Alphabet and Amazon Keep Buying Nvidia Chips Despite Their Own Silicon

Nvidia shares fell 18.29% over the past week, even as new earnings reports from Alphabet and Amazon showed both companies still buying Nvidia systems at enormous scale. That combination — a falling stock and rising commitments — is the story behind three quarterly reports that describe one connected business.

Nvidia (NASDAQ:NVDA), Alphabet (NASDAQ:GOOGL) and Amazon (NASDAQ:AMZN) have all now reported earnings. Google sells its own TPU chips. Amazon keeps expanding Trainium. Both keep paying Nvidia anyway, and the numbers explain why the AI hardware market still runs through one supplier.

How big are the cloud businesses that feed Nvidia?

Google Cloud revenue rose 103% to $30.71 billion, making it Alphabet's main growth engine. CEO Sundar Pichai pointed to "strong demand for our TPU systems across a wide range of customers." That demand has not displaced Nvidia. Nvidia lists Google Cloud among the sites where its next-generation Vera Rubin platform is in full production. Google sells its own chips while running Nvidia's in the same data centers.

AWS tells a similar story from a different angle. Revenue rose 39% to $45.98 billion. Most of Amazon's Bedrock AI service runs on Trainium, Amazon's in-house accelerator. CEO Andy Jassy said custom-silicon customers saved more than $2 billion in Q3 alone. That savings gives Amazon real leverage over its suppliers.

Even so, AWS will deploy an additional 2 million GPUs from Nvidia through the second quarter of Nvidia's fiscal 2029. Jassy noted that all advanced chips come from one foundry and "everybody gets an allocation," which limits how fast Trainium can scale. The single-foundry constraint — a nod to Taiwan Semiconductor Manufacturing — caps even the most ambitious custom-silicon plans.

The scale of the spending shows up in the capital expenditure lines. Alphabet posted capex of $51.36 billion for the quarter. Amazon posted capex of $77.26 billion. Nvidia's hyperscale revenue reached $49 billion.

Business Driver	Nvidia	Alphabet	Amazon
AI Engine	Vera Rubin, Blackwell Ultra	Google Cloud, TPU systems	AWS, Bedrock, Trainium
Key Quarterly Figure	Hyperscale revenue $49 billion	Capex $51.36 billion	Capex $77.26 billion

Why do custom chips fail to break CUDA's grip?

Jensen Huang framed the rivalry narrowly. "Many of these XPUs are inference-specific chips for one cloud or one service," the Nvidia CEO said of his competitors' silicon. His argument: one platform can serve every cloud and every phase of AI work, from training to inference.

Training frontier models takes the densest compute and fastest interconnects available. Most of the machine learning world writes its code for CUDA, Nvidia's software platform. Cloud providers have to offer the hardware developers ask for, regardless of what they build in-house.

The custom chips do serve a purpose — just not the one headlines suggest. For Amazon, Trainium is mainly about protecting margins. AWS posted a 38.7% operating margin, and cheaper in-house silicon helps defend it.

Alphabet's spending is harder to absorb. The company's free cash flow came in at negative $9.07 billion. It issued $28.7 billion in new notes. When a company of Alphabet's size borrows while spending at that pace, the strain becomes visible in the balance sheet rather than the income statement.

What does the Vera Rubin ramp reveal?

Nvidia forecast Q3 revenue of $108.0 billion, plus or minus 2%. It expects Vera Rubin to make up about 20% of data center revenue as the platform ramps. Huang described his fiscal 2028 growth outlook of about 70% as limited by supply, not demand.

That supply constraint matters for the whole market. If Nvidia cannot make enough Vera Rubin systems to meet orders from Google, Amazon and other hyperscalers, the incentive to qualify alternative silicon grows. If supply loosens, CUDA's gravitational pull strengthens.

The open question is timing. Amazon says it will start selling its AI chips to other companies in late 2027. Google sells TPU systems to external customers already. Both moves could pull workloads away from CUDA over time — if external Trainium sales gain real traction and if Google's TPU sales keep compounding.

What do the valuations say about the three-way bet?

After a 26.34% drop over the past month, Nvidia trades near 34x earnings. Amazon sits near 38x. Alphabet trades near 13x, the lowest earnings multiple of the three.

Each valuation rests on a different assumption. Nvidia's depends on the Vera Rubin ramp and on hyperscalers continuing to buy at scale. Amazon's leans on Jassy's forecast of AWS growth above 40% in both 2027 and 2028. Alphabet's discount reflects, in part, an $8.70 billion loss on equity securities that sits alongside otherwise strong operating results.

The dependencies run in every direction. Google needs Nvidia hardware to serve customers who want it, even as it competes with TPUs. Amazon needs Nvidia GPUs by the million, even as Trainium absorbs Bedrock's core workloads. Nvidia needs both companies' capex — $51.36 billion and $77.26 billion in a single quarter — to keep flowing into its data center products.

Where does this leave the AI hardware market?

Nvidia remains the supplier both giants keep paying, through a stock decline of 18.29% in a week and 26.34% in a month. The report's conclusion is conditional rather than permanent: Alphabet and Amazon will keep putting Nvidia hardware in their data centers until external Trainium sales gain real traction. Late 2027, when Amazon starts selling its chips to outside customers, is the first date that could test the thesis.

Until then, the market's structure holds: one chip supplier with a software moat, two cloud giants building their own silicon for margin rather than independence, and a single foundry allocating capacity to everyone at once.