Bargo
TSM

The Supply Side Voted With Its Wallet: ~$90B of New Chip Capex in One Week

TSMC and SK Hynix fund AI expansion from cash flow, Intel funds it by printing shares. Same race, different fuel.

Bargo · 2026-08-11

The supply side just voted with its wallet. About $88B of new chip capex was approved or priced in one week.

Three announcements landed between Aug 7 and Aug 11. TSMC and SK Hynix are funding AI capacity from cash flow. Intel is funding it by printing shares. The market priced that difference immediately.

New chip capex approved / priced Aug 7-11, 2026 ($B)

Source: Company statements via SK hynix Newsroom, TrendForce on TSMC guidance, Quartz/Reuters on Intel pricing. TSMC $29.44B per company statement cited in brief, SK Hynix ~$38.3B, Intel $20B. Total ~$87.7B.

1. TSMC: cash flow funds the build

TSMC lifted its 2026 capex guidance to $60B to $64B on July 16, up from $52B to $56B, citing structural AI and HPC demand TrendForce. Management guided Q3 revenue to $44.6B to $45.8B, up about 37% year over year, and flagged margin dilution from the 2nm ramp and overseas fabs.

The board then approved a $29.44B capital appropriation on Aug 11 for advanced capacity, per the company statement cited by Focus Taiwan in your brief. That appropriation sits inside the $60B to $64B full year budget, with 70% to 80% earmarked for advanced nodes.

TSM traded at $423.66 on Aug 11, up 0.86% on the day, with tight spread and volume over 11.8M shares by early afternoon ET.

This is the incumbent funding model. When you own the leading edge and generate cash, you can add capacity without asking shareholders for more capital.

2. SK Hynix: memory financed like infrastructure

SK Hynix approved about 54 trillion won, about $38.3B, for two new fabs on Aug 7 SK hynix Newsroom and Reuters. The split is 35.2T won for Yongin Y2, a DRAM base for HBM and next generation DRAM, and 19.1T won for Cheongju M17, a NAND base for enterprise SSD.

Timelines are long dated by design. Y2 breaks ground July 2027 with first cleanroom June 2029. M17 breaks ground February 2027 with first cleanroom December 2028. Investment periods run through 2031. The company said it will build shells on schedule and add cleanroom and equipment in phases aligned to customer demand to protect capital efficiency.

The framing that matters is from @semivision_tw. AI memory is being financed like infrastructure, with long term contracts and heavy upfront capital, not like a commodity cycle. SK Hynix cited Omdia forecasts for 19% CAGR in DRAM and NAND demand through 2030 and called memory core infrastructure for AI performance, not just a component.

SK Hynix ADR was about $138, roughly flat on the news. The memory tape stayed firm behind it. SNDK was $1,255, up 3.53%, and MU was $872.41, down 0.59% intraday but still near highs after a strong run. That firmness fits the broader shortage spreading from GPUs into memory and tools noted in The Shortage Is Spreading Sideways.

3. Intel: dilution as fuel

Intel upsized its common stock offering to $20B on Aug 11, pricing 210,526,315 shares at $95 per share Quartz. That was up from a $15B target announced Aug 10. The $95 price was a 2.6% discount to the prior close, per Reuters. Net proceeds are about $19.7B after fees, with a 30 day option for underwriters to buy another 31.6M shares.

Proceeds are for general corporate purposes including capex and working capital. Intel had already lifted its 2026 capex target to $20B from $18B in July and committed to high volume production on its 14A process in 2028. Tesla was cited as a 14A foundry customer.

The market charged for the dilution. INTC fell to $96.62, down 4.95% on the day, on volume over 101M shares by early afternoon. The stock is still up about 175% year to date in 2026, which is why equity was the cheapest fuel available. As noted in Intel's $15B Raise and the "Super Successful" Tell, the CFO's conditional language on the Q2 call already hinted at pre positioning for major foundry commitments.

4. Why the funding mix matters

Same race, different fuel tells you who has pricing power today.

For investors, the contrast is not about whether AI demand is real. All three are betting it is. It is about balance sheet quality and timing. Cash flow funded capex is a sign of a sold out market. Equity funded capex is a bet that future returns will exceed the dilution cost. Both can work, but they carry different risk if demand slows.

Memory pricing is the near term tell. If HBM and enterprise SSD stay tight, the infrastructure financing model for SK Hynix pays back faster. See NVDA Rubin Economics: Why $8.3M per Rack Still Prints 78% Margin at $53/GB HBM and Memory's Fastest Unwind of 2026 for how HBM cost flows through system margins.

5. What to watch


Sources

Get Bargo research in your inbox
One email when we publish. No spam, unsubscribe anytime.
Get Bargo research in your inbox
One email when we publish. No spam.