India AI Digest — Sunday, July 5, 2026

• PM Modi inaugurated CG Semi's ₹7,500 crore OSAT facility in Sanand on July 4 — the India Semiconductor Mission's third assembly-and-test plant to enter commercial production this year, with a second unit planned by end-2026.
• Business Today's July 4 tally puts the combined market value of India's five largest IT-services companies at ₹18.15 lakh crore — down more than 46% from the August 2024 record, and now barely above Reliance Industries alone — as markets price AI disruption into the services model.
• Anthropic is moving to close the offshore routes — Singapore subsidiaries, reimbursed VPN subscriptions — through which Chinese companies have been accessing Claude, per Financial Times reporting picked up July 3. Enforcement runs on account telemetry: time zones, usage patterns, identity verification.
• Position movements: compute_infrastructure +1 (CG Semi), capital_availability 0 (IT-services repricing — predictive), enterprise_adoption_depth 0 (access enforcement — predictive).

---

SEMICONDUCTOR · COMPUTE · POLICY · July 4, 2026

Modi inaugurates CG Semi's Sanand OSAT; third assembly-and-test plant online this year

Prime Minister Narendra Modi inaugurated CG Semi's outsourced semiconductor assembly and test (OSAT) facility in Sanand, Gujarat, on July 4, marking the start of commercial production. CG Semi is the joint venture of CG Power and Industrial Solutions with Japan's Renesas and Thailand's Stars Microelectronics, and the plant — built at an investment of over ₹7,500 crore — is one of the first four projects approved under the India Semiconductor Mission. Initial capacity is reported at about 20 crore chips a year, with government and company projections of up to 5 billion units annually at full build-out; a second unit, G2, is slated to come online by end-2026. The output serves automotive, industrial, consumer electronics, telecom, and power applications.

From the room. "Our youth will power the AI, robotics and next-generation technology revolution with Made-in-India chips" — PM Modi at the inauguration, per DD India.

What this means. This is the third assembly-and-test plant to enter commercial production in India this year, after Micron's ATMP facility in February and Kaynes Semicon's OSAT in March. The cadence — approvals in 2023–24, construction through 2025, output in 2026 — is the substantive signal: the India Semiconductor Mission is converting sanctioned projects into operating plants on roughly the timelines it promised, which was the part of the programme most observers doubted.

The ceiling matters as much as the milestone. OSAT is packaging and test, not fabrication, and the product classes here are automotive- and industrial-grade chips, not AI accelerators. The podium framing — made-in-India chips powering an AI and robotics revolution — describes an aspiration several tiers above what this plant ships. The chronicle records the step at its actual magnitude: real capacity, in the segment where Indian entry was most tractable.

India angle. A domestic OSAT shortens a supply-chain segment that currently routes through Southeast Asia, and the chip classes it packages are what Indian electronics manufacturing actually consumes at volume. For AI compute specifically, little changes today — accelerator-class advanced packaging is a different tier of the same discipline — but supplier base, talent pool, and logistics built at this tier are the substrate any later, higher tier would need.

Behind the news. The Gujarat semiconductor concentration has been building all year: the May 7 digest covered the Union Cabinet approving over ₹3,900 crore for two further semiconductor units in the state. Sanand's commercial output adds the production end to what had been an approvals pipeline.

What to watch. The G2 unit: CG Semi has put an end-2026 date on its second Sanand facility. Whether it holds that date is the next hard test of the mission's execution cadence.

Source: Akashvani News (newsonair.gov.in), July 4, 2026 → link; DD India, July 2026 → link.

Confidence: High on the inauguration, date, and JV structure — government outlets and multiple business dailies concur. Medium on capacity: initial-capacity and full-ramp figures vary across coverage (about 20 crore units a year initially; up to 5 billion projected at full build-out).

---

SERVICES · STRATEGY · ENTERPRISE · July 4, 2026

Top-five Indian IT's market value halves from its 2024 peak — and is now barely above Reliance alone

The combined market capitalisation of India's five largest listed IT-services companies — TCS, Infosys, HCL Technologies, Wipro, and Tech Mahindra — has fallen to ₹18.15 lakh crore, down more than 46% from the August 2024 record of ₹33.71 lakh crore, per a Business Today tally published July 4. Reliance Industries, the most valued Indian company, stands at ₹17.65 lakh crore — meaning the five firms that anchored India's equity story for two decades are together now worth roughly one conglomerate.

What this means. This is a two-year derating, not a bad quarter. The market is repricing the effort-based services model itself: clients expect AI-driven productivity to deflate contract values, and investors cannot yet see AI revenue lines large enough to offset that deflation.

The mechanism got more concrete the same week. Microsoft announced its Frontier Company on July 2 — a $2.5 billion operating unit that embeds roughly 6,000 engineers and industry specialists directly inside enterprise clients to build and run AI systems, with Rodrigo Kede Lima as president. Amazon committed $1 billion to a similar forward-deployed-engineering push days earlier. Coverage of both cites MIT Project NANDA's finding that some 95% of enterprise generative-AI pilots show no measurable profit-and-loss impact — the implementation gap is the prize, and it is precisely the gap Indian SIs have historically monetised. When model vendors and hyperscalers sell the implementation themselves, the SI position compresses from both sides: deflation on legacy effort pricing, and new competition for the AI-deployment work that was meant to replace it.

The counter-read has real weight. Market capitalisation is anticipation, not revenue; the majors' order books have not collapsed, and 6,000 Microsoft engineers is a rounding error against TCS's headcount alone. The forward-deployed model is a template threat today, not a volume threat. What the derating prices is transition risk — whether the majors can convert delivery scale into AI-era commercial models before the template scales.

India angle. IT services remains India's largest single exposure to the AI transition — by employment, by export earnings, and by equity-market weight. For the majors, the pressure lands on pricing-model conversion: outcome-linked constructs have to move from announcement to disclosed deal share. For Indian investors, the sector that was the public-market AI proxy has halved in index weight. For the talent market, the question is whether AI-skills redeployment inside the majors outpaces the hiring slowdown the derating implies.

Behind the news. The May 18 digest covered Bloomberg's argument that India's market-darling run is ending as the AI trade reshapes flows — including the Nifty IT index weight halving from over 17% in early 2022 to roughly 8%. The June 11 digest covered LTIMindtree's BlueVerse Currency, the cohort's most explicit attempt so far to reprice services for the agentic era. The July 4 tally is those arcs showing up in the aggregate number.

What to watch. TCS opens the Q1 FY27 earnings season on July 9 — the first hard revenue read of the fiscal year on whether AI-led pricing pressure is showing up in reported numbers, and on what the majors disclose about AI-attributable revenue.

Source: Business Today, July 4, 2026. → link

Confidence: Medium — the market-cap aggregates are Business Today's computation from exchange data as of early July; the direction and scale are corroborated by independent coverage of the Nifty IT drawdown, but the specific totals rest on one outlet's arithmetic.

---

POLICY · STRATEGY · ENTERPRISE · July 3, 2026

Anthropic moves to close the offshore routes Chinese firms use to reach Claude

Anthropic is moving to shut down the workarounds by which Chinese companies access Claude despite terms of service that bar Chinese firms and entities under their control, per Financial Times reporting syndicated across secondary outlets on July 3. The reported routes: Ant Group giving staff corporate Claude accounts tied to a Singapore-based subsidiary, ByteDance reimbursing engineers for personal subscriptions bought over VPNs, and foreign-incorporated subsidiaries running access through cloud infrastructure such as Azure. None of this violates US or Chinese law; it violates Anthropic's terms. Enforcement, per the reporting, runs on account telemetry — computer time zones and usage patterns that flag accounts acting as transfer stations — and, since April, identity verification for flagged accounts, including government-issued ID and live selfies before access is restored.

What this means. A private company is operationalising a geopolitical boundary with account-level telemetry. The June export-control episode was the blunt version of this: a government directive with no way to verify nationality in real time produced a hard global shutoff. Three days after that order lifted, the same company is reported to be building the granular enforcement the blunt instrument lacked — jurisdiction inferred from time zones, usage patterns, and identity documents rather than declared at signup. The direction of travel is legible: frontier-model access is becoming compliance-graded, verified, and revocable at the account level, not just the country level.

India angle. Nothing in the reporting touches Indian accounts, and the terms-of-service bar is China-specific. But the mechanics are jurisdiction-agnostic once built, and the pattern being audited — corporate access routed through Singapore-registered entities — is also how a meaningful share of Indian enterprises and GCC parents structure software procurement. The operational read for Indian procurement teams: assume hub-routed Claude contracts can be reclassified by telemetry, and that verification demands can arrive without notice. For BFSI and government buyers who started pricing availability risk into frontier-API procurement after June, this adds an identity-and-jurisdiction compliance layer to the same checklist.

What this is not. Not an India-access story. There is no reported effect on Indian users or entities, and no indication the enforcement targets anything beyond China-linked routing.

Behind the news. The arc runs through the export-control episode: the June 14 digest covered US export controls forcing Anthropic to disable Fable 5 and Mythos 5 worldwide, and the July 2 digest covered the restoration after the order lifted. Access came back; the enforcement layer being reported now is what revocability looks like as ongoing infrastructure rather than emergency action.

What to watch. Whether Anthropic converts reported practice into published policy — a named update to its usage terms or a stated enforcement framework for entity ownership and hub-country access, in the coming weeks. A published standard would be the signal that other labs face pressure to match it.

Source: Financial Times reporting, July 3, 2026, via BanklessTimes → link and Analytics Insight → link.

Confidence: Medium — the underlying reporting is the Financial Times', corroborated across multiple secondary outlets; Anthropic has not itself published the enforcement details.

---

Position movements