Wall Street is sizing up two very different things this week. One is a fresh estimate of how much new stock will hit the US market in 2026. The other is whether Nvidia still owns the silicon layer of artificial intelligence. Together the two stories describe what the market actually fears and what it still wants to own.

A $600 billion supply year is not the wall it sounds like

Goldman Sachs put the 2026 US equity issuance figure at roughly 600 billion dollars. About 160 billion of that is expected to come from IPOs. The rest is secondary offerings and convertible debt. The bank’s view is that supply at this level is manageable.

The reasoning is plain. Markets in 2024 and 2025 absorbed a heavy issuance calendar without choking. Passive inflows into US equity ETFs kept up. Corporate buybacks continued to sit on the other side of the trade.

Supply is only ever a problem when demand drops. In a market with steady index buying and a stable retirement bid, an extra 600 billion of paper is digestible. The headline looks large in isolation. Place it next to the trillions in retirement and index demand and it shrinks fast.

The mix of IPOs matters more than the total. A list dominated by profitable software and infrastructure names clears at fair multiples. A list stuffed with cash burning startups needs a willing risk tourist. The 2024 and 2025 cohorts skewed toward the first category. That is a clue, not a guarantee.

Nvidia and the ASIC question

The same week brings a louder debate around Nvidia. The bear thesis is simple. Hyperscalers like Google, Amazon, Meta, and Microsoft are pouring money into custom application specific chips designed for their own workloads. If even part of that effort lands, Nvidia loses a chunk of high margin demand.

The bull side answers in two parts. First, the customer base is wider than the four hyperscalers. So called neocloud providers and large enterprises still need general purpose accelerators. Most of them are locked into the CUDA software stack. Switching to a custom chip is not a hardware purchase. It is a multi year engineering migration with real risk.

Second, the inference market is opening up. New silicon aimed at low latency token serving expands the total addressable market rather than splitting it cleanly with training silicon. Inference workloads are growing faster than training workloads. That is where most of the next decade of compute dollars will go.

The numbers cited in the bull case sit at a price to earnings ratio near 25 times forward estimates and a projected 70 to 75 percent share of the AI accelerator market through 2030. Free cash flow keeps funding aggressive buybacks. For a company sitting at the center of a once in a decade buildout, that multiple does not look stretched.

Treasury yields are the quiet pressure under everything

Behind both stories is the same macro factor. Treasury yields have moved higher. That changes the math everywhere. Rate sensitive sectors take the hit first. Real estate trusts and utilities deflate. Banks see deposit costs climb back. Forward earnings estimates for the broad S&P 500 are still grinding higher, but not as fast as yields are biting into valuation multiples.

Higher yields also tighten the bar for new equity issuance. Investors will buy paper at sensible multiples. They will not pay growth premiums to companies that cannot grow into them. That is exactly the discipline that makes a 600 billion dollar supply year manageable.

The interaction nobody talks about

Here is the link between the two stories. A heavy IPO calendar plus a single dominant chip vendor creates a strange feedback loop. Many of the bigger 2026 listings will be AI dependent. They will pitch growth that depends on access to scarce compute. Compute that mostly comes from Nvidia.

If Nvidia’s grip stays at 70 percent or higher, those new issuers have a clear capex story but a concentrated supplier risk. If competition takes a bigger bite than expected, supply chain stress eases for the new issuers but Nvidia’s premium multiple comes under pressure. Investors cannot pick the optimistic scenario for both at the same time.

What this means

For ordinary investors the practical takeaways are short. Heavy issuance calendars rarely cause crashes on their own. They cap upside in the frothiest pockets and push capital toward profitable names. Nvidia’s position is dominant but not permanent. Concentration risk inside the S&P 500 is real. Anyone buying broad index funds is making a sizable bet on three or four semiconductor and software names whether they realize it or not.

What to watch

Three numbers over the next two quarters.

• The mix of IPOs that price in the second half of 2026. Profitable issuers clearing easily would confirm the manageable supply thesis. Aggressive postponements would not.
• Hyperscaler capital expenditure guidance. If the four big buyers keep raising spend, the AI accelerator demand picture stays intact regardless of who supplies the chips.
• The 10 year Treasury yield. It remains the single most important number on the screen. As long as it stays anchored, equity multiples have room. If it breaks higher again, every other story has to bend around it.