- Key insight: Upstart's stock plummeted 14.8% not because of poor profits, but because its own AI model intentionally "tightened the credit box," causing a miss on loan origination volume.
- Supporting data: The company reported $2.9 billion in loan originations, falling significantly short of the $3.3 billion analyst consensus, even as consumer applications hit a three-year high.
- Expert quote: CEO Dave Girouard called the model's behavior a "speed bump" and admitted it was "overreacting," but analysts from Jeffries found the company's explanation "confusing."
Overview bullets generated by AI with editorial review
Upstart, a lending marketplace powered by AI, saw its stock price decline 14.8% following its third-quarter earnings call Tuesday afternoon.
Despite reporting strong profitability that beat analyst expectations, the market reacted negatively to disappointing loan origination volume and cautious guidance for the remainder of the year.
Upstart shares, which closed at $46.24 on Tuesday, opened trading Wednesday at $39.38. Year-to-date, the company's stock remains under pressure, down 35.7% since the beginning of 2025.
Analysts noted that uncertainty surrounding the effectiveness of Upstart's underwriting model and lower forward-looking expectations contributed to the stock's performance.
Despite the negativity, Upstart, which had been unprofitable for 12 straight quarters until
The company had net income of $32 million, beating the analyst consensus of $10.2 million, according to estimates collected by S&P Market Intelligence. The company had a net loss of $6.8 million in the same quarter last year.
This net income also translated to a diluted earnings per share, or EPS, this quarter of $0.23, compared to the consensus of $0.08.
"In Q3, we continued to execute on our 2025 game plan of rapid growth, profitability and AI leadership — all anchored in exceptional credit performance," said Upstart CEO and founder Dave Girouard.
Loan originations slowed
The headline miss that concerned investors was in loan origination volume. Upstart had approximately $2.9 billion in originations on the quarter, falling below the consensus expectation of $3.3 billion, according to Jefferies analysts.
"3Q25 results show slowing momentum, driven by a more selective model," read an analyst note from John Hecht, an equity analyst at Jefferies.
Upstart's signature AI models intentionally pulled back on lending activity because of mixed risk indicators, according to company leadership.
Upstart's chief technology officer Paul Gu said during the earnings call Tuesday that Upstart's AI model sensed macroeconomic stresses during the summer, leading it to tighten the credit box.
The model is designed to "respond with speed and precision to changes in macro conditions," according to Gu. "A few months ago, that led the model to tighten on credit while certain risk signals were elevated before recently normalizing."
This behavior "partially reflects irreducible volatility in the outside world, but is also a function of our model design and sampling variance, both of which continue to improve," Gu added.
Hecht said in his analyst note that the company's explanation for why the model shifted during the quarter "was confusing to us and is in contrast to the overall market."
This temporary conservatism resulted in a reduction in the company's conversion rate from 23.9% in the previous quarter to 20.6% in Q3. The conversion rate reflects the number of loans approved, not their dollar value.
Upstart's AI model converted fewer loans despite consumer demand growing rapidly, evidenced by application submissions reaching the highest level in over three years.
CEO Girouard acknowledged the model was "overreacting" and called the result a "speed bump."
"I think in some sense, having a model that overreacts is better than having ones that underreact because it did revert," he told investors.
Analysts skeptical of underlying performance
David Scharf, managing director of equity research at Citizens Bank, said in an analyst note that the bumps "raise questions regarding how differentiated [Upstart's] model is compared to other machine learning-based lenders."
Upstart has marketed its underwriting models as a differentiator, but the "bumps or misreads inject a dose of incremental uncertainty in the company's ability to deliver industry-leading risk-adjusted yields to investors in its loans," according to Scharf.
Upstart's revenue of $277 million — a 71% improvement year over year — slightly missed the analyst consensus of $280 million. Hecht represented the company's third-quarter results as "largely weaker than consensus estimates."
Scharf said Citizens expects the third-quarter trends "to continue into next year and to see them incrementally lowering bottom-line performance."
As such, the bank slightly lowered its estimate for Upstart's 2026 performance. Specifically, it changed its estimate for the company's adjusted EBITDA (earnings before interest, taxes, depreciation and amortization) from $316 million to $313 million.
