India AI Digest — Monday, July 27, 2026

• Anthropic CEO Dario Amodei publicly denied ever advocating a ban on open-weight AI models, pushing instead for tighter chip export controls on China — a rebuttal issued as Nvidia's "Open Weights and American AI Leadership" letter grew to a fast-growing list of signatories over the weekend, with Anthropic and Amazon among the most conspicuous holdouts.
• Moonshot AI published Kimi K3's full open weights, reported at 2.8 trillion total parameters with 104 billion active — among the largest open-weight releases to date — continuing the story the White House escalated on July 22 by accusing Moonshot of distilling Anthropic's Fable.
• Ola's Krutrim cut 20-25 more jobs on July 28, roughly halving its already-shrunk workforce in its second restructuring of 2026, as the company continues its pivot from foundation models and chips to AI cloud infrastructure.
• A two-item day: both stories sit on the compute-and-open-weights fault line that's been running through the archive all week, but nothing India-specific and independently verifiable cleared the bar today outside the Krutrim item.

---

POLICY · COMPUTE · GEOPOLITICS · July 27, 2026

Amodei denies advocating an open-weight AI ban, as Nvidia's coalition letter nears 50 signatories

Anthropic CEO Dario Amodei published a statement on July 27, responding to a weekend of pressure following Nvidia CEO Jensen Huang's July 24 open letter, "Open Weights and American AI Leadership." "Anthropic has never advocated for a ban on open-weights models," Amodei wrote, adding: "Open-weights models that don't have dangerous capabilities are a public good: they don't cost anything besides the compute needed to run them." His stated concern is narrower — authoritarian governments, with the Chinese Communist Party named as the most capable one he worries about, using frontier AI for what he called "permanent military superiority" and domestic repression. In place of a ban, he called for tighter export controls on chips and chipmaking equipment to China, restrictions on model distillation, and international model-safety testing bodies that include China. Huang's letter carried Nvidia, Hugging Face, Meta, Microsoft, and Mistral among its named signatories at launch, per TechCrunch, and grew to a fast-growing list of signatories over the following days; Anthropic and Amazon remain the most visible non-signatories.

What this means. The letter and the rebuttal are both downstream of the same trigger: Kimi K3. Moonshot AI's open-weight model is the concrete proof point the open-weights coalition points to when it argues that restricting downloadable models mainly cedes ground to China rather than containing it. Amodei's rebuttal doesn't dispute that logic on its face — he explicitly calls open weights without dangerous capabilities "a public good." His disagreement is about where the line sits and who draws it, not about open weights as a category.

The distillation allegation from July 22 — the White House's claim that Moonshot built K3 by covertly distilling Fable, with Treasury floating sanctions — sits underneath this entire exchange without being resolved. Amodei's call for "restrictions on distillation" as a policy tool, rather than a ban on downloadable weights, reads as an attempt to keep that specific fight open while conceding the broader open-weights argument. Whether Washington's policy apparatus converges on distillation-specific enforcement or broader model-access restrictions is still unsettled; Amodei is trying to steer it toward the former.

India angle. India has no seat at this table and no frontier lab whose commercial interest the letter or the rebuttal directly serves — but the outcome shapes what's on the shelf for Indian builders. India's AI chip and model access has run on ad hoc, bilaterally negotiated terms since Washington rescinded the tiered AI Diffusion Rule in May 2025, with no fixed classification replacing it. A policy regime that treats open weights broadly as a public good — Amodei's own framing, even as he pushes distillation-specific restrictions — keeps cheap open-weight options like K3, DeepSeek, and Qwen available to Indian teams that can't afford frontier API pricing. A regime that criminalizes distillation more broadly, or that expands from "don't distill covertly" to "disclose your training provenance," would land on any Indian lab that has fine-tuned or evaluated against frontier-model API outputs — a category that plausibly includes parts of the Sarvam and Krutrim pipelines, neither of which discloses training methodology to a level that would clear that bar today.

Behind the news. The July 23 digest covered the initial White House-Treasury distillation accusation against Moonshot, which flagged India's compute access as structurally exposed to, but not currently named in, that dispute. Amodei's July 27 statement is the first instance of a major US lab publicly distancing itself from the ban framing that the Moonshot accusation implied was coming.

What to watch. Whether OpenAI and Google, both reported to have joined Huang's coalition after initially abstaining, stay on the letter as the distillation dispute with Moonshot develops — a split between "pro-open-weights-in-general" and "anti-Moonshot-specifically" positions among the signatories would be the next visible fracture line.

Source: TechCrunch, July 27, 2026. → link

Confidence: Medium-high. Amodei's quotes and the letter's signatory count are corroborated across multiple outlets; the exact current signatory count is a moving target and not independently verified as of this writing.

---

FOUNDATION MODELS · STARTUPS · INDIA · July 28, 2026

Krutrim cuts 20-25 more jobs, halving its remaining workforce in its second 2026 restructuring

Ola-backed Krutrim laid off 20-25 employees on July 28, roughly halving what remained of a workforce that stood above 550 people in August 2025 and had already fallen to 150-160 by March 2026. The cuts fell mainly on product and engineering roles, in some cases leaving single employees covering what had been team functions. Krutrim's statement described the move as "organisational restructuring" to "align with evolving priorities," and said the company "continues to be in an active phase of scaling its operations." The layoffs follow Krutrim's earlier 2026 pivot away from large language models and semiconductor work toward AI cloud infrastructure and enterprise services, and the discontinuation of its Kruti consumer AI assistant in April.

What this means. The gap between the stated framing and the underlying trend is now wide enough to state plainly: a company describing itself as "scaling" has cut its headcount by roughly 75% in eleven months, and this is its second restructuring round of the year. The strategic pivot from full-stack foundation-model ambitions to cloud infrastructure and enterprise services was itself a retreat from the original Krutrim pitch — an India-built LLM and chip stack to rival global labs. This layoff is a retreat from the retreat: a smaller team even for the narrower cloud-and-enterprise mandate.

Applying the substance diagnostic the archive has used on Krutrim before: shipping ratio and technical disclosure were already the weakest tiers for this company relative to Sarvam and AI4Bharat. A workforce this small — well under 150 people across all functions — makes sustained model or infrastructure R&D at any serious scale close to impossible going forward, regardless of what the company's public positioning says. The cloud-and-enterprise pivot itself now needs re-evaluating against a team of this size.

India angle. For the Indian foundation-model narrative, Krutrim's contraction sharpens a divergence the archive has tracked for over a year. Sarvam's HCLTech-and-Government-of-Odisha sovereign compute MoU, announced July 24 at a planned outlay of roughly ₹14,257 crore (~$1.4 billion, Business Standard) — a separate figure from Sarvam's own $1.5 billion valuation reported earlier this year — preceded Krutrim's layoffs by four days; Sarvam then disclosed its trillion-parameter model roadmap at its Epoch 2026 conference on July 30, two days after the cuts. Both announcements land in the same calendar week as Krutrim cutting engineering headcount to skeleton levels. Both companies started from comparable positioning — founder-led, well-capitalized, India-first framing — and are now visibly diverging on execution. For investors evaluating the next cohort of Indian AI companies, the gap between capital-and-narrative-led entry and sustained technical delivery is no longer theoretical; it has two years of concrete outcomes attached to two specific companies.

Behind the news. The July 21 digest explicitly flagged Krutrim's restructuring as a story that fell outside that day's event-date window, noting it would be covered on the day it actually happened. This is that day. Krutrim's layoffs stretch back through 2025 — roughly 50 people in July 2025, about 100 (mostly linguistics) in September 2025 — with July 28, 2026 marking Krutrim's own count of its "second round" of cuts this year, though the first 2026 round isn't independently dated in available reporting.

What to watch. Whether Krutrim discloses a specific enterprise or government customer for Krutrim Cloud at meaningful scale — the pivot's credibility now rests on cloud/enterprise revenue that hasn't been independently verified in available reporting, run by a team that keeps shrinking.

Source: Inc42, July 2026 (reporting on the July 28 layoffs). → link

Confidence: Medium. Layoff figures and company statement are corroborated across multiple outlets; historical headcount figures rely on secondary reporting not independently verified against company filings.

---

Position movements