Published on LinkedIn and amitabhapte.com | 17 May 2026
This week, three very different stories arrived at the same conclusion. A chipmaker went public and nearly doubled on its first day. India signed a deal to build its first semiconductor fab. And AI-powered robots quietly took over 40 Los Angeles neighbourhoods while AI rewired the world’s largest e-commerce search bar. The infrastructure of intelligence is no longer being planned. It is being built, listed, traded, and deployed on your nearest pavement.
1. The Capital Signal: Chips, Memory and Public Markets
The Cerebras Systems IPO was the largest US tech listing since Uber in 2019. Priced at $185 per share, it raised $5.55 billion and surged 68% on its first day of trading, valuing the company at nearly $50 billion. The company makes wafer-scale AI chips. Its WSE-3 processor is the size of a dinner plate, packing over 4 trillion transistors, and is purpose-built for inference, running trained AI models fast and cheaply rather than training them. With AWS and OpenAI already as anchor customers, the order book closed 20 times oversubscribed.
The headline, though, is not just the company. It is the queue behind it. SpaceX, OpenAI, and Anthropic are all reportedly preparing listings, with SpaceX and OpenAI alone expected to raise a combined $135 billion. The public markets are signalling something clear: AI infrastructure is now a provable, priceable asset class, not a speculative bet.
And yet, as capital celebrates the chip layer, a different bottleneck is quietly tightening. High-bandwidth memory (HBM) has become AI’s most critical constraint. Data centres now consume an estimated 70% of global high-end memory production. Samsung, SK Hynix, and Micron have reallocated the vast majority of their fab capacity to HBM for AI accelerators, producing 3x less conventional DRAM per wafer as a result. DRAM prices have surged sharply in early 2026, with memory sold out across the supply chain into 2027. The irony is structural: we are celebrating the intelligence layer while quietly starving the memory layer that makes it run.
| My PoV The Cerebras IPO is important, but memory is the signal that matters more. When three manufacturers control 95% of global DRAM production and have committed most of it to AI data centres through 2027, every enterprise architecture conversation should include a memory procurement question. If your AI roadmap does not account for infrastructure scarcity, it is optimistic by design. |
2. Geography as Strategy: Capital, Compute and Sovereignty
The most telling signal this week was not a product launch. It was a portfolio filing. A Reuters analysis of nearly 6,000 institutional investor 13-F filings showed that more than 4,000 funds added to or initiated positions in AI infrastructure stocks in Q1 2026. Only 146 sold. Not only that: there were no net sellers of utility stocks in the quarter, as pension funds and endowments moved money toward the power and cooling layer that keeps AI running. The Magnificent Seven are being treated selectively. The infrastructure beneath them is being bought without hesitation.
This capital movement is not random. It is following a geography. AWS committed €33 billion to Spain, framing the country as its European AI epicentre. Tata Electronics and ASML signed an MoU to build India’s first commercial 300mm semiconductor fab in Dholera, Gujarat, backed by an $11 billion investment. The deal, signed in the presence of both nations’ heads of government, covers ASML’s full lithography suite, talent development, and supply chain resilience, targeting chips for automotive, mobile, and AI from 28nm to 110nm. On services, TCS declaring its ambition to become the world’s largest AI-led technology services company, with 130 of its top 139 enterprise clients already on board and over 270,000 employees upskilled in AI in a single year. TCS Chairman N. Chandrasekaran describing AI as “the infrastructure of intelligence”, a deliberate echo of how the internet was framed in the late 1990s. Silicon, services, and strategy, converging at the same moment.
The broader pattern is clear. AI leadership is being contested not just in model benchmarks but in megawatts, fab nodes, data centre land, and sovereign infrastructure decisions. The race has moved from algorithms to atlases.
| My PoV When 4,000 institutions buy AI infrastructure and zero sell utilities in the same quarter, that is not a trend. It is a conviction. The smart money has concluded that the physical layer of AI, power, chips, cooling, connectivity, is where durable advantage compounds. For enterprise technology leaders, the implication is concrete: your AI strategy is increasingly shaped by decisions being made in fabs, power grids, and diplomatic meetings, not just in model releases. Understanding your infrastructure dependencies by geography is no longer optional. |
3. AI Meets Culture: From Cannes to the Pavement
Two very different scenes this week, both telling the same story about normalisation.
At Cannes, filmmakers shifted from resistance to pragmatism. French director Xavier Gens noted that AI would have cut his Netflix hit’s visual effects budget in half and saved eight months of post-production time. Demi Moore, serving as a jury member, said simply: “AI is here. To fight it is to fight something that is a battle we will lose.” The distinction that emerged from Cannes was useful and important: directors broadly agreed that using AI to generate scripts or entire films from a prompt should not be allowed, but its use in production and post-production is increasingly accepted. Guillermo del Toro put it well, noting that lumping all AI applications under one label makes productive discussion impossible. That nuance, creativity vs. craft, authorship vs. tooling, is the conversation every industry will eventually need to have. Cannes just had it in public first.
On the commerce side, Amazon replaced Rufus with Alexa for Shopping, embedding an AI assistant directly into the default search bar of the world’s largest e-commerce platform. No click required. AI answers now appear by default for complex queries, completing the shift from search as retrieval to search as conversation. The assistant can also shop third-party retailers and execute purchases autonomously through Buy for Me. Meanwhile, on the streets of Los Angeles, Serve Robotics has deployed over 500 sidewalk delivery bots across 40 neighbourhoods, up from just two in 2023. Cities are scrambling to draft regulations in real time.
The pattern across Cannes, Amazon, and the pavements of Los Angeles is the same. The debate has moved from “will AI arrive?” to “how do we live with it?” That is a meaningful shift. It means AI is no longer being evaluated. It is being negotiated.
| My PoV The Cannes conversation matters for leaders well beyond media. Every industry will face the same negotiation: which parts of creative and knowledge work can be augmented, which parts should remain human, and how do you draw that line without either losing competitive ground or losing your people? The organisations that will navigate this best are those that have the conversation explicitly, rather than letting it resolve through default or drift. |
My Takeaway This Weekend
Three things became measurably more real this week. The chip layer went public and priced at scale. The geographic layer hardened, with capital, fabs, and services converging by region rather than by company. And the consumer layer normalised, quietly and irreversibly, from the search bar to the pavement.
The window between “this is coming” and “this has already happened” keeps closing faster than planning cycles allow. The organisations best positioned are not necessarily those with the most AI projects. They are the ones that understand where their infrastructure dependencies sit geographically, which parts of their work they are willing to delegate, and which parts they are determined to protect.
Amitabh Apte 