Capital & Industrial Strategy

Cerebras Systems priced its Nasdaq IPO and surged approximately 68% on debut, reaching a market capitalisation of roughly $67 billion — the largest IPO of 2026. The offering is structurally significant not just as a liquidity event but as a direct challenge to Nvidia's dominance in AI compute. Cerebras' wafer-scale chip architecture targets inference workloads where Nvidia's GPU paradigm carries inefficiencies, and the IPO signals that public market investors are now willing to price in a credible alternative at meaningful scale. Bloomberg WSJ CNBC

SambaNova CEO Rodrigo Liang moved quickly to counter-position, arguing publicly that the next AI infrastructure war is not about training chips but inference economics — framing Cerebras' GPU-replacement narrative as only half the story. Bloomberg The post-IPO pullback on day two suggests institutional buyers are sizing positions cautiously relative to the retail-driven first-day spike, which is the normal pattern for high-profile tech debuts. The valuation requires Cerebras to demonstrate not just technical differentiation but commercial scale against an incumbent that controls the developer ecosystem.

Anthropic is raising an additional $30 billion, which would rank among the largest private funding rounds in technology history. The WSJ reports that a handful of frontier AI labs are now absorbing the majority of global venture capital deployment, with Anthropic, OpenAI, and xAI collectively pulling capital away from the broader AI ecosystem. WSJ This structural concentration is not incidental — it reflects investors' judgment that the value chain in AI will be captured at the model layer, making bets on application-layer companies increasingly risky relative to direct exposure to foundational model providers.

Concurrently, OpenAI CFO Sarah Friar confirmed the company may raise additional capital beyond its recent round, suggesting that even freshly capitalised frontier labs view their current funding as insufficient runway for the infrastructure and compute demands ahead. Bloomberg The AI boom is also reversing the multi-year trend of de-equitisation in US markets, with the FT noting that the volume of AI-related equity issuance is now large enough to shift aggregate market dynamics. FT

OpenAI's two-year partnership with Apple has become severely strained, with Bloomberg reporting that OpenAI is preparing possible legal action against Apple after failing to receive expected distribution benefits from the deal. Bloomberg Semafor The Apple relationship was central to OpenAI's consumer distribution strategy — iOS integration represented access to over a billion active devices without direct customer acquisition costs. A breakdown removes that channel and forces OpenAI to compete for consumer attention through direct app installs, a structurally inferior position against Google's native Android and search integration.

Simultaneously, OpenAI has reorganised its product leadership, with Greg Brockman now officially taking control of the ChatGPT and Codex product lines in a move to unify the consumer and developer product experience. Wired This is the latest in a series of executive reshuffles that signal ongoing internal tension over product direction. The Musk v. Altman trial concluded this week, with final arguments centering on governance and trustworthiness of OpenAI's leadership — a reputational overhang that complicates enterprise sales cycles even as the company pursues further fundraising. TechCrunch

President Trump disclosed that he discussed Nvidia H200 chip access and AI guardrails directly with Xi Jinping during their Beijing summit — an extraordinary elevation of semiconductor export policy into head-of-state diplomacy. Bloomberg Bloomberg The H200 sits at the boundary of current export control thresholds, and any presidential-level signals about relaxing access would have immediate implications for Nvidia's addressable market in China — historically a significant revenue contributor before the Biden-era restrictions. The SEMI workforce comment in the same discussion underscores that the US chip sector faces a dual constraint: geopolitical demand management externally and talent shortages internally.

The FTC's antitrust probe into Arm Holdings' chip technology licensing practices adds another regulatory dimension to the semiconductor landscape. Bloomberg Arm's architecture underlies virtually every mobile chip and is increasingly central to data centre AI inference silicon — an antitrust finding on licensing terms could structurally alter the cost base for every AI chip designer building on Arm IP, including Qualcomm, Apple, and a range of custom silicon programmes.

The Roundhill Memory ETF (DRAM) has reached $10 billion in assets at the fastest pace ever recorded for an ETF, driven by investor positioning around DRAM as the identified bottleneck in AI infrastructure scaling. CNBC Kioxia, the Toshiba flash memory spinout, is planning to list American depositary shares to access US capital markets, directly linking the AI compute buildout to non-US memory suppliers seeking dollar-denominated investment. FT Reuters also confirmed that institutional investors materially increased new stakes in semiconductor firms during Q1 2026. [Reuters] South Korean bond markets are pricing in semiconductor-driven growth as inflationary, with analysts forecasting yield extension on the basis that the chip boom is adding genuine economic heat. Bloomberg

Ackman's new Microsoft stake deserves reading alongside these memory and chip flows — Pershing Square is essentially making a capital stack argument that Microsoft is the most liquid, underpriced way to own the AI infrastructure buildout without taking single-chip-company concentration risk. WSJ Figma's strong Q1 results and raised full-year guidance, meanwhile, provided the first concrete evidence that AI disruption of the design software stack is not yet materialising as a revenue threat — a meaningful data point for the broader SaaS-versus-AI disruption debate. Bloomberg