Citi Sees Memory Chip Shortages to 2031; Nvidia Eyes Double Sales

Two separate research signals published this week point to the same conclusion: the AI hardware buildout has years of runway left, and supply is already falling behind.

Key points

• Citi forecasts memory chip shortages deepening through 2031, driven by AI adoption of continual learning and surging demand for HBM, DRAM, and NAND.
• Nvidia CEO Jensen Huang pledged to double chip sales next year, a vow that directly lifted expectations for HBM suppliers Samsung Electronics and SK Hynix.
• Samsung has begun trial production at its Taylor, Texas facility to supply AI chips to Tesla, marking a milestone for US-based advanced semiconductor manufacturing.
• Nvidia-backed cloud firm Nscale filed for a US IPO after reporting a 1,252% revenue surge to $140.6 million in H1 2026, alongside a $1.02 billion net loss.
• A Yahoo Finance analysis warns that the anticipated Anthropic IPO may disappoint public investors, drawing a parallel to SpaceX’s pre-IPO private market run-up.

What Citi’s 2031 shortage call means for investors

Citi’s research note, published Monday, frames continual learning as a central theme in AI over the next five years. Unlike static training runs, continual learning keeps AI models updating on new data persistently, which places far greater and more sustained demand on memory bandwidth and capacity.

The practical implication is that HBM demand does not peak when a model is trained. It keeps growing as the model operates. If Citi’s thesis holds, the beneficiaries include SK Hynix, Samsung, and Micron, all of which are major HBM suppliers. Investors who have been watching Micron closely (covered here last week) now have additional analyst cover for the multi-year demand thesis.

The caveat: Citi is an investment bank with its own positioning, and multi-year semiconductor forecasts carry wide error bars. The 2031 timeline is a directional call, not a precise prediction.

Jensen Huang’s sales pledge and what it does to the supply chain

Huang’s commitment to double chip sales next year is a significant data point for the entire AI semiconductor supply chain. Nvidia’s GPUs ship with HBM stacked directly on the package, so a doubling of GPU volumes means a near-proportional increase in HBM demand. That is the direct link to Samsung and SK Hynix expectations moving on this news.

Samsung’s parallel progress at Taylor, Texas adds another layer. Trial production has started there to supply AI chips to Tesla, with full-scale production originally scheduled later. The facility’s ramp-up matters beyond the Tesla contract: it signals that Samsung is building US-based capacity at a moment when geopolitical pressure on semiconductor supply chains remains high.

The Nscale IPO filing: revenue surge, massive losses

Nscale’s US IPO filing is the kind of number that grabs attention: 1,252% revenue growth to $140.6 million in the first half of 2026. The company, backed by Nvidia, provides AI cloud infrastructure. The flip side is a $1.02 billion net loss over the same period, a ratio that reflects heavy capital expenditure on GPU clusters and data center buildout.

The filing is a useful window into the economics of AI cloud infrastructure at this stage of the cycle. Revenue is scaling fast, but so are costs. Investors evaluating Nscale (and comparable plays) will need to form a view on whether the unit economics improve at scale, or whether this is a business that requires continuous capital raises to stay competitive.

Should IPO investors be cautious on Anthropic?

A separate analysis published Friday draws a pointed comparison between the Anthropic IPO outlook and SpaceX’s private market history. The core argument: blockbuster IPOs rarely live up to the hype, and companies that generate enormous excitement in private markets often see that enthusiasm priced in before the public ever gets a chance to buy. SpaceX is offered as the archetype of a company where early private investors captured most of the value.

For Anthropic specifically, the lesson is about valuation entry points. The company has attracted massive funding rounds and high private valuations. Public market investors who buy at IPO are not buying at seed prices. That is a straightforward point, but one worth keeping in front of mind given how much enthusiasm surrounds the company. Anthropic’s CEO has also recently called for slower pacing of AI development, a posture that may complicate the growth narrative investors typically expect at IPO.

None of this is a prediction about Anthropic’s IPO timing, pricing, or outcome. It is a reminder that the gap between a company’s quality and its public market return depends heavily on the price paid.

Nothing in this article constitutes investment advice. All figures sourced from the linked reports.