Three AI earnings, one big question: who keeps the profits?

Key takeaways

• AMD, Arista and Supermicro show that AI infrastructure demand remains strong, despite recent market doubts.
• Their different profit margins show why growing sales do not translate equally into shareholder returns.
• Watch customer spending, competitive pressure and cash generation, not just the next upbeat forecast.

The AI boom has a long shopping list: powerful chips, servers to house them and networks to connect them. But who makes the most money?

On 3 November 2026, AMD, Arista Networks and Super Micro Computer delivered better-than-expected results (versus Bloomberg consensus) and optimistic forecasts. Investors welcomed them after doubts about AI spending. Demand looks healthy, but profits are another matter.

Three pieces of the same puzzle

AMD makes processors for AI systems and computers. Data-centre sales almost doubled, while its forecast exceeded expectations. AMD is gaining ground, although Nvidia remains the leading supplier of specialised AI chips. AMD also challenges Intel in conventional server processors, giving it more than one route to growth.

Arista supplies the networks that allow thousands of processors to communicate. Bigger AI systems need faster connections. Its strong growth and improved forecast show that the boom extends beyond chips. Its networks can support chips from different manufacturers, rather than depend on one winner.

Supermicro assembles chips and cooling equipment into ready-to-use servers. Sales more than tripled from a year earlier, and management raised its annual forecast. Together, these results show customers still ordering across the infrastructure chain.

The same boom, very different businesses

Strong demand does not create equal profits. Consider gross margin, the share of sales remaining after direct production costs. For each dollar of revenue, Arista kept roughly 64 cents, AMD 54 cents and Supermicro 13 cents. These are not final profits, but the gap matters.

Arista combines hardware with network-management software. AMD designs valuable chips but competes with Nvidia and its established software ecosystem. Supermicro operates closer to assembly, where competition can leave thinner margins. Arista also relies on spending by a relatively small number of powerful cloud customers.

One customer accounted for 44% of Supermicro's quarterly sales. That creates risk if orders slow. Its margin also fell from the previous quarter, while operations consumed cash. Sales are not the same as cash in the bank.

Who pays for the AI shopping list?

These results suggest AI investment remains robust. They cannot prove whether the cloud companies funding it will earn attractive returns. Slower AI adoption, rising financing costs or disappointing returns for customers could eventually curtail orders. Memory shortages may also raise costs for other technology businesses. A strong forecast cannot guarantee returns when share prices already assume rapid growth.

Investor playbook

• If AI spending keeps rising: compare profit margins and cash flow alongside sales growth.
• If spending slows: examine dependence on a few customers and the strength of each company's competitive position.
• If share prices race ahead: ask how much future growth investors already expect.

The receipts matter more than the shopping list

The three reports confirm that AI's shopping list is still getting longer. That is encouraging for suppliers and explains why investors welcomed the forecasts.

But the important question for investors is not simply who receives the biggest order. It is who can deliver it profitably, withstand competition and turn sales into cash over many years.

AMD, Arista and Supermicro occupy different parts of the same industry, yet their economics are far from identical. The AI boom can be real, while the investment outcomes remain very different.

The shopping list is growing. The receipts still deserve a careful read.