India AI Digest — Friday, July 24, 2026

• Infosys put AI revenues at 8.2% of a $5.08 billion quarter — the third Indian services major to publish an AI-revenue figure this earnings season — while trimming FY27 guidance to 1.5–3% constant-currency growth.
• BUSINESSNEXT raised a $40 million Series C from ServiceNow Ventures as sole investor at a reported $700 million valuation, funding an autonomous-banking push across Asia Pacific.
• AMD and Cerebras announced a disaggregated-inference partnership pairing Helios racks with the Wafer-Scale Engine, claiming up to 5x tokens per second per watt, available first through Cerebras Cloud in H2 2026.

---

SERVICES · ENTERPRISE · July 23, 2026

Infosys puts AI at 8.2% of revenue as it trims FY27 guidance to 1.5–3%

Infosys reported Q1 FY27 results on July 23: revenue of $5,082 million, up 2.4% year-on-year in constant currency, operating margin of 21.1%, and large-deal TCV of $3.6 billion with 61% net new. The number the company put on AI: revenues from AI offerings at 8.2% of total — roughly $417 million in the quarter. FY27 guidance was revised to 1.5–3.0% constant-currency growth, trimmed from the prior 1.5–3.5% range. As with every AI-revenue figure this season, the metric is company-defined; there is no standard for what counts as AI revenue in a services book. Analytics India Magazine reads the guidance as lowered on uncertain demand.

What this means. Infosys is the third Indian services major to publish an AI-revenue figure in the Q1 FY27 season, and the third to define the metric differently. TCS reports an annualized run rate ($2.6 billion). HCLTech reports an advanced-AI revenue line ($171 million). Infosys reports a share of revenue (8.2%). Each figure is directionally informative and none is comparable to the others — investors now have three disclosures and no common denominator.

The more interesting tension is inside Infosys's own print. An 8.2% AI share with 61% of large-deal TCV net new sits against a top line guided to grow 1.5–3.0%. AI revenue is growing inside a nearly flat book. Two readings circulate: AI work is genuinely additive and the rest of the book is shrinking under demand softness, or AI is partly a relabeling of work the book already contained. The disclosure as published cannot distinguish the two, and Infosys has not broken out how the 8.2% is counted.

India angle. For the Indian SI layer, the season's pattern is now established: publishing an AI number has become the disclosure norm within one earnings cycle, after none of these firms reported one a year ago. The pressure this puts on the remaining large firms is mechanical — investors will ask on every call. The definitional looseness is the cost of speed: three firms, three metrics, no auditor. Whether the disclosures converge on a comparable standard, or stay as marketing-adjacent directional indicators, is the open structural question for how Indian IT's AI transition gets measured.

Behind the news. The July 12 digest's TCS item flagged exactly this as the thing to watch — whether Infosys, Wipro, or HCLTech would publish a comparable AI-revenue metric. HCLTech answered on July 13 with its $171 million advanced-AI line, up 62% year-on-year. Infosys's 8.2% is the third datapoint in eleven days. The disclosures are stacking; the definitions are not.

What to watch. Infosys Q2 FY27 results in October — whether the 8.2% metric repeats as a standing disclosure and whether the company discloses how AI revenue is counted. A second print makes it a series; a definition makes it comparable.

See also: TCS reports a $2.6B annualized AI revenue run rate in Q1 FY27, HCLTech books $171M in advanced-AI revenue

Source: Infosys Q1 FY27 press release, July 23, 2026; Analytics India Magazine coverage. → link

Confidence: medium. Financial figures are company-reported and corroborated via press-release syndication; the AI-revenue share is a company-defined metric without a published definition.

---

FUNDING · BFSI · July 22, 2026

BUSINESSNEXT raises $40M Series C from ServiceNow Ventures at a reported $700M valuation

BUSINESSNEXT, the financial-services CRM and AI platform with CRMNEXT lineage, raised $40 million in Series C funding from ServiceNow Ventures as sole investor, at a $700 million valuation per press reporting (Business Standard, Inc42). The stated use of funds: R&D and expansion into Australia and New Zealand, around what the company frames as an autonomous-banking push across Asia Pacific. The company's announcement confirms the $40 million round and the autonomous-banking framing; the $700 million valuation and ServiceNow's roughly 5% stake come from press reporting (TechCrunch, Inc42), not the announcement itself.

What this means. The structure is the signal. A US enterprise-software strategic taking a sole-investor growth-stage position is a different transaction from a financial-VC round: ServiceNow is buying proximity to an Indian BFSI AI platform with an Asia-Pacific book, not just a return. It is also ServiceNow's second India-facing move in two months, after the expanded Wipro agentic-AI partnership in late May — a pattern of the platform placing itself into the Indian enterprise-AI delivery chain from two directions, integrator and application vendor.

For BUSINESSNEXT, the autonomous-banking framing is positioning ahead of product evidence. The company has a shipped CRM base in Indian and Asian banks; what "autonomous banking" adds beyond that base is not specified in the coverage. The capital is real and the strategic alignment is real. The category claim rides on both.

India angle. Fintech remains the most mature Indian AI sector by deployment, and this round adds strategic validation at growth stage — a segment where Indian applied-AI companies have had fewer datapoints than at seed. A $700 million valuation from a strategic buyer of enterprise workflows is also a reference point for the cohort of Indian BFSI software firms weighing platform partnerships against independence.

Behind the news. ServiceNow's India thread is two months old and two moves deep: the expanded Wipro agentic-AI partnership announced May 28, now strategic capital into an application vendor. The round also extends the funding acceleration documented in H1 — $676 million across 57 deals, up more than 4x year-on-year — into H2 and into growth stage.

What to watch. Whether BUSINESSNEXT surfaces in ServiceNow's platform catalog or joint product announcements over the next two quarters — the concrete test of whether the strategic stake converts into distribution rather than sitting as a balance-sheet position.

See also: Wipro and ServiceNow widen agentic-AI tie-up, Indian AI startup funding quadruples to $676M in H1 2026

Source: Business Standard, July 22, 2026; Inc42. → link

Confidence: medium. Round size and investor are company-announced; the $700 million valuation and roughly 5% stake are from press reporting (TechCrunch, Inc42); announcement date is July 22 per Business Standard with some coverage dated July 23.

---

COMPUTE · INFRA · July 23, 2026

AMD and Cerebras pair Helios racks with wafer-scale engines for disaggregated inference

AMD and Cerebras announced a partnership on July 23, in matching press releases, combining AMD's Helios rack-scale Instinct GPU infrastructure with the Cerebras Wafer-Scale Engine in a disaggregated inference architecture: Helios handles prompt processing, the Wafer-Scale Engine handles token generation. The companies claim up to 5x higher tokens per second per watt. First availability is through Cerebras Cloud in H2 2026. The 5x figure is a vendor projection; no independent benchmarks exist yet.

What this means. Disaggregated inference — splitting prefill and decode onto hardware optimized for each — has been moving from research architecture to production practice inside the large labs. This is the first prominent cross-vendor productization of the split: two different silicon architectures, each taking the phase it is better shaped for. If the efficiency claim survives independent benchmarking, the architecture bends the serving-cost curve at the layer where most inference spend sits.

The claim deserves its stated conditions. Tokens per second per watt at announcement is a projected figure on unshipped configuration; availability is two quarters out and initially through a single cloud. It also lands in a visible wave — specialized non-NVIDIA inference silicon drawing capital and alliances as inference disaggregates from the NVIDIA training stack. The wave is real; each individual claim in it still needs benchmarks.

India angle. Tokens per second per watt is the denominator of inference unit economics, and Indian workloads — consumer apps at ₹80–200 monthly ARPU ceilings, high-volume enterprise deployments — are among the most cost-sensitive riders of that curve. Cerebras Cloud will be reachable by Indian builders in H2 2026, but neither press release makes any India-region commitment, so residency-constrained BFSI and healthcare workloads are unaffected until regional availability exists. For Indian data-centre operators, a second credible rack-scale inference architecture widens the procurement conversation beyond a single vendor.

Behind the news. The inference-silicon wave has been building through July: SambaNova closed the first $1 billion of a Series F at an $11 billion valuation on July 8, with JPMorganChase named an inference-infrastructure partner. The AMD–Cerebras pairing is the alliance version of the same thesis: inference is a distinct market with distinct hardware economics.

What to watch. Cerebras Cloud availability in H2 2026 with published pricing, and the first independent tokens-per-second-per-watt benchmarks against the 5x claim.

See also: SambaNova closes first $1B of Series F at an $11B valuation

Source: AMD press release, July 23, 2026; Cerebras press release. → link

Confidence: high on the announcement facts; the 5x tokens-per-second-per-watt figure is a vendor projection, not a benchmark.

---

Position movements