India AI Digest — Thursday, May 28, 2026

• Cognition (maker of the Devin coding agent) raised $1 billion at a $26 billion valuation, per a Bloomberg report on May 27, sharpening the pressure on the Indian IT-services replacement-vs-augmentation narrative.
• Avendus Capital's third annual data-centre report projects India's built data-centre capacity to grow from 1.6 GW to ~5 GW by 2030, with AI workloads driving 650,000–700,000 GPU deployments and ~$23 billion of investment.

---

FUNDING · DEV TOOLS · SERVICES · May 27, 2026

Cognition raises $1B at $26B valuation; the agentic-coding capital curve steepens against the Indian IT-services book

Bloomberg reported on May 27, 2026 that Cognition AI, maker of the Devin autonomous-coding agent, raised $1 billion in new financing at a $26 billion post-money valuation. The round is among the largest single financings disclosed for a US agentic-coding company so far in 2026, and the largest disclosed for Cognition specifically since its $400 million September 2025 round at $10.2 billion post-money. It follows a sequence of frontier-lab and integrator moves through May aimed at landing autonomous-coding agents inside enterprise engineering workflows. Bloomberg's reporting discloses the lead syndicate (Lux Capital, General Catalyst, 8VC, with Ribbit, Atreides, Founders Fund participating), a revenue run rate of ~$492 million up 13x from $37 million in May 2025, and named enterprise customers including Goldman Sachs, Mercedes-Benz and parts of the US federal government. What is not disclosed: per-seat pricing, contract sizes, renewal rates, and any India go-to-market posture.

What this means. The capital is real; the deployment evidence is not yet at the scale the valuation implies. A $26 billion mark on an autonomous-coding product is venture conviction that the per-task pricing of an AI engineer is a replacement substrate for the billable-hour pricing of a human one — at enterprise scale, across the offshore-services cost base, on a horizon investors believe is now inside the next two to three years. The structural reading is that frontier capability has finally landed on a product surface (the agent) where the cost displacement is legible to enterprise procurement, not just to engineering buyers. The conservative reading is that the ~53x ARR multiple at $26B valuation against the $492M run-rate sits above comparable enterprise-software multiples; contract sizes, per-customer concentration, and renewal rates that would discipline that multiple are not in the public record at the round close. Both readings sit on the same fact: the capital is now in place to fund enterprise GTM at a scale that closed-API agentic-coding offerings have not previously had access to.

The competitive frame around the round is also worth marking. The two convergent moves of May — OpenAI's Dell on-prem and hybrid Codex distribution announced May 19 (covered in the May 20 digest), and the wider Anthropic enterprise-distribution stack via AWS, PwC, KPMG, and the Blackstone-Goldman-H&F venture — describe the major frontier labs building procurement substrate for agentic-coding deployment. Cognition's raise is the standalone-product alternative to the same end state: rather than ride a hyperscaler or consultancy channel, fund the direct GTM. Whether that direct path scales as fast as the hyperscaler-channel path is the next 12–18 months' empirical question.

India angle. The Indian IT-services majors — TCS, Infosys, Wipro, HCLTech, LTIMindtree — are the cohort most directly exposed to the unit-economics displacement an autonomous-coding agent at this funding scale implies. Together they employ roughly 1.5–1.6 million people, the bulk in software engineering and application-development roles where labour-arbitrage pricing has been the structural moat for two decades. A per-task agent priced at venture scale compresses the margin under the wage arbitrage. Two responses are visible in the cohort so far. LTIMindtree announced its ~€160 million offer for Randstad's Europe-and-Australia technology and consulting business on May 22 (covered in the May 23 digest), the most explicit inorganic move up the AI-services stack by the cohort this year. The other majors have so far stayed on organic AI-services repositioning through quarterly investor disclosures rather than M&A.

For India's coding-tools and developer-experience cohort — the Indian-headquartered companies building AI-assisted IDE, code-review, and engineering-workflow products — Cognition's raise sets a global ceiling on what a category-leading agentic-coding product can fund-raise in 2026. None of the Indian dev-tools cohort is yet at frontier-agent scale; the question is whether the funding signal pulls Indian investors toward the category at adjacent layers (review, security, observability for agent-generated code) where the build-cost is materially lower than full agent IP.

Behind the news. The enterprise-AI distribution arc through May has had two consistent through-lines: frontier labs competing on procurement substrate rather than model, and the Indian IT-services cohort visibly responding through inorganic and organic AI-services repositioning. The Cognition raise sits on the supply side of the same arc. Devin-class products, named in the April 12 digest as a comparison set when discussing Indian agentic-coding product gaps and again in the April 20 digest on the open-weights side, have moved in two months from research-stage benchmarks to a $26 billion private-market mark. The cohort-level question for Indian IT services is whether this changes the speed at which AI-services revenue must replace volume-led revenue, not whether the displacement is coming.

What to watch. Quarterly investor reporting from TCS, Infosys, Wipro, HCLTech, and LTIMindtree across the FY27 Q1 results window (July–August 2026) — specifically, whether AI-services revenue is broken out as a distinct line item, and whether any of the five names an autonomous-coding-agent partnership or in-house competing product in the same disclosure cycle. Absent any such named move from the cohort by end of Q2 FY27, the bear thesis on the SI book — that the cohort defers the repositioning question rather than answering it — gains a quarter's worth of evidence.

Source: Bloomberg, May 27, 2026. → link

Confidence: high on the round announcement and the company identity (single Bloomberg report; no Cognition press release at the time of this digest). Medium on the strategic-displacement framing — directional from the funding signal, not from deployment evidence. Low on Indian go-to-market specifics — none disclosed.

---

COMPUTE · INFRA · FUNDING · May 27, 2026

Avendus Capital projects 700,000 GPUs and ~$23 billion India AI data-centre opportunity by 2030

Avendus Capital released its third annual India data-centre report on May 27, 2026, summarised across Business Standard, IANS, TechStory, and tele.net.in (the underlying report has not been posted on Avendus's public site at the time of this digest). The headline projections: India's built data-centre capacity grows from 1.6 GW in 2025 to roughly 5 GW by 2030 (~26% CAGR), AI workloads drive deployment of 650,000–700,000 GPUs and roughly $23 billion of investment over the same window, 1 GW of AI capacity is already in the construction pipeline, three new subsea cables are expected to land in Chennai across 2026–27, Mumbai retains roughly half of installed and upcoming capacity, and three to four data-centre or AI-infrastructure IPOs are expected within three years to fund the build-out.

What this means. This is a third-party projection from an investment bank with a visible book in the same sector, not an operator-reported capacity tally and not a government build-plan. Read it as a financing thesis with infrastructure scaffolding rather than a measured forecast. The 1.6-to-5-GW capacity range is consistent with operator-disclosed expansion pipelines from CtrlS, Yotta, AdaniConneX, NTT India, STT, and Nxtra over the past 18 months, but the conversion of pipeline capacity to live workloads is the lever that bends the curve up or down — and that conversion has historically slipped in India on power, land, and water-cooling permitting. The 700,000-GPU figure depends on the AI-workload share of those additions converting at hyperscaler-comparable utilisation, which is itself a function of demand from Indian enterprise deployments, captive global-product workloads landing in India (the Uber-Adani facility announced May 13 is one anchor of that type, covered in the May 15 digest), and the IndiaAI Mission's GPU-allocation discipline.

The financing-narrative side of the report is the part most likely to carry weight independent of the capacity projection. Avendus is an investment bank; signalling 3–4 IPOs over three years from the Indian DC and AI-infrastructure cohort is the part of the report that is closest to the bank's own commercial activity, and is the most readable forward signal in the document. If two of those three-to-four IPOs price within the next 24 months, the financing thesis is validated regardless of whether the 5-GW figure lands by 2030 or by 2032.

India angle. Three reads, by layer. For the Indian colocation and DC-operator cohort — CtrlS, Yotta, AdaniConneX, NTT India, STT, Nxtra, ESDS, Web Werks — the report frames the demand-side ramp the operators have been pricing into capacity expansion through 2025–26. The report is not new information for these operators; it is third-party validation that the financing market is willing to underwrite at the projected scale. The Competition Commission's May 20 clearance of Blackstone's controlling stake in Neysa Networks (covered in the May 22 digest) is the most recent concrete data point on the financing side of that thesis — $600 million of equity into a domestic AI-cloud build, against the financing narrative Avendus is now formalising as a multi-IPO horizon.

For the IndiaAI Mission and MeitY allocation discipline, the report's 700,000-GPU figure is a benchmark against which the Mission's empanelled-provider GPU capacity (currently in the tens of thousands, layered across Yotta, Tata, NeevCloud, and others through 2025–26 tendering) will be compared. If the Avendus projection holds even at the low end, the IndiaAI Mission accounts for a single-digit-percent share of national AI compute by 2030; the rest is private-sector capacity built on private-sector contracting. Whether the Mission's sovereign-compute framing is read as catalytic or as residual depends on which side of that comparison the policy framing eventually lands.

For BFSI, public-sector, and healthcare buyers with residency-bound workloads, more domestic AI capacity is the substrate that makes DPDP-bound and sector-specific data-handling rules workable for AI deployments at scale. The Chennai subsea-cable landings, if they materialise on the 2026–27 timeline, add east-coast connectivity redundancy that the current Mumbai-concentrated landing-station map does not provide — a connectivity-geopolitics signal independent of the GPU-count headline.

Behind the news. The Indian DC and AI-infrastructure thread through May has consolidated around foreign-platform compute localisation (Uber-Adani on May 13 in the May 15 digest) and large private-capital tickets clearing regulatory review (Blackstone-Neysa on May 20 in the May 22 digest). The Avendus report formalises what each of those individual items has been pointing toward: the private capital is being underwritten on a multi-year ramp, not on single-tenant deals. The 26% CAGR figure is closer to the operator-pipeline reading than to a sovereign-compute push driven by the IndiaAI Mission alone.

What to watch. First, the listing pipeline — DRHPs filed by Indian DC or pure-play AI-infrastructure operators over the next 12 months would convert the IPO projection from a banker's slide to a market reality. Second, the Chennai subsea-cable landings; named operators and timelines from TIC, Bharti Airtel-led consortia, or Reliance Jio for the three projected systems will determine whether the 2026–27 window holds. Third, the next semi-annual capacity-disclosure cycle from the named domestic operators (Q2 FY27, expected November 2026) against the Avendus 5-GW glide path.

Source: Business Standard, May 27, 2026. → link Also: IANS; TechStory; tele.net.in.

Confidence: medium on the projection numbers — the underlying Avendus report has not been posted to a public Avendus URL at the time of this digest; the figures are consistent across four secondary-Indian publications citing the report directly. High on the broader financing-narrative direction. Low on the timeline specificity of the Chennai subsea-cable landings and the IPO window — both are forward-looking projections from a banker's report, not committed pipelines.

---

Position movements