Google’s chip business is on a faster track than Nvidia’s own AI boom, according to Yahoo Finance’s reporting as of November 9, 2026. The evidence starts with the calendar. Alphabet Inc.’s Google introduced two new AI chips, the TPU 8tx and the TPU 8i Max, joining the TPU 8t and TPU 8i that arrived earlier. The report says this was the first year Google introduced four TPU systems.
That cadence matters because Google used to ship roughly one data-center TPU per year. Four in a single year signals a company that treats its Google TPU line as a product family rather than an internal side project. Bloomberg Intelligence puts a number on the ambition: it estimates TPU sales could climb from about $12 billion in 2026 to more than $135 billion by 2028, excluding chips rented through Google Cloud.
Nvidia is not standing still, and the comparison deserves care. The company posted $96.2 billion in fiscal second-quarter 2027 revenue, a figure we unpack in our roundup of the AI earnings week. This analysis separates what the reports confirm from what rests on projections, because the headline claim blends both.
Add Shattered once in the Google app and our stories show up in your news suggestions.
Google’s Four-TPU Year: What Changed in 2026
For years, Google’s TPU roadmap moved at a measured pace. The company would unveil one data-center part, tune its software stack around it, and let internal teams and a handful of cloud customers absorb the capacity. Yahoo Finance’s report describes a break from that rhythm. In 2026 Google introduced the TPU 8tx and TPU 8i Max, adding them to the TPU 8t and TPU 8i for a total of four systems.
The naming itself hints at segmentation. The “t” and “i” suffixes in Google’s lineup sit alongside the new “tx” and “Max” variants, which suggests Google is carving its silicon into distinct tiers. The reports available to us do not spell out what each letter means, and we will not guess at specifications. What the naming does tell readers is that Google wants more than one answer to the question of what an AI customer needs.
Why would a company split one product into four? Customers run very different workloads. A lab training a frontier model has different priorities from a trading firm running inference on market data, and both differ from a university lab with a fixed budget. A broader lineup lets Google pitch each buyer a tailored option instead of one chip for everyone.
The shift also changes how analysts read Alphabet. Chips used to appear in coverage as a cost-saving tool for Google’s own search and cloud services. Now Bloomberg Intelligence models them as a sales line that could rival the scale of established silicon vendors. That reframing, more than any single chip, explains why the story is dominating AI-market coverage today.
TPU 8t, 8i, 8tx and 8i Max: What the Reports Confirm
What is confirmed
Three facts hold up across the reporting. Google introduced the TPU 8tx and TPU 8i Max. They join the TPU 8t and TPU 8i, bringing the total to four TPU systems introduced this year. And Google previously released about one data-center TPU annually. Those points come straight from the Yahoo Finance report and form the factual core of the story.
What is not confirmed
The reports give no performance figures, no memory specifications, no power ratings and no pricing for any of the four chips. They also give no shipment volumes, margins or recognized TPU revenue. Readers will see benchmark claims circulating on social media, but none of them appear in the verified reporting, so we leave them out. Anything that sounds like a spec sheet should be treated as unsourced until Google publishes one.
That gap is not a minor footnote. Without prices and volumes, nobody outside Google can compute how much revenue the TPU line generates today. The Bloomberg Intelligence figure is an analyst estimate, not reported company revenue, and the distinction runs through every section below.
Bloomberg Intelligence’s $12 Billion to $135 Billion Path
The most quoted number in the story comes from Bloomberg Intelligence. The firm estimates TPU sales, including related software and hardware, could rise from roughly $12 billion in 2026 to more than $135 billion by 2028. It excludes chips rented through Google Cloud, so the estimate covers only part of Google’s potential silicon income.
Run the arithmetic and the ambition becomes clear. Going from $12 billion to $135 billion is an increase of about 11 times in two years. If the path were smooth, it would imply annual growth of roughly 235 percent in each of those years. That is our calculation from Bloomberg Intelligence’s endpoints, not a figure the firm published.
Compare that with the growth Nvidia just reported. Its data center revenue rose 117 percent year over year in the latest quarter. Even if Nvidia repeated that pace for two straight years, the business would grow about 4.7 times. A tripling-plus every year for Google’s TPUs would outrun that rate by a wide margin, which is the mathematical heart of the “faster track” framing.
Several caveats apply. The estimate is a forecast, and forecasts for fast-moving hardware markets miss in both directions. It relies on assumptions about customer demand, chip supply and pricing that the available reports do not disclose. The figure also leaves out cloud rentals, which means the true Google chip opportunity could look bigger or smaller depending on how analysts count rented capacity.
The takeaway: Bloomberg Intelligence’s number is a credible directional signal and a poor basis for hard claims. It tells readers that a serious research shop sees TPUs scaling quickly. It does not tell them that Google has already booked anything close to $135 billion.
Nvidia’s Fiscal Q2 2027: The Baseline Google Is Chasing
Any claim that Google is outpacing Nvidia needs a fair look at Nvidia’s numbers. For the quarter ended July 26, 2026, Nvidia reported revenue of $96.2 billion, up 106 percent from a year earlier and up 18 percent from the previous quarter. Data Center revenue came in at about $89 billion, up 117 percent year over year. For the following quarter, Nvidia guided to roughly $108 billion in revenue with a gross margin near 74 percent.
Two details stand out. First, Data Center is roughly 92 percent of total revenue by our arithmetic ($89 billion of $96.2 billion), so Nvidia’s fortunes track AI infrastructure spending almost one for one. Second, the guided $108 billion implies the sequential growth rate is cooling from 18 percent to about 12 percent, even as the absolute dollars keep climbing.
Cooling sequential growth on a bigger base is normal for a company this large, but it matters for the narrative. Investors who have watched the stock swing on headlines, as covered in our look at the $4.5 trillion Nvidia crisis, now have a competing story to weigh. The reports available to us do not quantify how much of Nvidia’s order book any TPU customer might redirect, so that link stays a question rather than a finding.
Faster Track or Smaller Base? Reading the Growth Race
The headline compares two growth stories, but the two sit on very different foundations. Nvidia’s figures are reported results. Google’s figures are an analyst projection that excludes cloud rentals. Putting them side by side shows why “faster” and “bigger” point in opposite directions.
The scale gap is stark. Nvidia’s single-quarter Data Center revenue of about $89 billion is roughly 7.4 times Bloomberg Intelligence’s full-year 2026 TPU estimate. Annualize Nvidia’s latest quarter and you reach around $356 billion. The 2028 TPU forecast of $135 billion would equal roughly 38 percent of that run rate. We offer this as an illustration of scale, not a forecast for either company.
So is Google’s chip business on a faster track? On growth rate, the projections say yes. On absolute dollars, Nvidia leads by a wide margin and would keep leading under any plausible reading. The unconfirmed flag on the sheet captures it well: comparisons showing Google’s TPU business growing faster than Nvidia’s AI business rest partly on projections and are not established by reported figures alone.
A smaller base always grows faster in percentage terms. A business that doubles from $12 billion adds $12 billion. A business at $89 billion per quarter that grows 20 percent adds more than $17 billion in one quarter. Percentages flatter small bases, and readers should hold both lenses at once.
Oracle Puts Google TPUs Outside Google’s Walls
One development in the reporting changes the strategic picture more than any chip name. Oracle announced that it would offer Google TPUs, letting customers use them outside Google’s own infrastructure for the first time. Until now, anyone who wanted TPU capacity had to buy it through Google.
That access point matters for three reasons. It widens the buyer pool to companies that already run their workloads on Oracle’s cloud. It gives Google a second sales channel that does not depend on convincing customers to move onto Google Cloud. And it puts TPUs into more direct contact with Nvidia GPUs on neutral ground, since customers can compare options inside an environment they already trust.
It also shows a shift in posture at Alphabet. A company that kept its best silicon in-house for years is now willing to let a rival cloud resell it. That choice only makes sense if Google believes TPU demand will outrun what its own data centers can serve, or if it sees more upside in chip volume than in locking customers into one platform.
The reports we reviewed do not state commercial terms, capacity levels or pricing for Oracle’s offering, and the 2027 availability window is the only timing mentioned. We will not fill those gaps with guesses. Readers evaluating the Oracle route should ask both companies for the specifics before assuming any cost or performance advantage.
Anthropic’s Million-TPU Deal and Nearly 100 Customers
The biggest named commitment in the story belongs to Anthropic. Reports say the AI lab agreed to use as many as 1 million TPUs in an agreement valued in the tens of billions of dollars. The exact contract value and the final chip quantity are not specified in the available results, so “as many as” is the correct framing. Anthropic’s compute appetite also shows up in our coverage of its reported gas-power purchase, which underscores how much physical infrastructure sits behind a deal of this size.
Thomas Kurian, CEO of Google Cloud, framed the broader demand picture this way: “Nearly 100 customers use TPUs today, and we’re seeing strong and growing demand from many others.” That is a company executive describing his own business, so readers should weigh it as a sales statement rather than an independent audit. Still, it is a useful count. Ninety-odd named or unnamed buyers is far from a single-customer story.
Reported TPU customers include Anthropic, Apple, Physical Intelligence, Citadel Securities, Jump Trading, Caltech, Cassava and Cirrascale. The mix spans AI labs, a consumer hardware giant, trading firms, a university and cloud-adjacent providers. That breadth supports Google’s claim that TPUs are not a niche product for one workload.
The Anthropic relationship also has a strategic edge. Anthropic does not rely on a single chip family, and its choice to commit at this scale tells the market that TPUs now sit in the same conversation as Nvidia GPUs for frontier work. For investors tracking the lab’s finances, our piece on Anthropic’s IPO commitments shows how large compute obligations loom over its balance sheet.
Google TPU vs Nvidia GPUs vs Other Custom Chips
Google is not the only hyperscaler building its own silicon. Amazon and Microsoft both pursue custom accelerators, and we compared two of them in our look at Trainium3 versus Maia 300. What sets Google apart in the current reporting is the combination of a fast release cadence, a growing list of outside customers and a third-party cloud route through Oracle.
Where Nvidia still leads
Nvidia leads on revenue scale by a huge margin, as the $96.2 billion quarter shows. It also holds the advantage of a broad customer base that buys GPUs without committing to one cloud. A buyer can purchase Nvidia hardware through nearly any provider, while TPU access has historically meant going through Google, with Oracle planning to add a second door in 2027.
Where Google is gaining
Google gains where customers want an alternative supplier. Large AI labs have strong reasons to avoid depending on one vendor for something as critical as compute. The Anthropic commitment, the customer list and Kurian’s account of growing demand all point to buyers who want a second option, and some of them appear to be treating TPUs as a primary one.
Neither company has published a head-to-head benchmark in the sources we reviewed, so we cannot rank the chips on speed, efficiency or cost. Treat any claim of a clear technical winner as unverified. The choice depends on the workload, the software stack a team already uses and the price a seller will actually quote.
Historical Context: From One TPU a Year to Four
The release pattern is the most concrete piece of history in the reports. Google previously introduced about one data-center TPU per year. In 2026 it introduced four TPU systems, the first time it has done so. The jump from one to four in annual releases is a fourfold change in cadence, which is the kind of statistic that stays useful regardless of how the revenue estimates age.
A faster cadence tends to reflect two pressures. Demand pushes a vendor to cover more workloads, and competition pushes it to shorten the gap between generations. Nvidia has long run an aggressive refresh cycle, and Google’s move suggests it wants to close that gap. The 2026 launch slate reads like a response to a market that now rewards speed.
The financial backdrop has shifted too. AI infrastructure spending has swollen to levels that would have sounded implausible a few years ago, and our coverage of record Big Tech bond issuance shows how heavily the industry is borrowing to fund it. In that environment, a hyperscaler with its own chips can protect margins in ways a pure GPU buyer cannot.
Historical context also warns against overreading one year. A company can ship four products and still struggle to sell them at scale. The cadence shows intent and engineering capacity. Revenue tells the rest, and the revenue data is the part the public still lacks.
Market Impact for Nvidia, Alphabet and the Supply Chain
For Nvidia, the TPU story adds pressure at an awkward time. The stock has already faced volatility, which our coverage of the $2.5 trillion AI sell-off explored from the regulatory side. A credible alternative supplier gives large buyers more leverage in price and allocation talks, even if they keep buying Nvidia hardware in volume.
For Alphabet, the story reframes the investment case. Analysts who once valued Google mostly on search and cloud can now add a chip line with an analyst-estimated path to $135 billion. If that path proves even partly right, Alphabet would earn income from a market that Nvidia has dominated, and from customers who also pay it for cloud services.
For the supply chain, a larger TPU business raises demand for the same constrained inputs that GPUs need, such as advanced packaging and high-bandwidth memory. More chip designs competing for those inputs could tighten supply for everyone. That dynamic fits with the shortage themes in our report on sold-out HBM and data center permitting.
Customers sit in the middle. More choice in accelerators is good for buyers, but it also raises switching costs in software. Teams that tune their code for TPUs may find it harder to move later, and teams locked into Nvidia tooling face the same issue in reverse. The market effect is not a simple swap of one supplier for another.
What Stays Unconfirmed in the TPU-vs-Nvidia Story
A fair read of this story requires a clear list of unknowns. The available reporting does not establish specific TPU prices. It does not give shipment volumes or margins. It does not show Google TPU revenue already recognized. And it does not deliver a settled Google-versus-Nvidia growth-rate comparison.
The claim that Google’s TPU business is growing faster than Nvidia’s AI business is a comparison built partly on projections. Nvidia’s figures are results. Google’s are an estimate. Any article that treats the two as equivalent data points overstates what the evidence shows, and we have tried to avoid that here.
The Anthropic deal carries its own caveats. Reports describe it as valued in the tens of billions of dollars and covering as many as 1 million TPUs. Neither the final contract value nor the final quantity appears in the available results. Phrases like “up to” describe a ceiling, not a delivery.
Finally, Kurian’s customer count comes from Google itself. It likely reflects real demand, but it is not an independent measure and it does not say how much any customer spends. Until Google or its customers disclose more, the numbers that matter most remain estimates.
Five Predictions for Google TPUs and Nvidia
These are our own analytical judgments, drawn from the reported facts. They are not forecasts from Google, Nvidia or Bloomberg Intelligence.
- More clouds will follow Oracle. Once one third-party provider offers TPUs, others face pressure to match, since customers like to buy accelerators inside the cloud they already use.
- Nvidia will stay far ahead in absolute dollars through 2028. Even under the most bullish TPU path, Nvidia’s guided $108 billion quarter dwarfs a $135 billion annual TPU forecast on a run-rate basis.
- Analyst estimates for TPU sales will get revised. A figure that spans an 11-fold increase leaves enormous room for error, and new customer disclosures will push the numbers up or down.
- Google will face pressure to disclose TPU revenue. Investors who now rely on third-party estimates will want recognized figures, and a growing external business is harder to keep opaque.
- Chip supply, not demand, will set the pace. With HBM and packaging already tight, the winner may be whoever secures components, not whoever designs the best part.
Each prediction hinges on facts we cannot yet confirm, so treat them as prompts for what to watch. The first earnings reports that break out TPU revenue, or a second major cloud partner, would be the clearest signals that the track is as fast as advertised.
What Buyers and Developers Should Do Now
Teams planning AI infrastructure spending should avoid treating this story as a reason to switch suppliers overnight. Start by benchmarking your own workload on both platforms. Ask Google, Oracle and Nvidia’s channel partners for written pricing and delivery dates. Treat any claim about speed or cost savings as unproven until a vendor puts numbers in a contract.
Developers should look at portability. Frameworks that run on several accelerators reduce lock-in, and a small pilot on TPU capacity can reveal friction before a large commitment. Keep the pilot scoped to a workload where a failure is cheap to absorb.
Finance teams should model scenarios rather than point estimates. With Nvidia guiding to about 74 percent gross margin and a competitor emerging, pricing could move in either direction as supply loosens or tightens. Build contracts with flexibility for volume changes. For wider coverage of model releases, chips and industry shifts, see our AI and machine learning hub.
FAQ: Google TPUs and Nvidia’s AI Boom
What new chips did Google introduce in 2026?
Google introduced the TPU 8tx and TPU 8i Max. They join the TPU 8t and TPU 8i, making four TPU systems this year, the first time Google has introduced that many in one year, per Yahoo Finance.
How big could Google’s TPU sales get?
Bloomberg Intelligence estimates TPU sales could grow from about $12 billion in 2026 to more than $135 billion by 2028, excluding chips rented through Google Cloud. It is an analyst estimate, not reported company revenue.
Is Google’s chip business really growing faster than Nvidia’s?
On projected growth rates, the Bloomberg Intelligence path implies faster percentage growth than Nvidia’s 117 percent Data Center increase. But the comparison rests partly on projections, and Nvidia’s business is far larger in dollars.
What did Nvidia report for its latest quarter?
For the quarter ended July 26, 2026, Nvidia reported $96.2 billion in revenue, up 106 percent year over year and 18 percent from the prior quarter. Data Center revenue was about $89 billion, up 117 percent, and guidance for the next quarter is roughly $108 billion.
Who uses Google TPUs?
Reported customers include Anthropic, Apple, Physical Intelligence, Citadel Securities, Jump Trading, Caltech, Cassava and Cirrascale. Google Cloud CEO Thomas Kurian said nearly 100 customers use TPUs today.
What does the Oracle announcement change?
Oracle said it would offer Google TPUs, which lets customers use them outside Google’s own infrastructure for the first time. Terms and pricing were not disclosed in the reports we reviewed, and availability on Oracle Cloud Infrastructure is reportedly planned for calendar year 2027.
How many TPUs did Anthropic agree to use?
Reports say Anthropic agreed to use as many as 1 million TPUs in a deal valued in the tens of billions of dollars. The exact contract value and final quantity are not specified.
Should buyers switch from Nvidia GPUs to TPUs?
The sources offer no head-to-head benchmarks, prices or shipment data, so there is no basis for a blanket answer. Test your own workload on each platform and compare written quotes before committing.