TechWithAdit (TWA)

Today is a day of industry reality checks delivered at the highest level. In an internal town hall, Mark Zuckerberg told Meta staff that AI agent development has not "accelerated in the way we expected" — despite $145 billion in infrastructure spending and a restructuring that cut 10% of the workforce. On the same day, Microsoft moved in the opposite direction, deploying a $2.5 billion Frontier Company with 6,000 engineers embedded directly inside enterprise clients to make AI actually work at scale.

Back home, HCL Technologies signed the biggest AI deal of the new quarter — a $1.14 billion, five-year contract with a European Fortune Global 50 company that made HCL the top NIFTY 50 gainer today. And Anthropic is entering the custom chip race: early-stage talks with Samsung to build its own AI processor, one week after OpenAI unveiled Jalapeño.

HCL Technologies announced this morning it has signed a $1.14 billion, five-year deal with a European Fortune Global 50 company to establish an AI-driven operating model for the client's global digital workplace and enterprise networks. The contract runs July 2026 to December 2031, with an option to extend a further five years — generating approximately $230 million annually for HCLTech. The deal is entirely net-new business. HCLTech shares jumped as much as 5.68% to ₹1,138.75 in morning trade, making it the top NIFTY 50 gainer today.

The announcement also came alongside HCLTech completing its acquisition of Jaspersoft from Cloud Software Group — a business intelligence and embedded analytics platform — and its continued backing of Sarvam, the Indian sovereign AI startup. Three moves in one morning, all pointing in the same direction.

The insights: Mega AI transformation deals of this scale have become rare as enterprise clients tighten discretionary spending. HCLTech landing one worth $1.14B on the first working day of Q3 is a strong signal that the Indian IT services sector still has pricing power in high-complexity AI engagements — even as the commoditisation threat from smaller AI-native players grows.

Microsoft announced Microsoft Frontier Company — a new operating unit embedding 6,000 engineers, trainers, and deployment specialists directly inside enterprise client organisations to build, deploy, and run AI systems on-site. The unit is backed by a $2.5 billion investment and led by Rodrigo Kede Lima, formerly president of Microsoft Asia. The team includes 2,000 solution architects, 1,800 deployment engineers, 1,200 trainers, and 1,000 strategists. It is model-agnostic — it can deploy OpenAI, Anthropic, or open-source models alongside Microsoft's own stack. Microsoft says client data will not train its AI models. It is already working with Unilever, Novo Nordisk, and the London Stock Exchange Group.

The move arrives two days after Amazon committed $1 billion to a similar forward-deployed engineering initiative, and follows comparable ventures launched by OpenAI and Anthropic in May. The approach was pioneered by Palantir two decades ago and has now become the dominant enterprise AI playbook of 2026. Microsoft's urgency is financial: its stock is down 21% this year, Microsoft 365 Copilot has failed to achieve widespread adoption, and GitHub Copilot has lost market share.

The insights: The $2.5 billion bet tells you what Microsoft now believes: selling AI software is not enough. The deployment gap — between what AI can do in demos and what it actually delivers in production — is wide enough that closing it requires human beings sitting inside client organisations. For any operator deploying AI at enterprise scale, the question is no longer which model to use. It's who is going to make it actually work.

The Information reported Thursday that Anthropic has begun early-stage talks with Samsung Electronics to manufacture its first custom AI chip, leveraging Samsung's 2-nanometer foundry and HBM memory production capabilities. The project is nascent — Anthropic has not yet finalised the chip's target workload, performance specs, or server configuration. The company could still abandon the effort. Anthropic recently hired Clive Chan, an engineer previously on OpenAI's custom chip program, signalling the project has moved past pure exploration. When reached, Anthropic told TechCrunch its existing stack — AWS Trainium, Google TPUs, and Nvidia GPUs — "will continue to be pivotal." Samsung declined to comment.

The move follows OpenAI unveiling Jalapeño — its Broadcom-built custom inference processor — just last week. With Anthropic's annualised revenue surpassing $30 billion (tripled from $9B at end-2025), the economics of owning its own silicon are becoming impossible to ignore.

The insights: Every major AI lab is now trying to reduce Nvidia dependence. OpenAI has Jalapeño. Google has TPUs. Amazon has Trainium. Anthropic is the last frontier AI lab without its own chip strategy — and it is clearly fixing that. For Samsung, landing Anthropic as a foundry client would be the strongest proof yet that its 2nm process can challenge TSMC at the frontier.

A new Economic Times investigation finds that the energy demands of the AI boom are forcing Google and Amazon to weaken their climate commitments, as data centre power consumption accelerates beyond what their renewable energy programmes can absorb. Google's 2024 sustainability report showed its emissions rose 48% over five years, driven by AI infrastructure. Amazon has extended its net-zero target timelines. The AI data centre boom — fuelled by the race to build compute capacity — is consuming electricity at a pace that outstrips both the build-out of renewable supply and the companies' own stated carbon reduction roadmaps.

Both companies had made public, aggressive net-zero commitments. Both are now quietly acknowledging that the AI infrastructure race makes those timelines unrealistic in the near term.

The insights: AI governance conversations have so far focused almost entirely on safety and regulation. The environmental cost — real, measurable, and growing — is the accountability gap that no major AI lab has fully answered for. As the UN's scientific panel flagged yesterday, this is precisely the kind of systemic risk that safeguards have not kept pace with.

Operator's Spotlight Read

At an internal Meta town hall on Thursday, CEO Mark Zuckerberg told staff that the "trajectory of agentic development over at least the last four months hasn't really accelerated in the way that we expected," according to a recording reviewed by Reuters. He added that the company's bets on its new AI-focused structure "haven't come to fruition yet" and acknowledged the recent restructuring was not as "clean" as it should have been. Earlier this year, Meta laid off approximately 8,000 employees — 10% of its corporate workforce — and reassigned a further 7,000 to AI-focused teams, including a unit called Agent Transformation. The move was driven by fear that Meta "wasn't going to move fast enough to adapt," Zuckerberg said. He also noted that executives entering 2026 had been "super optimistic" about tools like Anthropic's Claude Code — an optimism that has not yet translated into agent productivity at the scale the company needed.

What makes this remarkable is the specificity. Zuckerberg is not vaguely managing expectations. He is telling staff that four months of hard effort produced less progress than projected, despite a company-wide restructuring, a headcount reduction, and a capital expenditure plan of up to $145 billion — one of the largest AI infrastructure commitments in history. Meta's CTO Andrew Bosworth addressed the same town hall, confirming that a review of the company's controversial mouse-tracking employee surveillance program found no employee data had been included in AI training. Meta has now shifted the program to opt-in.

Zuckerberg did express optimism, saying he expects "more significant benefits" from AI investments within the next three to six months. But that language — carefully hedged, time-bound — is the language of a CEO who has already learned that AI deployment timelines are not what they appear on paper.

The insights: Meta is spending more on AI than almost any company on earth. If Zuckerberg is standing in front of his entire staff saying agents haven't progressed as expected, every operator who has made similar assumptions about their own AI agent roadmaps should be asking the same question. The gap between what agents can do in controlled environments and what they deliver in messy, real-world workflows is wider and harder to close than the demos suggested. That is not a Meta problem. It is an industry-wide problem — and the CEO of Meta just said so out loud.

Was this email forwarded to you? Don't miss any updates — Subscribe to TechWithAdit for sharp, no-noise tech intelligence. Stay sharp. — Adit