Credit Markets Flash Caution as DeepSeek Undermines AI Capex Narrative
The Magnificent Seven stocks—Nvidia (NVDA), Apple (AAPL), Alphabet (GOOG), Microsoft (MSFT), Amazon (AMZN), Meta Platforms (META), and Tesla (TSLA)—have spent years convincing investors that their tremendous artificial intelligence spending would translate into sustainable growth, powerful cash flows, and fatter valuations. That assumption is now up against a major test, according to Bank of America strategist Michael Hartnett.
Wall Street has largely treated hyperscaler spending as a powerful long-term growth engine. However, Hartnett just flagged a major risk that could test how much investors are willing to pay for the AI trade. The contrast is becoming incredibly tough to ignore. Stock markets remain somewhat resilient, but parts of the credit market are flashing more caution around AI spending. BofA now sees one major market signal as critical to the Mag 7’s ability to shrug off that threat: the price strength of the Roundhill Magnificent Seven ETF (MAGS), holding around $70, which serves as a confidence gauge.
Why Cheap Chinese Compute Changes Everything
Chinese AI developers are showing they can deliver highly capable models using cheaper hardware, more efficient architectures, and dramatically lower inference costs. DeepSeek first exposed that flaw in early 2025. CNBC reported that its V3 model was developed using less-advanced Nvidia H800 chips, citing training costs of under $6 million. The reaction was immediate: Nvidia dropped nearly 17% on Jan. 27, 2025, wiping $593 billion from its market value in a single session.
That threat has only gotten more tangible. DeepSeek’s new V4-Flash costs just $0.14 per million input tokens and $0.28 per million output tokens, according to Artificial Analysis data reported by Reuters. Even though it was remarkably cheap, the model was much more competitive than more expensive systems, including Alibaba, Z.ai, Moonshot, and ByteDance. If businesses can achieve similar AI performance with far fewer GPUs or cheaper models, the economic return on those billions of dollars becomes much less certain.
Credit Spreads Widen as Lease Burden Mounts
Big Tech’s AI buildout is now up against a major cash-flow problem. Microsoft, Meta, Oracle, Amazon, and Alphabet have collectively written down nearly $1.09 trillion in future lease payments, much of it tied to data centers, according to Reuters. Moreover, credit markets are noticeably more cautious, with Oracle’s five-year credit-default swaps trading around 200 basis points, compared with nearly 53 basis points for a broader investment-grade CDS index.
Hartnett pointed to two major signs: rising U.S. investment-grade tech credit spreads and Oracle’s five-year CDS as evidence that credit investors are growing increasingly cautious about the AI infrastructure trade. Though stocks are still rewarding the AI story, credit markets are beginning to question its cost. For investors, the biggest risk might therefore be valuation compression instead of an immediate earnings collapse.
Counterargument: Jevons Paradox Could Rescue Demand
There’s the counterargument, though: cheaper AI could raise compute demand. Yahoo Finance reports that Microsoft CEO Satya Nadella made that argument following the original DeepSeek shock, arguing that greater AI efficiency will drive significantly more demand. So far, U.S. hyperscalers haven’t responded by slashing spending. Case in point: Amazon recently raised its 2026 capex forecast to $220 billion, citing healthy AWS demand and ongoing capacity constraints, according to the Financial Times.
MAGS ETF: High Risk, High Concentration
The Roundhill Magnificent Seven ETF has slowed significantly compared to its lofty year-over-year gains. The ETF has gained 4.82% over the past week and 4.38% over one month, outperforming the S&P 500’s 3.57% and 3.38%, respectively. However, things look shakier over a six-month period, with the ETF posting a 9.55% gain compared to the broader market’s 12%. Longer-term, Roundhill has been a massive money spinner, surging 121% over three years compared to 71.7% for the S&P 500.
It’s important to note that MAGS carries a higher-than-average risk profile, with its holdings heavily concentrated in a small group of stocks. Nearly 96% of its assets sit in its top 10 holdings, double the typical ETF level of 45%. Its annualized volatility of 22.2% is well above the ETF median of 14.1%, underscoring larger price swings. Moreover, its standard deviation is elevated at 24, compared to the ETF median of 13.
FAQ
What are the Magnificent Seven stocks and their current market caps?
- Nvidia (NVDA) — $5.424 trillion
- Apple (AAPL) — $4.572 trillion
- Alphabet (GOOG) — $4.322 trillion
- Microsoft (MSFT) — $3.712 trillion
- Amazon (AMZN) — $2.960 trillion
- Meta Platforms (META) — $1.508 trillion
- Tesla (TSLA) — $1.297 trillion
Source: CompaniesMarketCap, as of Aug. 7, 2026.
Why is the MAGS ETF considered a confidence gauge for the AI trade?
The Roundhill Magnificent Seven ETF (MAGS) holds concentrated positions in all seven mega-cap tech names. Because it trades around $70 and holds ~$70 in assets per share, its price action directly reflects investor sentiment toward the entire AI capex narrative. If MAGS maintains pricing strength despite cheap Chinese compute fears, it signals continued belief in the long-term CapEx story.
What is the Jevons Paradox and how does it apply to AI?
The Jevons Paradox states that increased efficiency in resource use leads to higher total consumption, not lower. Applied to AI: if models become cheaper to run (like DeepSeek’s V4-Flash), total compute demand may surge as more applications become economically viable. This is the bullish counterargument championed by Satya Nadella—that efficiency gains expand the addressable market rather than shrink spending.
