A former MediaTek employee, in an expert-network interview that investor Rihard Jarc posted on X on 26 October 2026, put Google’s external TPU hardware revenue at about $8 billion this year, $51 billion in 2027 and around $104 billion in 2028. These are one person’s estimates, not Alphabet’s figures, and we could not verify them.
Alphabet reports its third quarter on Wednesday, 28 October, with its call at 1:30pm Pacific Time. On its second-quarter call on 22 July, CFO Anat Ashkenazi said Google had started delivering TPU systems to customer data centers and that it anticipates “the vast majority of the revenues from these agreements will be realized in 2027.”
Video explainer · Google TPUs before Alphabet’s Q3
A 16:9 video explainer narrated from the published text of this piece and four related FourWeekMBA pieces.
The key insight: As we read it, the expert and Alphabet point to the same year, 2027, but only Alphabet’s words are on the record, and in the TPU remarks we read Alphabet gave timing, not a dollar figure.
Ashkenazi told analysts in July that Google “began to recognize revenues from TPU system sales, which we delivered to customer data centers for the first time in Q2.” Cloud revenue rose 82% to $24.8 billion, and she said growth “accelerated meaningfully, even after excluding the impact of TPU system sales.”
She said Google expects “to recognize a relatively small portion of the revenues from our existing TPU system sales agreements this year, ramping as we exit 2026.” The cost side showed up too: Other Cost of Revenues rose 22%, which she attributed partly to “inventory costs, primarily from the sales of TPU systems to customers.”
Google Cloud’s backlog reached $514 billion, up more than $50 billion in the quarter, with just over 50% expected to turn into revenue over the next 24 months. Asked about TPU sales, Ashkenazi said “the vast majority of the $514 billion of cloud backlog is the GCP agreements. But the TPU system sales are reflected in that backlog.” In the TPU remarks we read, Alphabet gave no dollar figure for TPU system sales.
CEO Sundar Pichai pointed to one example of where the systems go: TPUs placed “in their data centers, or in other data centers, like the project we are doing with Blackstone.” Alphabet raised its 2026 CapEx guidance to $195-205 billion and said it expects CapEx to increase significantly in 2027.
Google external TPU hardware revenue as estimated by one former MediaTek employee in an expert-network interview relayed by Rihard Jarc on X: about $8B in 2026, $51B in 2027 and around $104B in 2028. Not Alphabet figures; unverified. 13x is our arithmetic.Google external TPU hardware revenue as estimated by one former MediaTek employee (expert-network interview relayed by Rihard Jarc on X, 26 Oct 2026): $8B in 2026, $51B in 2027, around $104B in 2028. Not Alphabet figures; unverified.
What Was Posted on X
Interview with a Former MediaTek employee on why a few hyperscalers will determine how the AI chip market evolves ($NVDA, $AMD, $AVGO, $AMZN, $GOOGL):
1. The expert describes $AMD’s possible role in $GOOGL’s TPU program as an early-stage discussion tied to v10, with nothing confirmed for v11. $AMD would mainly bring packaging and heterogeneous integration expertise, plus CPU IP, since placing CPU cores alongside the TPU could cut communication overhead and improve latency for agentic AI and reinforcement learning workloads. He does not see this as $AMD replacing $AVGO, whose significant share in the TPU is not in danger, but as part of $GOOGL’s consistent strategy of diversifying its vendors.
2. He puts $GOOGL’s external TPU hardware revenue at about $8 billion this year, rising to roughly $51 billion in 2027 and around $104 billion in 2028. $GOOGL began recognizing external TPU sales in Q2 of this year, and management has said most revenue from existing agreements will be recognized in 2027. The expert notes these sales sit inside $GOOGL’s roughly $514 billion cloud backlog, which could scale TPU system sales from single-digit billions in 2026 to tens of billions in the following years.
3. The expert agrees that $NVDA can keep growing in dollar terms even as custom silicon takes share, because total AI demand is expanding so quickly. The real risk is the precedent rather than the loss of any single customer. If $GOOGL shows that custom accelerators have compelling economics, $MSFT and $AMZN may gain even more confidence in their own ASIC programs, which could pressure $NVDA’s long-term share and pricing power.
4. According to the expert, $MSFT is the largest hyperscaler buyer of $NVDA GPUs, driven by heavy Azure AI demand, with $META also highly exposed because it relies on GPUs for large-scale training and inference despite its MTIA effort, which the expert says could have gone better. $AMZN is running both strategies, deploying $NVDA GPUs alongside Trainium and Inferentia, while $GOOGL is the clearest custom ASIC counterexample, remaining a meaningful $NVDA customer but a smaller one than $MSFT and $META.
5. The expert notes that more than half of $NVDA’s data center revenue comes from these four hyperscalers, with 2025 shares of roughly 18% from $MSFT, 13% from $META, 9% from $AMZN, 9% from $GOOGL, and about 5% from $ORCL. The expert expects that split to shift in the coming year, with $GOOGL’s dependency on $NVDA falling while $MSFT, $META, and $AMZN stay heavily reliant.
6. He expects CapEx at major cloud providers to keep rising in absolute dollars, from roughly $880-900 billion this year to about $1.3 trillion next year and around $1.5-1.7 trillion in 2028. The expert is confident the 2028 dollar increase will be larger than the 2027 increase, even though the percentage growth rate is slowing.
Rihard Jarc posted excerpts from an expert-network interview with a former MediaTek employee, labelled as found on AlphaSense. In them the expert puts Google’s external TPU hardware revenue at about $8 billion this year, $51 billion in 2027 and around $104 billion in 2028, describes AMD‘s possible TPU role as an early-stage discussion tied to v10, and says Nvidia‘s risk is the precedent rather than losing any single customer.
What the Expert Said
One former MediaTek employee’s estimates of Google’s external TPU revenue, 2026 to 2028. Alphabet said on 22 July that the vast majority of revenue from its existing TPU system sales agreements will be realized in 2027. Sources: Jarc on X; Alphabet Q2 2026 call.
In the transcript screenshots Jarc posted, which he labels as found on AlphaSense, the expert says: “$8 billion was the external TPU hardware revenue this year.” For later years: “2027, that’s going to be $51 billion, not $75 billion. 2028, that’s going to be around $104 billion.” The excerpt does not say whose $75 billion figure the expert is rejecting.
By our arithmetic, the expert’s 2028 figure is 13 times the 2026 one, and 2027 is 6.4 times 2026. The expert ties the path to the backlog, saying TPU system sales “are included in that backlog that could plausibly scale from a single-digit billions in 2026 to tens of billions of dollars.”
On suppliers, the expert describes AMD’s role in Google’s TPU as “an early discussion phase right now”, tied to the v10 program, and says: “They are not replacing Broadcom.” On Nvidia, the expert says: “NVIDIA’s risk is not just towards losing one customer here. It’s the precedent.”
The expert also puts 2025 shares of Nvidia’s data center revenue at 18% for Microsoft, 13% for Meta, and about 9% each for Amazon and Google. These are the expert’s figures.
The Numbers to Check on 28 October
Three figures from the July call can be compared with the third-quarter release and call: the $514 billion cloud backlog, the TPU system revenue that management said would ramp as it exits 2026, and the $195-205 billion CapEx range.
Update, 27 Oct: TPUs on Oracle’s cloud. Oracle and Google said on 27 October that Google TPUs will be available through Oracle Cloud Infrastructure, “the first time Google TPUs will be offered on a hyperscale cloud platform outside Google,” in calendar year 2027, according to their joint release. Our follow-up has the details.
As we read it, the backlog is the place to look: Ashkenazi said TPU system sales sit inside the $514 billion, while the vast majority of it is GCP agreements, so the TPU share is not given in the passages we read.
As we read it, the cost line moved first: Ashkenazi tied part of the 22% rise in Other Cost of Revenues to TPU system inventory, and said Google starts building inventory to be able to sell those systems before it recognizes the revenue.
As we read it, the expert’s Nvidia point is about precedent rather than lost dollars: the expert says Nvidia can keep growing as total AI demand rises, and that if Google shows compelling economics, Microsoft and AWS can gain confidence in their own chip programs.
“We anticipate the vast majority of the revenues from these agreements will be realized in 2027”
Three Implications
FOR NVIDIA The expert names Microsoft and Meta as Nvidia’s largest hyperscaler buyers and Google as a smaller one, and puts Google at about 9% of Nvidia’s 2025 data center revenue; these are the expert’s figures.
FOR BROADCOM AND AMD The expert says AMD’s TPU role is an early discussion tied to the v10 program and that AMD is not replacing Broadcom.
FOR ALPHABET’S Q3 Alphabet’s call is on Wednesday, 28 October, at 1:30pm Pacific Time; the July call gave the 2027 timing, the $514 billion backlog and the $195-205 billion CapEx range.
The Business Engineer Lens
This story maps onto the Business Engineer framework The AI Capex Map.
The framework starts from the hyperscalers’ order books: “the four largest U.S. cloud providers now hold roughly $2.1 trillion in revenue backlog”.
As we read it, Google’s TPU system sales put part of that backlog into chips Google sells rather than rents, which is why the TPU share of the $514 billion matters. The framework was published in May 2026 and does not discuss this interview.
What Is Not Established
The $8 billion, $51 billion and $104 billion figures come from one unnamed former employee of a chip company, in an excerpt relayed on X; we did not see the full transcript. In the passages we read, Alphabet does not give TPU revenue, unit volumes or customer names beyond the Blackstone project, and it does not say how much of the $514 billion backlog is TPU system sales.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
One former MediaTek employee puts Google’s external TPU sales at $8 billion this year and $51 billion in 2027; Alphabet itself has said only that most TPU system revenue from its existing agreements lands in 2027, inside a $514 billion cloud backlog it reports again on 28 October.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Alphabet facts are from its Q2 2026 earnings call transcript (22 July 2026) and its 6 October 2026 notice of the Q3 call, read on Alphabet’s investor relations site on 27 October 2026. The TPU revenue estimates, the AMD, Broadcom and Nvidia remarks and the customer shares are from an expert-network interview with a former MediaTek employee, as excerpted in screenshots Rihard Jarc posted on X on 26 October 2026; they are one person’s views and we could not verify them. 13x and 6.4x are our arithmetic. This is not investment advice. We did not contact Alphabet.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
