Dynex Eats a 15% Book Value Mark and Refuses to Blink

A mortgage REIT blames the AI capex boom for its losses, adds preferred capital for the first time in six years, and lets leverage drift past its own stated range.

DX · Earnings Call · 2026-10-19

The mark is taken; the carry is not

Dynex lost 15.4% of book value in a single quarter. Book value per common fell to $10.91 from $12.90, total economic return was negative 11.4%, and comprehensive loss ran $1.49 per share against income of 80 cents in Q2. The stock has already voted: down 18.5% over the last 90 trading days, some 22% below its July 16 peak of $13.53. So this call is not a shock — it is a confirmation, and the interesting question is what management is asking you to price next. “Generic repricing is a normal occurrence in the markets. And the mark to market impact of the third quarter's repricing is already reflected in our assets, our balance sheet, our book value, and our share price.” — Smriti Laxman Popenoe, Co-Chief Executive Officer and President · 2026-10-19 The tell that the mark may be trough: book value quarter-to-date through Friday October 16 was estimated at $10.95, net of the accrued dividend — a small but real uptick from the $10.91 quarter-end print, which the tape's -18.5% slide does not yet reflect.

AI as a rates story, not a mortgage story

Here is what is genuinely new. In six quarters of prior Q&A, AI showed up only as a passing curiosity — refinancing algorithms, an "advent of AI" aside. This quarter it became the entire macro frame. Dynex argues the US is in a significant positive investment cycle “Driven by AI and the infrastructure required to support it” — Smriti Laxman Popenoe, Co-Chief Executive Officer and President · 2026-10-19, and that the consequences — demand for capital, the price of that capital, commodity prices, and then inflation — are all pushing yields higher. The punchline is a rare admission from a mortgage REIT: “agency MBS have actually been a sideshow, not the main act” — Smriti Laxman Popenoe, Co-Chief Executive Officer and President · 2026-10-19. That reframe is worth flagging against the global tape. The AI data centers theme is up 53.8% over 360 days across 96 tickers — but negative over the last 90 days, with 47 decliners against 6 advancers. The AI infrastructure trade has cooled at the margin, even as its rate consequences are what just marked Dynex's book. Meanwhile data center buildout is a live theme at Prologis. Dynex is, functionally, on the wrong end of that capex cycle: a levered duration holder watching the term premium pay for the boom.

The deliberate trade: more spread risk, less rate risk

Management did not hide behind the loss — it leaned in. Spread sensitivity rose from roughly 4.6% of common equity per 10 basis points at June 30 to 5.7% at quarter end. Rate sensitivity collapsed in the opposite direction, as T.J. Connelly put it:

An instantaneous 50 basis point rise in rates as of September 30 is modeled to reduce common equity by only about 0.5%. Compared with 3.5% at June 30.
Terrence J. Connelly Jr., Chairman and Co-Chief Executive Officer · 2026-10-19

The mechanics: only about 1% of outstanding mortgages carry more than a 25 basis point refinance incentive, much of the universe has already extended, and convexity selling from servicers and levered-cap holders has largely run its course. That is why the convexity profile argument — ranked the company's third-highest keyword this quarter — is the load-bearing beam of the bull case. With current coupon spreads at roughly 162 basis points to 7-year swap, Connelly pegs static ROE on new swaps-hedged investments at around 20% in lower coupons and 25% in current coupons. Two things complicate that. First, the ROE math has not structurally improved — the entry point reset. In January, management guided to mid-teens hedged ROE at 7x leverage and mid-to-high teens in the low eights; in October 2025 it was high-teens net, mid-twenties gross. Those numbers were quoted against spreads the team expected to settle at “100, 120 basis points. I expect that to be the equilibrium over time.” — Terrence J. Connelly, Head of Capital Markets and Investor Relations · 2026-07-20 Spreads went the other way. The 20%+ figure is the same trade, repriced. Second, leverage. It ended at 9.0x, up from 8.1x, and Connelly noted net purchases since quarter end. That is past the 7.5x–8.5x band called comfortable in July and past the philosophical line Smriti Popenoe drew three months ago: “high levels of leverage are sort of out of the picture, out of the scope, simply because of our deep respect for the macro environment” — Smriti Laxman Popenoe, Co-Chief Executive Officer and President · 2026-07-20. The defense is that leverage drifted mechanically as equity fell and that not delevering preserved earnings power — “We did not delever.” — Terrence J. Connelly Jr., Chairman and Co-Chief Executive Officer · 2026-10-19 Fair, but the resulting risk posture is one the prior call explicitly disavowed.

A capital-structure promise finally kept

The quiet win is the balance sheet. Dynex raised $384 million of accretive capital, including $268 million of common through the ATM, and priced a $116 million Series D — its first preferred in more than six years — lifting preferred from roughly 3% to 7% of the capital stack. This closes a loop opened 15 months ago, when the answer on preferred was blunt: “The preferred markets have generally been very spotty and perhaps not even open at this point... for now, the focus is on the common.” — Bose George, Analyst · 2025-07-21 Now the language is “That number has room to go higher.” — Smriti Laxman Popenoe, Co-Chief Executive Officer and President · 2026-10-19 Scale is real, and it is what the filing confirms rather than the call: stockholders' equity reached $3.2 billion, up 97% year over year, with repo financing at $22.6 billion, up 163%. Net interest income of $94 million was up 305% year over year — a balance-sheet artifact, not a margin story, since interest expense rose 136% to $209 million and the economic net interest spread held flat at roughly 1.18%. Leverage, in the accounting sense, sits at 88.0% of assets.

What went quiet

The most telling absence: Agency CMBS — a centrepiece of the 2025 diversification push, five-year lockouts at swaps plus 90 basis points, pitched to banks and insurers — gets zero mentions this quarter. Spec pools, the company's top-ranked theme in the prior quarter, is likewise gone from the narrative. So is the security-selection alpha story, replaced by something much blunter: hold the spread, earn the carry, wait. Even the GSE backstop has softened to “anecdotal evidence of GSE buying in late September” — Terrence J. Connelly Jr., Chairman and Co-Chief Executive Officer · 2026-10-19, versus the confident structural framing of prior calls. The marginal demand case now leans on Norges Bank chatter and $8 trillion of money market cash. What matters is that the downside is now a financing question, not a duration question. Repo-to-SOFR at 12–15 basis points and central clearing expansion are doing more work in this thesis than any mortgage fundamental. If AI capex keeps pressing the term premium, the 0.5%-per-50bp sensitivity protects book value — but spread widening does not, and that is exactly where Dynex has deliberately added risk.