Samsung, SK Hynix, TSMC and Micron Top U.S. Free Cash Flow

By Sophie Lindqvist

2027 hyperscaler capital-spending estimates have reached $1.15 trillion, 94% above initial estimates, according to Sonu Varghese.

Where is the spending going?

Current hyperscaler free cash flow is negative $40 billion, he said. He expects it to remain negative through 2027. The comments came during episode 212 of the “Facts Versus Feelings” podcast.

His wider forecast puts six hyperscalers at $1.4 trillion of 2027 spending, almost 4% of GDP. He described that transfer bluntly: “what’s happening is the buyer’s cash flow is going straight to the sellers.”

That spending supports new data-center construction and, by his account, has driven an unprecedented capital-expenditure cycle across the technology companies buying advanced artificial-intelligence hardware systems.

The estimates are forward-looking, not spending already disclosed for the year. The source does not identify the six hyperscalers by name or specify the GDP base.

What does the cash-flow ranking actually show?

On the podcast, Chief Macro Strategist Varghese grouped Samsung, SK Hynix and Micron as memory makers, alongside Taiwan Semiconductor Manufacturing Company (TSMC). He said: “each generated more free cash flow in the third quarter than any other listed U.S. company, including Nvidia.”

The claim rests on third-quarter financial data, but the published item provides no company totals. It gives no year, calculation method, or comparison table for the ranking. Readers cannot independently test the result from the material presented.

The ranking supports his capital-flow thesis, not a measured profit ranking. It also does not disclose each supplier’s margin, investment burden, or cash conversion. Those omissions limit any comparison with Nvidia.

How does Nvidia fit into the chain?

Varghese’s supply-chain explanation starts with technology companies purchasing processing units. He said those units include Micron memory, with Samsung and SK Hynix also represented in the mix, before Nvidia applies its own margin.

This arrangement lets one infrastructure budget touch memory suppliers and the processor vendor at the same time. The analysis therefore presents Nvidia as one beneficiary among several, rather than the sole recipient of AI-related spending.

The published analysis supplies no product families, process nodes, package configurations, wafer sizes, shipment volumes, or memory content per system for any named company involved. The explanation cannot establish each supplier’s dollar share.

Nvidia’s margin is mentioned only as the last step in that sequence. The analysis does not quantify it or show how much cash Nvidia retains.

Are the memory suppliers reinvesting their cash?

The report says memory makers are returning cash to infrastructure even as their free cash flow strengthens. Varghese said their capital expenditure has doubled over the past year.

He did not provide a base amount, annual budget, or project list. Nor did he separate replacement spending from capacity expansion.

The doubling establishes faster investment, not the amount of new capacity, expected pricing, or the share of cash retained after the next heavy spending cycle.

It also leaves open how much of the cash-flow gain will remain after the next spending cycle.

His broader account says data-center buyers now absorb the cost, while chip vendors collect the cash. That flow is the commercial mechanism behind the reported ranking, according to Varghese. Neither the ranking nor the spending forecast proves that pricing power will remain intact.

What could happen by the end of 2027?

Varghese projected aggregate free cash flow for Nvidia Corp., Micron Technology Inc., Broadcom Inc. and Applied Materials Inc. will climb to $150 billion-$200 billion by the end of 2027. This is a forecast, not a reported result.

The forecast follows the source’s infrastructure-cycle argument; it does not provide a separate pricing target. No pricing path, unit-demand path, or capacity-utilization assumption is provided. The list of four companies also excludes Samsung, SK Hynix, and TSMC, despite their leading roles in the earlier third-quarter cash-flow claim as Varghese described.

That omission makes the $150 billion-$200 billion projection a narrower basket than the four-company ranking. The source does not explain whether the difference reflects business mix, reporting scope, or a different forecast methodology.

The source asks how the related stocks performed but supplies no share-price returns, valuation multiples, or earnings estimates. Cash-flow leadership alone does not establish which share will outperform.

Benzinga disclosed that AI tools assisted part of the item’s production and that its editors reviewed the result. That review does not independently verify the podcast’s underlying figures.

If the projected $1.4 trillion of 2027 hyperscaler spending materializes, memory suppliers must convert strong cash generation into infrastructure without weakening returns, and their next capital decisions will test whether current cash-flow gains persist.