Asia’s AI Boom Is Rewriting the Bill

What Indian IT contracts, Korean memory pricing and Alibaba’s financing reveal about who captures AI’s gains.

As of October 26, 2026.

The practical test for an AI business is who gets to rewrite the invoice. A supplier can automate work and discover that its customer expects a lower bill. A chipmaker can produce a more sophisticated device while earning more from a less celebrated product. A cloud company can sell substantially more computing and still seek additional capital.

These situations put the economics of adoption in different hands. Customers negotiate productivity commitments. Manufacturers allocate scarce production resources. Shareholders finance capacity whose eventual returns remain uncertain. The common thread is a contest over prices, contracts and capital, rather than a single verdict on whether AI works.

Three cases illustrate that contest. They concern Indian export services, Korean memory production and Alibaba’s Chinese cloud expansion. Their implications differ for customers, operators and owners. None supplies a shortcut from an impressive technology to an attractive stock.

Indian IT firms’ customers want a share of the efficiency gain

Infosys offers a useful picture of a transition already under way. In its September quarter, company-defined AI revenue reached 10.5% of sales and grew 29% sequentially. Total revenue grew 1.2% sequentially in constant currency, while operating margin was 21%. Its full-year growth guidance was 1–2% in constant currency. These are different measures: subtracting their growth rates would not reveal how much existing work AI displaced. Infosys, October 23, 2026, p. 1

The more revealing evidence concerns negotiations. On its July earnings call, Infosys said customer demands for AI productivity and competitive pressure had limited the price increases it expected. Management described additional AI-related deflation across large and other contracts, with productivity discussions sometimes occurring between renewals. Crucially, it also said net pricing was still increasing. The finding is pressure on the revenue opportunity, not a universal fall in prices. Infosys, July 23, 2026, pp. 12–16

Independent reporting shows the bargaining process extending beyond one company. TCS’s chief executive told Reuters that about 80% of contracts in its finance, human-resources and other business-services segment used outcome measures. That figure does not cover all TCS contracts. Persistent’s chief executive described clients demanding the same work for 25–30% less; those requests were not a measured industry-wide price reduction. Reuters, August 21, 2026

The mechanism matters. When payment depends on people’s time, fewer hours can mean less revenue. When payment depends on an outcome, the provider can retain some efficiency gain—but the customer can negotiate that gain into the next agreement. New implementation work may enlarge the relationship while automation reduces the bill for existing tasks. Gross demand for AI services therefore need not equal incremental spending on the supplier.

That gives buyers a concrete bargaining opportunity: connect productivity promises to price, service quality and responsibility for failures. For suppliers, the valuable capability is delivering a measurable result at a cost the contract can support. Merely attaching an AI label to existing revenue does not answer that commercial question.

An outcome agreement also assigns measurement risk. The parties need a baseline for the work, a definition of an acceptable result and a way to account for implementation costs. A faster task is not necessarily a cheaper business process if checking and repair consume the saving. Conversely, a provider that achieves a dependable result with fewer resources can improve its own economics even after sharing some benefit. The contract determines which party keeps that residual; a productivity benchmark alone cannot.

There is also a substantial countercase. TCS told Reuters that new work had so far offset AI-related revenue deflation. Reuters, August 21, 2026 Outcome pricing can reward a provider that delivers efficiently, and Infosys’s maintained profitability demonstrates that pressure and profit can coexist. Infosys, October 23, 2026, p. 1 Neither observation establishes that vendors keep all the gains.

Nor should every disappointing contract be blamed on AI: Infosys explicitly said the particular European manufacturing deals it declined, and a separate termination, were unrelated to it. Those qualifications prevent a plausible mechanism from becoming an all-purpose explanation. This is principally a story about exported services and overseas clients, rather than a measure of domestic Indian technology adoption. Infosys, July 23, 2026, p. 13, Infosys, October 23, 2026, client geography

Memory profits follow the pricing clock

Memory supplies a different example of bargaining power. Dynamic random-access memory (DRAM) provides working memory for computers and servers; high-bandwidth memory (HBM) is a specialized form used in AI data centres. In January, S&P Global described HBM as carrying higher margins than conventional DRAM, helping explain manufacturers’ preference for it. That was a dated assessment, not a promise that the ranking would last throughout 2026. S&P Global/Visible Alpha, January 28, 2026

By April, Samsung acknowledged that conventional DRAM was showing higher profitability than HBM. Its explanation included the pricing calendar: HBM agreements were negotiated annually, while conventional-memory prices adjusted quarterly. Faster repricing could therefore change the relative economics before the next HBM negotiation. This was reported contemporaneously in Korean, rather than inferred from Samsung’s group profit. Yonhap Infomax, April 30, 2026

Taiwan-based TrendForce supplied an independent research comparison. Using estimates of die size, manufacturing yields and prices per gigabit, it concluded that HBM’s revenue per wafer—and profitability—had been overtaken by DDR5 64GB registered server-memory modules in the first quarter. This concerns a specific server product and an estimated production comparison. It does not mean every conventional chip had a higher margin. TrendForce, June 2, 2026

The distinction helps explain why technological importance and current production returns can separate. A high selling price per unit says little by itself about the manufacturing resources needed to produce that unit. Equally, a product can be essential to a customer’s AI system without repricing as quickly as another constrained component.

For manufacturers, this makes qualified product flexibility economically valuable. Samsung said it would maintain balanced supply despite the profitability difference. Maintaining an HBM relationship can support future business even when another product offers better immediate economics. Maximizing this quarter’s product profit and maximizing the lifetime value of a customer relationship are different decisions. Yonhap Infomax, April 30, 2026

The inversion also carries its own expiry risk. TrendForce’s September outlook anticipated sharply higher blended HBM selling prices in 2027. Product mix and specification changes contribute to that forecast; it cannot be read as an equivalent margin increase. It nevertheless supplies a clear reason not to project the earlier profitability ranking indefinitely. TrendForce, September 29, 2026

For an investor, none of these comparisons establishes return on invested capital or a cheap security. Per-wafer revenue, profit margin and shareholder return have different denominators. For a customer, however, the contract clock is immediately relevant: the date at which a supplier can reset prices can matter as much as today’s quoted price.

Alibaba’s expansion has a financing side

Alibaba shows how commercial progress and capital demands can advance together. June-quarter revenue in AI Cloud and Compute Services rose 45% year on year to RMB48.4 billion, with RMB5.6 billion in adjusted EBITA, its non-GAAP measure of operating earnings. The reorganized segment includes chip business T-Head; it is broader than a pure AI-model revenue measure. Alibaba, August 20, 2026, segment reporting

At group level, capital expenditure of RMB67.7 billion was approximately three times operating cash flow of RMB22.9 billion, by the author’s calculation using unrounded disclosures. Reported non-GAAP free cash flow was negative RMB44.7 billion. These figures do not identify the standalone cloud business’s cash return. They show the scale of investment being financed alongside commercialization. Alibaba, August 20, 2026, cash flows

Operating earnings and a capital budget answer different questions. A business can earn money serving current customers while paying for equipment intended to serve future ones. Financing bridges that timing gap. The economic test is whether the additional capacity eventually generates enough cash to cover its full costs and capital commitments. Neither rising segment earnings nor a negative group cash-flow quarter supplies that answer on its own.

In August, Alibaba completed a placement of 710 million ordinary shares, raising approximately HK$79.7 billion net. From that placement alone, a holder who did not participate saw their proportional ownership decline by 3.57%, calculated from the disclosed share counts. The company also received cash: a smaller ownership percentage is not an equivalent percentage loss of wealth. Alibaba, August 26, 2026, placement

This is the capital-allocation bargain. Existing owners accept a different fraction of the enterprise while the enterprise gains resources to expand. If investment produces sufficiently valuable future cash flows, they can benefit. If competition, costs or further financing absorb those flows, commercial growth can leave them with less than the operating story suggests. Neither outcome follows from issuance alone.

There is no demonstrated funding crisis here. Alibaba reported RMB474.5 billion of cash and other liquid investments at June’s end. Front-loaded spending can finance profitable capacity; a quarter’s outflow cannot settle its lifetime return. Broader research from Rhodium likewise treats financing as one constraint in China’s AI expansion, alongside access to compute. Its financing analysis does not establish that each project will earn poor returns. Alibaba, August 20, 2026, liquidity, Rhodium Group, September 17, 2026

Across these cases, the next useful evidence is specific: how productivity commitments change service revenue, how memory contracts reset relative profitability, and how cloud investment converts into cash available per diluted share. Technology adoption creates the opportunity. Prices, obligations and financing determine how its benefits are divided.