[Memorandum] Nvidia Siege or “SpaceX Impairment”? — The Reality Behind Alphabet's 8% Stock Drop Despite 100 Google TPU External Sales and OpenAI Contact
When lining up the news that flowed in succession from October 27th to 28th, one cannot help but lean in and wonder, “Has Nvidia’s stronghold finally begun to crumble?” First, on the 27th, U.S.-based Oracle announced that it would offer Google’s proprietary AI chip, the “TPU,” on its own cloud platform, “OCI.” This is the first time a hyperscaler other than Google has sold TPUs externally, with the contract being multi-year and multi-gigawatt in scale, and service scheduled to begin in 2027. Just a few hours later, U.S. media outlet The Information reported that “OpenAI is considering purchasing Google’s TPUs.” OpenAI has been the largest customer that has continued to buy Nvidia’s GPUs in the greatest volume, so the fact that this company is in contact with the Google camp carries significant impact.
Furthermore, on the 28th, Google’s parent company, Alphabet, announced its third-quarter financial results. Revenue was $127.6 billion (approximately 20.4 trillion yen, a 25% increase year-on-year), and cloud division revenue was $30.7 billion (approximately 4.9 trillion yen), a 103% increase year-on-year, effectively doubling. In the earnings call, it was revealed that about 100 companies are using TPUs, and partnerships with Anthropic, Apple, Oracle, and others are following one after another. Despite this, Alphabet’s stock fell by over 10% at one point in after-hours trading and by over 8% based on the closing price. The twisted development of “Why would the stock of a company that is supposed to be cornering Nvidia with TPUs plummet?” is, to me as a humanities-focused amateur, honestly quite an interesting event.
1. The Reality of “100 TPU External Sales”: Last Year’s “Rumors” Have Become Numbers
The story of Google’s TPU external sales itself is actually not new. Looking back to November 25, 2025, when The Information reported that “Meta is considering adopting Google’s TPUs for its 2027 data centers,” Nvidia’s stock fell by up to 7.1% during trading hours and 2.6% at the close, wiping out about $115 billion in market capitalization. At that time, it was still at the “rumor under negotiation” stage, and the contract for Anthropic to obtain up to 1 million TPUs (announced in October 2025) was the most prominent material. In these financial results, however, the number of TPU-using companies was announced with a concrete scale of about 100, and specific figures were even provided, such as the training “TPU v8t” having up to 2.7 times the performance of the previous generation, and the inference “TPU v8i” having an 80% improvement in price-performance ratio. Google is also providing “TorchTPU,” which works with Meta to run PyTorch on TPUs, and with external sales via Oracle and contact with OpenAI, the largest Nvidia customer, last year’s “rumors” seem to have emerged with quite a genuine outline.
However, Nvidia itself does not seem to be watching this situation idly. The “landlord/guarantor” strategy of guaranteeing customer financing to the tune of $164.5 billion, which I touched upon in a previous memorandum, is likely one such move, and reports also suggest moves to use capital to penetrate potential switchers to Google through investments in OpenAI and Anthropic. The depth of the CUDA development environment remains Nvidia’s strength, and Google’s own CEO Sundar Pichai has positioned TPUs as “one of the platforms that will be polished by competing with other companies,” so it does not seem to be a simple story of “TPUs immediately replacing Nvidia.”
2. The “Real Reason” for the 8% Stock Drop: SpaceX Impairment Rather Than the Siege
This is the point that was most anticlimactic this time, but the direct trigger for the sharp drop in Alphabet’s stock seems to have had little to do with the momentum of the TPU external sales competition. Earnings per share (EPS) were $2.12, significantly lower than the market expectation of around $3, but the main cause of this downside is said to be the deterioration of “other income/expenses” unrelated to the core business, specifically an unrealized valuation loss of approximately $8.2 billion incurred when marking the shares of investee SpaceX to market value. In other words, while the core businesses of advertising and cloud were strong (the cloud division’s operating margin rose from 23.7% the previous year to 36.2%), an unrelated investment valuation loss worsened the overall impression of the financial results.
In addition, the capital expenditure (capex) plan also became a source of concern. The guidance for 2026 was revised upward from the previous $180-190 billion to $195-205 billion, and for 2027, it was only warned that it would “increase significantly further” without specific figures being provided. Perhaps the sense of caution regarding expenditures ballooning without seeing when the returns on massive investments will appear, combined with the SpaceX impairment, pushed the stock down. What is interesting is that Nvidia’s stock on the same day, the 28th, actually rose by 0.9% to $217.94 during the day. While headlines like “Google TPU threat and Nvidia’s debt risk shake growth prospects” flew around the world, if one follows the actual stock price movements honestly, the one that bled the most that day was not the Nvidia camp but the Google camp; I think this gap between headlines and reality is very characteristic of this market.
3. Implications for Japan: Is Ibiden’s Bet Really a “Winner”?
What concerns me here is the story of Ibiden, which I covered in a memorandum six days ago. Nvidia had even made “prepayments” to secure orders from Ibiden in Gifu to triple its GPU substrate production. In that article, I wrote about Samsung Electro-Mechanics entering Nvidia’s substrate supply chain as a “threat,” but the expansion of Google TPUs that we learned about this time seems to pose a question one level larger than that. In other words, before the competition over which company, Ibiden or Samsung Electro-Mechanics, takes the substrate order share, if the hyperscalers’ computing resources themselves shift from Nvidia GPUs to Google TPUs, there is a risk that the demand source for the factories that Ibiden expanded even after receiving prepayments will itself diminish.
Moreover, looking at the Google TPU memory (HBM) supply chain, reports as of the end of 2025 indicated that Samsung Electronics would handle about 60% of HBM3e and was expected to be a major supplier for HBM4 as well (I must honestly add that I have not been able to confirm updated information for 2026 regarding these figures, so they may be somewhat old). In other words, it is a structure where “bonuses” for the TPU camp are likely to flow heavily to South Korea, and for Japan’s Ibiden, which finally showed its presence in substrate demand for the Nvidia camp, if the shift from GPUs to TPUs progresses, it looks like a two-tiered risk where part of the special demand it worked so hard to capture could be redistributed to South Korea beyond its reach. However, as long as the total amount of AI computing resources continues to increase, one could also take the view that demand for advanced packaging and HBM itself is solid regardless of whether GPUs or TPUs grow, so it would be premature to be too pessimistic. It seems necessary to look calmly at both the risks of betting on Nvidia alone and the benefits of the continued growth in overall AI demand.
(Afterword/Memo)
What was most surprising while researching this time was that on the day the headlines “Google is cornering Nvidia” were lined up, it was Google itself whose stock was actually plummeting. I think the figure of 100 companies for TPU external sales has a certain weight, but stock prices seem to move for more myopic reasons (in this case, the SpaceX valuation loss). As a humanities-focused amateur, every time I see this kind of “gap between headlines and reality,” I feel that it is dangerous to conclude the collapse of a monopoly structure based solely on the first line of the news. The risk of Ibiden’s single-legged approach also feels one level more complex than when I wrote about it before, with the addition of the new variable of TPUs. By the time we reach next year’s GTC or Ibiden’s financial results, I think a clearer picture will emerge as to which way this “TPU vs. GPU” tug-of-war is turning.
※Previous article (Related 1): [Memorandum] Nvidia “Guarantee” $164.5 Billion, 72% of Shareholders’ Equity — Behind Huang’s Boast That “GPUs Are the Strongest Collateral,” Is Masayoshi Son’s 10 Trillion Yen OpenAI Investment Safe?
Reference articles below
Earnings call transcript: Alphabet Q3 2026 profit miss sparks 10% afterhours drop (Investing.com)
Broadcom Snaps Up Half Of Samsung’s HBM Output For Google AI Chips (Benzinga)