large model business really a high-pricing power business?
This is critical for Google's valuation. The previous market premium on Google's AI was based on the idea of "search cash cow + AI cloud growth + model monetization" as a triple asset. Now, with price wars, the premium for model applications is first being discounted. In other words, Google isn't falling because it lacks AI; it's because the market is beginning to doubt: AI might increase Google's revenue but not necessarily improve its profit margins.

Google's current problem is somewhat like Microsoft's in the past.
Microsoft's shift from Windows licensing to cloud subscriptions was mainly a business model transition, not a technical challenge. Google now faces the same: moving from a "keyword advertising company" to an "AI answer and task execution platform."
But this is even harder than Microsoft's transition, because Microsoft’s cloud shift involved moving Office and enterprise software to subscriptions with clear user payment paths.
Google's AI Search is about拆开重装 its most profitable search ad model. Users may feel more comfortable, advertisers may be more precise, but the auction mechanisms, click paths, ad inventory, publisher ecosystem—all need to be redesigned.