India AI Digest — Saturday, April 18, 2026

• xAI dropped Grok 4.3 Beta on April 17 silently into the model selector, behind a $300/month SuperGrok Heavy paywall, with native video input and document-generation tooling.
• Polaris Smart Metering closed Rs 710 crore (~$80M) of debt financing from British International Investment for a 2.2M-meter West Bengal AMI rollout, underwriting AI-on-grid telemetry rather than AI directly.
• Thin day; one global-with-India read and one infrastructure adjacency.
• Position movements: foundation_model_capability +1 (xAI, global frontier), compute_infrastructure 0 (India, AMI substrate touched not moved).

---

Grok 4.3 ships into the model selector with no announcement post

xAI made Grok 4.3 Beta available on grok.com and the iOS/Android apps on April 17, 2026, flagged as Early Access for SuperGrok Heavy subscribers at $300/month. There was no press release or blog post; Elon Musk confirmed availability on X and said xAI plans to ship updates roughly daily. The release adds native video input, in-chat generation of PDFs, PowerPoint slides, and spreadsheets, and is reported to extend long-context handling. Persistent cross-session memory is not present. Standard SuperGrok subscribers ($30/month) see the model in the dropdown but cannot access it; full rollout to that tier is targeted mid-to-late May 2026 [TBV].

What this means. The release pattern is the substantive part. xAI is moving to continuous-deployment cadence for a frontier-class model — beta drops without announcement, daily improvements, formal release notes after the fact. That's a different tempo from OpenAI's versioned releases or Anthropic's model-card-anchored launches. Whether the daily-shipping approach holds quality across the rollout is the empirical question; iterating in production has historically traded off against stability.

The capability disclosure is uneven. Native video input and document generation are concrete additions that change agent workflows. The long-context claim and the underlying architecture are not detailed in any release-equivalent post; benchmark numbers will surface in third-party evaluations over the coming weeks rather than in xAI's own materials. The price-feature pairing is what to track. $300/month for the Heavy tier is positioning Grok 4.3 against ChatGPT Pro and Claude's higher tiers; access to the same capability for the $30/month tier in mid-to-late May resets where the model sits in builder evaluation sets.

The cautious read on the daily-shipping framing: continuous deployment of a frontier model means the model a builder evaluates today may not be the model that responds to a production query next week. That's an operational property, not a feature, and it makes Grok 4.3 harder to anchor evaluations on than versioned alternatives.

India angle. Two reads cluster.

For Indian consumer use, the $300/month tier is unreachable at Indian discretionary ARPU. SuperGrok Heavy will sit in the same category as ChatGPT Pro for the Indian market — used by a small cohort of operators, founders, and high-end professional users, not a mass-consumer product. The relevant tier for Indian consumer signal is the $30/month standard tier when 4.3 lands there.

For Indian agent and SI-layer builders, video input and in-chat document generation matter. Indian SI offerings around enterprise AI agents — TCS, Infosys, Wipro, HCLTech, plus the application-layer builders shipping agentic systems — gain a candidate model where the workflow primitives now include video understanding and presentation output. Whether Grok 4.3 enters production evaluation depends on the daily-deployment property: SI engagements need stable APIs across release cycles, and a model that ships changes daily creates a different vendor-management problem than versioned alternatives.

For Indian AI-builder evaluation against Indian-built models — Sarvam, AI4Bharat-lineage work — Grok 4.3 is not a directly comparable peer (different scale, different positioning). The indirect comparison is: every quality jump at the global frontier widens the headroom that Indian Indic-optimized models trade off against, and the unit-economics question for Indic-first products gets harder when the global cost of capability keeps falling.

Source: xAI / grok.com, April 17, 2026 (no formal release post); secondary coverage of feature set and access tiers. → link

Confidence: low — release confirmed via secondary reporting and product surface; specific architecture, benchmark numbers, and license details not in any primary release post.

---

Polaris Smart Metering closes Rs 710 crore from BII for West Bengal AMI rollout

Polaris Smart Metering, an I Squared Capital portfolio company, secured Rs 710 crore (~$80M) in debt financing from British International Investment (BII) for its Hooghly Smart Metering subsidiary, per Entrackr reporting on April 16, 2026. The capital underwrites deployment of 2.2 million smart meters in West Bengal. Polaris's order book sits at roughly 7.5 million electric meters across Uttar Pradesh, West Bengal, and Manipur, plus 1.6 million smart gas meters in operation [TBV against company filings].

What this means. This is energy infrastructure financing, not an AI deal. The reason it sits in a digest tracking India's AI position is the substrate question. Advanced Metering Infrastructure rollouts at this scale are the data layer that AI-on-grid applications run on top of — load forecasting, AT&C loss detection, distribution-side anomaly detection, time-of-day pricing optimization. Without the AMI rollout, the AI-on-grid story is benchmark-heavy and deployment-thin. With it, the data substrate exists; what's built on top is a separate question.

The honest read on what this announcement does and doesn't do: it adds 2.2M meters of telemetry-capable endpoints in West Bengal. It does not name an analytics or AI partner, does not commit to data-sharing arrangements with utilities or third parties, and does not specify whether the AMI software stack supports model-driven analytics out of the box. The financing is debt, which constrains how the operator structures partnerships downstream. The dimension being touched here is compute and data infrastructure for AI applications in a specific sector; it's not being moved by this transaction alone.

The BII participation is the directional read. UK development-finance capital flowing into Indian energy infrastructure — at Rs 710 crore on a single subsidiary — is consistent with the pattern of patient foreign capital underwriting India's grid digitisation, separate from the venture-AI flows that get most of the attention. Whether this becomes part of an AI-on-grid story depends on what software stack Polaris's utility customers eventually deploy on top of the meter telemetry.

India angle. Sectoral, narrow read.

For Indian energytech AI builders — load-forecasting, demand-response, fraud-detection startups — the AMI substrate becoming available at the 7.5M-meter order-book scale is the prerequisite to actually shipping product into Indian distribution utilities. The dependency runs: AMI deployment → utility data accessibility → vendor procurement for analytics → AI-driven grid operations. The first link of that chain is what's getting funded; the later links are not addressed by this transaction.

For the IndiaAI Mission's sectoral agenda on energy and infrastructure, the rollout is consistent with the framing that AI applications in Indian energy infrastructure depend on physical and digital grid modernisation that's still mid-progress. The investment thesis — that Indian utilities will eventually procure AI-driven grid analytics on top of AMI — remains forward-looking; the substrate accumulating is the testable signal.

What this is not. Not an AI deal. Treating it as one inflates the AI-relevant signal. The signal is that the data layer that AI-on-grid applications need is being laid in the largest Indian state-utility geography.

Source: Entrackr, April 16, 2026. → link

Confidence: medium — funding amount and geography confirmed via secondary; AI implications are inferred from sector adjacency and remain forward-looking.

---

Position movements

---

Thin day. Two items shipped where 3-5 is the target. Indian-source AI events on April 18, 2026 itself were not surfaced by primary-source search; closest Indian-domiciled item was Polaris on April 16, framed honestly as adjacency rather than direct AI. Sarvam AI's $350M raise remains in advanced talks per April-mid reporting but no closing-date primary surfaced for this date window. Better to ship two well-grounded items than to pad with weakly-anchored ones.