Alibaba’s AI Chip Challenge: A Threat to Nvidia?

Alibaba's newest AI chip comes at a time when Nvidia says it is effectively shut out of China's data-centre compute market. That makes the Zhenwu V900 a credible strategic threat to Nvidia's future business there, even before it has proved itself in mass production. For Alibaba shareholders, though, a different question is just as pressing: will stronger chips and fast-growing cloud sales earn enough to cover the company's rising spending and recent share dilution?

Let's break down where the V900 could challenge Nvidia, what Alibaba is earning from AI today and how investors can judge the valuation of Alibaba stock.

What did Alibaba announce with the Zhenwu V900 AI chip?

At its Apsara conference on 22 September 2026, Alibaba introduced the Zhenwu V900, developed by its T-Head semiconductor unit. The company says it delivers three times the performance of the Zhenwu M890 announced in May. The V900 is designed for both training AI models and running them for users, a process known as inference. Alibaba expects mass production and commercial availability in the first quarter of 2027, according to Reuters' account of the conference. That timetable matters: unveiling a chip is different from producing and deploying it at scale.

Reported specifications put the V900's memory at 216 GB and its inter-chip bandwidth at 1,200 GB per second, up from 144 GB and 800 GB per second for the M890. More memory lets a chip hold more of a model or its working data nearby. Faster connections help many chips cooperate when a model is too large for one chip. Alibaba also described a system architecture intended to link large numbers of its chips, including a potential cluster of up to 500,000 accelerators. The cluster figure is an engineering ambition, not a count of V900 chips already deployed.

The three-times figure compares the V900 with Alibaba's own M890. There is no independently comparable V900-versus-Nvidia result in the material reviewed, so a headline about three times the speed cannot establish which supplier delivers a better price per AI task. The 20 GW figure is a long-term capacity target, rather than capacity already online or a revenue forecast. Sources: Alibaba's May chip announcement; Reuters' report on the September Apsara conference; T-Head specifications reported by TechNode.

Alibaba's chip launch came alongside a broader AI roadmap. Its current Qwen3.8-Max model has 2.4 trillion parameters and the company says future models could scale to 5 trillion to 10 trillion. Larger parameter counts can raise computing requirements, though model size alone does not establish that a model is better. The strategic connection is straightforward: Alibaba wants to design the models, the hardware they run on and the cloud service through which customers use them.

Why is Alibaba building its own AI chips?

Alibaba sells computing power to other companies. When AI customers need more of it, the cloud provider must obtain accelerators, servers, networking equipment and electricity before it can collect the resulting revenue. A reliable in-house chip could give Alibaba another source of supply and more control over the cost of each AI task. US restrictions on exports of advanced AI hardware to China make that control especially valuable, although designing a chip does not by itself solve manufacturing and component constraints.

There are two possible benefits, and they should be kept separate. Alibaba may use V900 chips inside its own data centres to improve cloud margins or expand capacity. It may also sell chips or chip-based services to outside customers. An internal deployment can improve economics without appearing as a separate chip sale in consolidated revenue. That distinction is important when judging the launch against Alibaba's financial results.

This is also why a chip should be evaluated as part of a system. Customers care about how quickly a model completes a task, whether the software runs reliably and how much they pay. A strong processor can disappoint if the network between chips is slow, software support is weak or the supply is too small. Alibaba's earlier M890 has a more meaningful commercial record than the newly announced V900: in its August earnings release, Alibaba said its Zhenwu chips, including the M890, were being used through its cloud services by more than 650 external customers across over 20 industries. The company has not yet shown that the V900 will reach the same level of deployment.

Is Alibaba's AI cloud growth turning into profit?

The June 2026 quarter gives the best starting point. Alibaba changed its reporting structure to combine Cloud Intelligence Group and T-Head into AI Cloud and Compute Services, so the segment captures more than a stand-alone chip business. Its revenue rose 45% year on year to RMB48.44 billion and its adjusted operating profit, or adjusted EBITA, more than doubled to RMB5.63 billion. Within that segment, Alibaba reported RMB12.38 billion of AI-related product revenue. These numbers come from its 20 August 2026 earnings release.

Calculations are rounded from Alibaba's June-quarter earnings release. Adjusted EBITA is a company-defined, non-GAAP measure. Segment revenue shares do not add directly to 100% because Alibaba reports inter-segment eliminations.

There is a real operating improvement in the cloud segment. But looking only at its RMB5.63 billion adjusted EBITA would miss the wider cost of Alibaba's AI strategy. AI Labs and Applications reported an adjusted EBITA loss of RMB13.86 billion in the same quarter, while capital expenditure reached RMB67.68 billion across the group. Those amounts represent different accounting categories and should not simply be subtracted from one another. Together they explain why cloud growth alone does not settle the investment case.

Alibaba's group net income fell 75% year on year to RMB10.44 billion. That decline should not be attributed entirely to the new chip or AI investment: the company's filing also identifies lower investment gains and other factors. Free cash flow, however, is harder to overlook. It was negative RMB44.67 billion versus negative RMB18.82 billion a year earlier, with Alibaba citing higher cloud infrastructure expenditure as the main reason for the deterioration. The company had RMB474.51 billion in cash and other liquid investments at the end of June, providing funding capacity, although that figure is not the same as net cash after all obligations.

Is Alibaba's Zhenwu V900 a threat to Nvidia?

In China, yes, potentially. For Nvidia's global business, the V900 has yet to demonstrate a comparable threat. The distinction comes from Nvidia's own disclosures. In its filing for the quarter ended 26 July 2026, Nvidia said US export controls and Chinese restrictions had effectively prevented it from competing in China's data-centre compute market. The company said the limited H200 units shipped under licences accounted for less than 1% of its latest quarterly data-centre revenue. Its outlook for the following quarter assumed no data-centre compute revenue from China. These statements describe Nvidia's position before the V900 reaches commercial production.

Alibaba therefore is not primarily taking a large stream of current Chinese data-centre chip revenue away from Nvidia. Its more immediate opportunity is to supply Chinese buyers while Nvidia cannot serve them at scale. If Alibaba installs its hardware in customers' systems, builds software support around it and trains developers to use it, Nvidia could find it harder to regain those customers even if restrictions later ease. Nvidia's filing explicitly warns that its exclusion from China helps competitors develop customer and developer ecosystems that could challenge it elsewhere. That is a more credible threat than claiming one untested chip has already displaced Nvidia's leading global products.

Nvidia still has scale Alibaba's announcement does not erase. It reported $89.0 billion of global data-centre revenue in its July 2026 quarter, up 117% year on year, and its software and networking products work alongside its chips. The comparison is also between different businesses: Nvidia primarily sells AI infrastructure to others, whereas Alibaba earns much of its AI revenue by operating cloud infrastructure for customers. Alibaba can benefit from using an in-house chip even if it never sells chips at anything resembling Nvidia's volume.

The V900's ability to compete in China will turn on four practical tests. First is manufacturing: can T-Head get enough usable chips, advanced memory and packaging at a competitive cost? Second is software: will customers' models run efficiently without expensive rewriting? Third is system performance: how does a full cluster handle long training runs and high-volume inference, rather than a single-chip demonstration? Fourth is economics: what does an actual completed AI task cost the customer after power, networking and utilization? We have no verified public answer to all four for the V900 as of 23 September 2026.

Our judgment is that Alibaba threatens Nvidia most in China's future installed base. For global markets, Nvidia's disclosed sales momentum remains far stronger than the evidence available for a chip scheduled for release in 2027. This is a meaningful strategic challenge, with a longer path to becoming a measurable threat to Nvidia's worldwide earnings. Sources: Nvidia's August 2026 fiscal Q2 earnings release and Form 10-Q; Alibaba's August 2026 results; Reuters' September Apsara report.

Is Alibaba stock fairly valued after its AI spending and share issue?

The chip story needs to be measured against the whole company. In the June quarter, AI Cloud and Compute Services produced RMB48.44 billion of revenue, while Alibaba E-commerce Group produced RMB205.86 billion. Cloud is growing much faster, but commerce still produces the bulk of sales and RMB39.75 billion of adjusted EBITA, compared with cloud's RMB5.63 billion. A forecast that values Alibaba as if it were only an AI chip company ignores the business that currently supports most of its operating profit.

The balance sheet and the share count complicate a simple 'cheap or expensive' label. Alibaba finished June with RMB474.51 billion of cash and other liquid investments, yet its free cash flow was negative RMB44.67 billion for the quarter. It had previously committed to invest at least RMB380 billion in AI and cloud infrastructure over three years. In August, it issued 710 million new ordinary shares for HK$80 billion gross proceeds to fund AI infrastructure. The Hong Kong exchange filing says that issue represented 3.57% of the enlarged share count. It supplied cash to build capacity but also reduced an existing shareholder's proportional claim on future earnings.

Here is an indicative valuation snapshot using a $116.31 US share price quoted for 22 September 2026, the 19.885 billion ordinary shares reported immediately after the placement and Alibaba's conversion of eight ordinary shares per US-listed ADS. Dividing the ordinary shares by eight gives about 2.486 billion ADS-equivalent units; multiplying by $116.31 gives an approximate $289 billion equity value. This is a dated snapshot, not a live valuation, and the ADS calculation uses the disclosed post-placement share count before any subsequent changes.

Sources: Alibaba's FY2026 and June 2026 earnings releases; its 26 August 2026 Hong Kong share-placement filing; a 22 September US share-price quote carried by StockAnalysis. 

Why do these earnings multiples disagree? For FY2026, Alibaba's reported earnings included investment gains and other items that its non-GAAP calculation removes. The subsequent weak June quarter changes the last-12-month picture again. A stock that looks inexpensive at 18 times an older full year's reported earnings looks less obviously inexpensive at around 27 times updated reported earnings, while its latest free cash flow is negative. None of these measures is a definitive estimate of what Alibaba will earn once its AI infrastructure matures.

To understand what the valuation requires, start with what cloud contributes now. Its June-quarter adjusted EBITA margin was about 11.6%. If future quarterly cloud revenue reached RMB60 billion at an 18% margin, adjusted EBITA would be RMB10.8 billion. That is approximately RMB5.2 billion above June's RMB5.63 billion. At the June filing's exchange rate, this would represent less than $1 billion of extra quarterly adjusted segment profit, before considering the cost of AI Labs, group expenses, taxes and future capital spending. A single good quarter would not transform a roughly $289 billion equity valuation.

The future rows illustrate arithmetic, not company guidance or share-price targets. The scenarios do not subtract AI Labs' RMB13.86 billion June-quarter adjusted EBITA loss and do not estimate capital expenditure, cash flow or returns for the entire group.

The practical valuation test is whether Alibaba can increase profit per unit of capital committed. If its own chips cost less to acquire or let it serve more requests with the same equipment, cloud margins could improve. If the V900 needs costly capacity, high-priced components or low utilization, faster cloud revenue may still leave poor cash returns. Investors should also judge the new share issue against the extra earnings it eventually creates: expanding earnings per share is a higher bar after issuing more shares.

A useful balance-sheet perspective is that the June figure of RMB474.51 billion is cash and liquid investments, not net cash or freely distributable surplus. Debt, customer obligations and the spending already planned matter. Equally, Alibaba's negative free cash flow reflects both AI infrastructure and spending in other businesses, so it cannot be allocated entirely to V900 development. We would give the chip a higher valuation weight only after Alibaba demonstrates scaled deployments, rising cloud margins and an improvement in group free cash flow.

What does Alibaba's AI chip mean for Indian investors?

For an Indian investor looking at the US-listed BABA shares, this is exposure to a Chinese e-commerce and cloud company undertaking an expensive AI expansion. Its chip gains must be assessed alongside its much larger commerce operations, new shares issued to finance construction and the cash needs of the wider group. The US listing also exposes an Indian holder to movements in the rupee against the dollar, while Alibaba's underlying business remains exposed to China's economic and technology policies.

For someone following Nvidia, Alibaba's progress is most relevant to the possibility of a lasting domestic chip ecosystem in China. Nvidia's July-quarter global data-centre revenue shows that its near-term earnings story extends well beyond that market. Future changes to US export permissions and China's purchasing restrictions could change the size of the contest, so the V900 should be read in that wider market context.

What should investors watch after the Zhenwu V900 launch?

The first proof point is whether Alibaba meets its Q1 2027 V900 commercial-release target, reports actual shipment volumes and shows customer deployments beyond the older M890. The next is comparative cost per AI task in a full system, ideally with verifiable tests across training and inference. In subsequent financial results, the decisive evidence will be external cloud revenue, adjusted EBITA margin, AI Labs losses, capital expenditure, free cash flow and earnings per share after the placement.

Our assessment as of 23 September 2026 is that the V900 makes Alibaba a more serious prospective competitor for China's AI computing business and could complicate Nvidia's eventual return there. Alibaba's 45% cloud-segment growth and improving cloud profit give that strategic claim a commercial base. At the same time, its negative free cash flow, large AI investment programme and August share issue make BABA's valuation more demanding than an old earnings multiple suggests. The case becomes more convincing when V900 deployments raise cloud profits and the wider company begins turning those profits into sustainable cash for each share outstanding.