Bank of America's top strategist has found value in an unloved corner of the AI trade: the bonds of the companies paying for it.

"Buy tech bond 'diamonds-in-rough' as AI hyperscalers up equity issuance, cut buybacks," Michael Hartnett, chief investment strategist at Bank of America, said in his latest Flow Show note.

The call comes after a punishing year for tech credit.

Why Hartnett Calls Tech Bonds 'Diamonds in the Rough'

According to Hartnett, U.S. technology bonds have lost about 9% so far in 2026.

That leaves their total return relative to the broad investment-grade bond index at the lowest level in 30 years. In other words, tech debt has rarely looked this cheap compared with the rest of the high-grade market.

The main reason is supply. AI hyperscalers have sold $182 billion of bonds this year, up from just $17 billion in 2024, as they borrow to build data centers.

Investors have demanded higher yields to absorb that wave. As a result, long-dated bonds of the Magnificent Seven and other AI leaders now yield between 6% and 8%, a level Hartnett considers attractive.

Source: Bank of America Global Research, Bloomberg

Further down the credit ladder, the 2031 high-yield bond of CoreWeave Inc. (NASDAQ:CRWV) yields 14.26%.

Why Bondholders Are Gaining the Upper Hand

The second part of Hartnett's argument concerns how the hyperscalers are now raising money.

Meta's recent share sale, he said, signals a new respect for credit investors. Selling stock rather than adding more debt strengthens the balance sheet that bondholders rely on.

Hartnett also believes the long stretch in which AI bonds lagged AI stocks is coming to an end.

Hyperscalers have raised $89 billion through equity issuance in 2026. At the same time, their share buybacks are running at an annualized pace of just $14 billion, down from $105 billion in 2024.

For Hartnett, that means the era of big stock buybacks is over. Cash that used to flow back to shareholders is now being kept inside the companies, which in turn favors those who own the debt.

The pressure comes from spending. Bank of America estimates AI capital expenditure will climb from $240 billion in 2024 to $1.2 trillion in 2027. Over the same period, free cash flow is seen swinging from $240 billion to negative $180 billion.

For this reason, Hartnett said, bond markets are now policing the AI boom.

S&P Global Data Show Credit Markets Are Still Nervous

The bond market has not given the all-clear yet.

Five-year credit default swaps, which show the annual cost of insuring a company's debt against default, have widened sharply across Big Tech since early September, according to S&P Global Market Intelligence data as of Friday.

Oracle's spread rose to 276 basis points from 210 on Sept. 1. Nvidia's climbed to 129 from 84, while Meta's increased to 118 from 94.

Alphabet Inc. (NASDAQ:GOOGL) moved to 85 from 61, and Amazon to 90 from 67. Microsoft remained the safest name, at 55 from 47.

However, there are early signs of stabilization. Oracle's spread has eased from a peak of 325 basis points on Oct. 20.

S&P Global has flagged a deeper shift. In an Oct. 13 update, it said spreads on AI-exposed tech companies have widened even as the broader investment-grade market has held up.

Investors, S&P Global said, increasingly treat AI exposure as a distinct credit risk. In addition, rising correlations among hyperscalers, chipmakers and other AI issuers mean tech credit is starting to trade as a single theme, eroding the diversification benefits of holding several names.

Supply will not fade soon. S&P Global said hyperscaler-related bond issuance reached about $240 billion in the first three quarters, on pace for its $250 billion to $300 billion full-year estimate.

The Levels Hartnett Is Watching

Hartnett expects risk-off conditions to persist until the global rate-hiking cycle ends. Central banks have delivered 42 hikes over the past six months.

He sees a release valve in the first half of 2027: geopolitical peace, a collapse in oil prices and the end of rate hikes.

Before any year-end rally, Hartnett said two levels must hold: $70 on the Roundhill Magnificent Seven ETF (CBOE:MAGS) and $520 on the VanEck Semiconductor ETF (NASDAQ:SMH).

On Friday, MAGS slipped 2.2% to $69.70, just below that line. SMH remains about 8% above its threshold.

Treasury yields offer little relief. The 10-year yield stood at 4.93% and the 30-year at 5.29%. The iShares iBoxx $ Investment Grade Corporate Bond ETF (NYSE:LQD) is down 5.5% this year.

Hartnett's case rests on a simple shift. Shareholders have funded the AI race so far. Now bondholders are being paid more to finance the next leg, even as default-protection costs show that credit markets remain wary.