Although still at a nascent state, quantum computing promises advances in healthcare, from drug discovery to personalised treatments. But it also threatens current cryptographic systems that protect medical data and infrastructure. The concept of “Q-Day” highlights risks such as “harvest now, decrypt later” attacks, with particular concerns for medical devices and sensitive applications in fields like femtech. Preparing for this future requires the rapid adoption of post-quantum cryptography, the coordination of time-phased and scalable “technology rollout” strategies, and revised regulatory frameworks to safeguard patient safety, privacy, and trust.
You’re reading the web edition of STAT’s Health Tech newsletter, our guide to how technology is transforming the life sciences. Sign up to get it delivered in your inbox every Tuesday and Thursday.
Abridge vs. OpenEvidence, round one
New announcements from Abridge and OpenEvidence, two of the most prominent health tech startups to emerge in recent years, highlight how despite different initial offerings, many artificial intelligence companies in health care will just end up competing with each other for physician eyeballs.
On Monday morning, Abridge, best known for its AI scribe that helps doctors automate the writing of clinical notes, announced a new product that will surface “real-time insights, prompts, and pathways” from the widely-used medical resource UpToDate based on things said in a clinical conversation and that are written in the patient’s record.
Background: Obesity is a chronic, relapsing disease influenced by environmental, lifestyle, biological, and genetic factors, affecting over 1 billion people globally. Treatment for adults typically involves multicomponent lifestyle interventions—diet, physical activity, and behavior change—for at least 6-12 months. However, adherence is often low, and in-person sessions can be time-consuming and costly. Digital therapeutics (DTx), which enhance patient engagement and support long-term outcomes, have proven effective in managing chronic and mental health conditions. DTx offer scalable, evidence-based solutions with the potential to improve obesity management. Objective: The Digital Therapy to Promote Weight Loss in Patients With Obesity by Increasing Their Adherence to Treatment (DEMETRA) study is a prospective, multicenter, pragmatic, randomized, double-arm, single-blind, placebo-controlled trial evaluating the 6-month efficacy of an innovative, multicomponent digital intervention for obesity, which combines dietary, physical activity, and behavioral strategies in people with obesity (primary objective). Secondary objectives were assessing changes in BMI, waist circumference, blood pressure, glucose metabolism, lipid profile, adherence, and factors associated with absolute 6-month weight loss. Methods: The trial was conducted at 2 obesity centers in Italy with 246 participants aged 18-65 years (BMI 30-45 kg/m2), randomly assigned to either the Digital Therapeutics for Obesity (DTxO) app or a placebo app. DTxO offered personalized diet plans, exercise routines, and psycho-behavioral support, while the placebo app only allowed users to log data without feedback. Both groups followed a Mediterranean-style low-calorie diet with an 800 kcal/day deficit. On average, participants used the DTxO app for 42 minutes/day and the placebo app for 35 minutes, primarily for physical activity tracking. Univariable and multivariable generalized linear models were used to assess associations with 6-month absolute weight change (primary end point) and percent weight change (secondary end point). Results: Overall, 207 participants (84.1%) completed the 6-month visit. Both arms achieved a statistically significant absolute (DtxO: –3.2 kg, IQR –6.0 kg to –0.9 kg; placebo: –4.0 kg, IQR –6.9 kg to –0.5 kg; P<.001) and percent loss in body weight (DtxO: –3.0%, IQR –5.7% to –0.8%; placebo: –4.0%, IQR –8.5% to –0.5%; P<.001) after 6 months, without significant between-group differences (univariable generalized linear models: P=.34 and P=.17, respectively). Univariable regression analyses showed a significant association between adherence to app use and 6-month absolute weight loss (β=–.06, SE 0.02, P=.01) as well as percent weight loss (β=–.05, SE 0.01, P=.01). Adherent participants, defined as those with overall adherence at or above the 75th percentile of daily usage, included 35 individuals in the intervention group and 10 in the placebo group. In this subgroup, the estimated 6-month mean absolute weight change was –7.02 kg (95% CI –9.45 to –4.59) in the DTxO-adherent group and –3.50 kg (95% CI –7.01 to 0.01) in the placebo-adherent group (P=.02). The estimated 6-month mean percent change in weight was –6.31% (95% CI –8.86 to –3.76) in the DTxO-adherent group and –2.78% (95% CI –6.48 to 0.92) in the placebo-adherent group (P=.03). A significantly greater weight loss (P=.01 for study arm, either on absolute or percent change in weight from baseline) among adherent participants randomized to the DTxO app was also confirmed by analyses using mixed linear models for repeated measures. Conclusions: Although overall weight loss did not differ significantly between the DTxO and placebo groups, participants who used the DTxO app for at least 40% of the expected time achieved significantly greater weight loss. These results suggest that higher engagement with DTx can improve obesity outcomes. Further research should explore combining DTxO with pharmacological treatments or bariatric surgery. Trial Registration: ClinicalTrials.gov NCT05394779; https://clinicaltrials.gov/ct2/show/NCT05394779
Background: Retrieval-augmented generation (RAG) is increasingly used to improve large language models in the medical and nursing domains. However, a comprehensive understanding of its specific architecture and applications in medical and nursing reasoning remains limited. Objective: We aimed to summarize the current state, existing limitations, and future development directions of RAG in the medical and nursing domains. Methods: The PubMed, Web of Science, IEEE Xplore, and arXiv databases were searched for relevant articles using queries that combined terms related to RAG, medical, and nursing domains, covering the period from November 1, 2022, to May 31, 2025. This review was conducted following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines. Results: A total of 917 articles were retrieved, of which 67 met the inclusion criteria. Most studies focused on the medical domain (63/67, 94%), while only a few addressed nursing applications (4/67, 6%). The RAG frameworks included in this review were categorized into 5 functional types: text-based RAG (36/67, 54%), knowledge graph–enhanced RAG (17/67, 25%), agentic RAG (6/67, 9%), multimodal RAG (2/67, 3%), and plug-and-play RAG (6/67, 9%). On the basis of the Simon decision-making process theory, we divided the RAG workflow into 4 stages: intent recognition, knowledge retrieval, knowledge integration, and generation. Only 26 studies included explicit reasoning support, and few were aligned with real-world clinical workflows. Only 12 studies attempted to address ethical considerations related to RAG. Conclusions: We identified 4 key shifts in recent RAG development: shifting from surface-level matching toward contextualized intent recognition, from vague semantics toward logic-driven dynamic retrieval, from passive toward active knowledge retrieval, and from simple aggregation toward coherent context construction. However, most RAG systems in the medical and nursing domains have not yet introduced reasoning methods, and those that have are still predominantly reliant on data‑driven associations without causal modeling. This highlights the need to integrate causal mechanisms for more effective and domain-relevant reasoning in health care. Trial Registration: OSF Registries 10.17605/OSF.IO/WBSV5; https://osf.io/wbsv5
Lab Invest. 2025 Oct 15:104253. doi: 10.1016/j.labinv.2025.104253. Online ahead of print.
ABSTRACT
There are often performance differences between intra-dataset and cross-dataset tests in machine learning (ML) modeling. However, reducing these differences may reduce ML performances. It is thus a challenging dilemma for developing models that excel in intra-dataset testing and are generalizable to cross-dataset testing. Therefore, we aimed to understand and improve performance and generalizability of ML in intra-dataset and cross-dataset testing. We evaluated 4,200 ML models of classifying lung adenocarcinoma (LUAD) deaths using the The Cancer Genome Atlas (TCGA, n=286) and Oncogenomic-Singapore (OncoSG, n=167) datasets, and 1,680 models of classifying glioblastoma deaths using TCGA (n=151) and Clinical Proteomic Tumor Analysis Consortium (CPTAC, n=97) datasets. After examining performance distributions of these ML models, we applied a dual analytical framework, including statistical analyses and SHapley Additive exPlanations-based meta-analysis, to quantify factors' importance and trace model success back to design principles. We also developed a framework to identify the best generalizable model. Strikingly, Jarque-Bera test revealed significant deviations of model performances from normality in both cancer types and testing contexts. Simple linear models with sparse feature sets consistently dominated in LUAD experiments, whereas non-linear models dominated in glioblastoma ones, suggesting that the best modeling strategy appears cancer-type/disease dependent. Importantly, both robust Analysis of Variance (ANOVA) and Kruskal-Wallis tests consistently identified differentially expressed genes as one of the most influential factors in both cancer types. The proposed multi-criteria framework successfully identified the model that achieved both the best cross-dataset performance and similar intra-dataset performance. In summary, ML performance distributions significantly deviated from normality, which motivates using both robust parametric and non-parametric statistical tests. We quantified and provided possible exploitability on the factors associated with cross-dataset performances and generalizability of ML models in two cancer types. A multi-criteria framework was developed and validated to identify the models that are accurate and consistently robust cross datasets.
Front Immunol. 2025 Oct 2;16:1673098. doi: 10.3389/fimmu.2025.1673098. eCollection 2025.
ABSTRACT
INTRODUCTION: Pancreatic ductal adenocarcinoma (PDAC) continues to exhibit resistance to immunotherapy. In this study, we evaluated the efficacy of combining immunotherapy with chemotherapy for the treatment of advanced pancreatic cancer. Additionally, we employed a multimodal analytical approach to elucidate the immune landscape and conduct transcriptomic profiling in PDAC.
METHODS: A retrospective analysis was conducted on the clinical data of 52 patients diagnosed with advanced PDAC who underwent a combined treatment regimen of immunotherapy and chemotherapy. The study evaluated the objective response rate (ORR), disease control rate (DCR), and progression-free survival (PFS). To characterize the immune landscape in treatment-naive pancreatic ductal adenocarcinoma (PDAC) tumors and in the systemic circulation, flow cytometry, multiplex immunohistochemistry (mIHC), and whole transcriptome sequencing were employed.
RESULTS: The study reported an ORR of 32.7%, a DCR of 67.3%, and a 6-month PFS rate of 38.5%, with a median PFS of 5.5 months. Patients treated with a combination of immunotherapy and gemcitabine achieved the longest PFS. The first-line treatment cohort exhibited a significantly higher DCR (79.3% vs. 52.2%, P = 0.038) and a longer median PFS (6.6 vs. 3.5 months, P = 0.032) compared to the second-line treatment cohort. The efficacy of treatment varied depending on the drug combinations used. Flow cytometry analysis revealed a greater frequency of CD45- CD64+ cells in the peripheral blood of patients with progressive disease (PD) compared to those with a partial response (PR). Multiplex immunofluorescence (MIF) analysis indicated an increased intratumoral infiltration of CD8+ T cells and CD137+ CD8+ T cells in patients with PR. Whole transcriptome sequencing (WTSS) identified key genes involved in immune regulation, signal transduction, and digestive function. Hemopexin (HPX) and regulatory factor X-associated protein (RFXAP) were upregulated in PR patients and showed a positive correlation with survival, whereas Interleukin-6 (IL-6) expression was linked to poor prognosis.
CONCLUSIONS: These findings indicate that immunochemotherapy shows potential for the treatment of advanced PDAC. Our study elucidates the immune landscape associated with PDAC and provides critical insights for the identification of prospective therapeutic targets, which could guide the development of innovative combination immunotherapy strategies.
A study shows that creativity enhances brain health by improving connectivity in age-vulnerable regions, highlighting the importance of supporting creative activities in public health strategies.
Background: Research typically shows a higher preference for professionally-led face-to-face mental health interventions over digital ones. It remains unclear in which circumstances digital self-help tools are preferred. To address this gap, it is important to examine user characteristics that may help predict when digital interventions are more desirable, ultimately guiding their design to enhance engagement and appeal. Objective: To examine how distress severity and capacity to invest effort relate to intervention preferences, using an ecological assessment of individuals who seek to receive feedback on their mental health. Methods: A comprehensive digital mental health screening tool providing automated feedback was developed and advertised on social media. The sample comprised 684 adult participants aged 18-82 who opted to complete the screening to receive feedback on their mental health state. Participants completed questionnaires measuring general psychological distress, depression, generalized anxiety and demographics. Kessler Psychological Distress Scale–6 was used as the primary measure for distress. Participants were also presented with questions measuring capacity to invest effort and preferences for a professional vs digital self-help tools and for psychotherapy vs a mobile application. The effectiveness of distress, capacity to invest effort, and background characteristics in predicting preferences (a professional vs digital self-help tools; psychotherapy vs a mobile application) was examined using hierarchical linear regressions. The distributions of dichotomized preferences were plotted against distress and capacity to invest effort for transparent visualization. Results: A hierarchical linear regression found that distress, capacity to invest, and currently being in psychotherapy significantly predicted preference for a professional vs digital self-help tools. Distress (β=.25, 95% CI .18 to .32, P<.001 and capacity to invest effort ci .16 .30 p were the strongest predictors with similar effect size. model explained of variance in preference uniquely contributing most distressed participants low preferred digital self-help tools whereas high favored a professional. results obtained when using phq-4 as an alternative distress measure. remained significant .10 .26 predicting for psychotherapy vs mobile application while was not .05 conclusions: this study highlights that interventions is driven by reduced intervention. attempts reduce mental health treatment gap through should focus on optimizing elicited users improve desirability engagement.>
In Silico Pharmacol. 2025 Oct 17;13(3):154. doi: 10.1007/s40203-025-00440-3. eCollection 2025.
ABSTRACT
Breast, ovarian, lung, cervical, and colorectal cancers are among the most prevalent malignancies affecting women worldwide. This study aimed to elucidate the common molecular mechanisms of tumorigenesis and identify potential biomarkers using an integrative bioinformatics and network-based approach. Integrative profiling of five microarray datasets identified 66 differentially expressed genes (DEGs) that are common across five cancer types. Gene ontology and KEGG pathway analyses of common DEGs were performed using the DAVID database. The cell cycle processes were the most enriched functions, and oocyte meiosis, oocyte maturation, the p53 signaling pathway, cancer pathways, and cellular senescence were the most important pathways identified. Protein-protein interaction (PPI) networks for the DEGs were constructed using the STRING database, and the resulting networks were visualized in Cytoscape. Through PPI network analysis, ten hub genes were identified, and subsequent survival analysis confirmed that CHEK1, DLGAP5, CCNB2, and CCNA2 are significantly associated with poor patient survivability, establishing them as common biomarkers across multiple cancer types. Subsequently, ten transcription factors (TFs) and ten post-transcriptional regulators were identified through the assessment of regulatory networks involving TFs-DEGs and miRNAs-DEGs. Finally, drug-gene association analysis from the GSCA library was used to anticipate drug-like compounds using the drug repurposing approach. Overall, this comprehensive investigation holds promise for future in vitro and in vivo studies, offering a molecular foundation for the diagnosis, prognosis, and treatment of malignant cancers.
SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s40203-025-00440-3.
OMICS. 2025 Oct 17. doi: 10.1177/15578100251387873. Online ahead of print.
ABSTRACT
Cancer is a disease with heterogenous molecular signatures that ought to be unpacked to achieve the overarching aim of precision oncology. A pan-cancer omics approach provides a systems science framework to explore shared and distinct mechanisms across cancers. We report here pan-cancer analyses of gene expression data from 17 cancers, for example, adrenocortical cancer, lung cancer, kidney cancer, and colorectal cancer, and 26 tissue types, using public datasets to construct disease-specific transcriptional networks. Using the hypergeometric test, 1005 microRNAs (miRNAs), 314 transcription factors (TFs), and 332 receptors were identified as regulatory molecules interacting with differentially expressed genes. Kyoto Encyclopedia of Genes and Genomes pathway analysis was performed to explore their functional roles. Accordingly, we found miR-124-3p, miR-6799-5p, and miR-7106-5p as common miRNAs; Specificity Protein 1 (SP1), RELA Proto-Oncogene, NF-κB Subunit (RELA), and Nuclear Factor Kappa B Subunit 1 (NFKB1) as shared TFs; Cyclin-Dependent Kinase 2 (CDK2), Histone Deacetylase 1 (HDAC1), and ABL Proto-Oncogene 1, Non-Receptor Tyrosine Kinase (ABL1) as common receptors; and pathways in cancer, PI3K-Akt signaling, and p53 signaling as commonly enriched. Survival analysis in an independent dataset confirmed these findings: SP1 and NFKB1 were significant in 9 cancers, RELA in 6, whereas CDK2, HDAC1, and ABL1 were significant in 11, 10, and 10 cancers, respectively, out of the 17 cancers researched herein. In conclusion, these findings provide system-level insights on tumor heterogeneity and inform future cancer classification, for example, according to shared and distinct molecular signatures and development of therapies that might prove effective across several cancers. We underline that unpacking molecular signatures across multiple cancers also offers new prospects to move beyond the "One Drug, One Disease" paradigm of pharmaceutical innovation.
Animal models provide insights that alternative approaches can’t. A heated debate about which method is better is distracting researchers from doing good science.
Biomedical research techniques that don’t involve the use of animals are gaining momentum, but those using innovative approaches still face resistance from some quarters.
Oral Oncol. 2025 Nov;170:107760. doi: 10.1016/j.oraloncology.2025.107760. Epub 2025 Oct 17.
ABSTRACT
Non-viral head and neck squamous cell carcinoma (HNSCC) has poor survival and high recurrence rates. Circulating tumor DNA (ctDNA) is a promising biomarker for understanding tumor biology, assessing treatment response, and monitoring disease progression. While extensively studied in virally mediated HNSCC, its role in non-viral HNSCC remains underexplored. This systematic review and meta-analysis consolidates evidence on the diagnostic, prognostic, and therapeutic value of ctDNA in non-viral HNSCC. A systematic search across Medline, PubMed, Embase, and the Cochrane Library identified 1,915 records, of which 47 were included. Data extraction followed PRISMA guidelines, with overall survival (OS), progression-free survival (PFS), and recurrence-free survival (RFS), pooled as hazard ratios (HRs) with 95% confidence intervals (CIs) using a fixed-effect model. Among 3,574 patients, the most common tumor sites were the oral cavity (35 %) and oropharynx (22 %), with the majority presenting with stage IVA/IVB disease (29 %). Pre-treatment ctDNA detection rates ranged from 50 % to 100 % (median: 83 %), while post-treatment detection rates varied between 28 % and 100 % (median: 48 %). ctDNA detected recurrence in 80 % of patients, with a median lead time of 4.6 months. ctDNA detection was significantly associated with worse OS (HR 10.26, 95 % CI 3.58-29.40; P < 0.0001). Residual ctDNA was strongly correlated with worse PFS (HR 7.32, 95 % CI 4.17-12.86; P < 0.00001) and RFS (HR 7.33, 95 % CI 2.75-19.58; P < 0.0001). ctDNA holds potential for improving diagnostic accuracy, monitoring progression, and predicting survival outcomes in non-viral HNSCC. However, further large-scale studies and standardized guidelines are needed for validation and clinical implementation.
Nat Commun. 2025 Oct 17;16(1):9232. doi: 10.1038/s41467-025-64292-3.
ABSTRACT
Recent advancements in spatial transcriptomics technologies have significantly enhanced resolution and throughput, underscoring an urgent need for systematic benchmarking. Here, we generate serial tissue sections from colon adenocarcinoma, hepatocellular carcinoma, and ovarian cancer samples for systematic evaluation. Using these uniformly processed samples, we generate spatial transcriptomics data across four high-throughput platforms with subcellular resolution: Stereo-seq v1.3, Visium HD FFPE, CosMx 6K, and Xenium 5K. To establish ground truth datasets, we profile proteins on tissue sections adjacent to all platforms using CODEX and perform single-cell RNA sequencing on the same samples. Leveraging manual nuclear segmentation and detailed annotations, we systematically assess each platform's performance across capture sensitivity, specificity, diffusion control, cell segmentation, cell annotation, spatial clustering, and concordance with adjacent CODEX. The uniformly generated and processed multi-omics dataset could advance computational method development and biological discoveries. The dataset is accessible via SPATCH, a user-friendly web server for visualization and download.
Mol Med Rep. 2026 Jan;33(1):6. doi: 10.3892/mmr.2025.13716. Epub 2025 Oct 17.
ABSTRACT
R‑loops, three‑stranded nucleic acid structures composed of an RNA:DNA hybrid and displaced single‑stranded DNA, have emerged as important regulators of gene expression and genome maintenance. Although physiological R‑loops participate in normal cellular processes, their dysregulation can threaten genomic integrity by inducing DNA damage and replication stress. The present review explores the role of R‑loops in hepatocellular carcinoma (HCC), a malignancy characterized by marked genomic instability. In the present review, the formation mechanisms of R‑loops, their dual functions in transcriptional regulation and DNA damage, and their specific implications for HCC pathophysiology were discussed. HCC cells exhibit altered R‑loop homeostasis with aberrant accumulation linked to hepatitis B virus infection, inflammatory signaling and oncogene activation. The present review highlighted how HCC cells exploit or manage R‑loops to promote tumor progression, particularly through the epigenetic silencing of differentiation genes and modulation of replication stress responses. Furthermore, emerging therapeutic strategies targeting R‑loop biology were examined, including small molecules that induce synthetic lethality, gene‑based interventions and combination approaches that exploit R‑loop vulnerabilities. Challenges in targeting R‑loops and future directions, including multi‑omics profiling and biomarker development, were also addressed. Understanding the complex interplay between R‑loops and HCC offers promising avenues for novel diagnostic and therapeutic approaches for this malignancy.
The surge of multi-billion-dollar investments in AI has sparked growing debate over whether the industry is heading for a bubble similar to the dot-com boom.
Investors are watching closely for signs that enthusiasm might be fading or that the heavy spending on infrastructure and chips is failing to deliver expected returns. A recent survey by BofA Global Research found that 54% of fund managers believe AI stocks are already in bubble territory, while 38% disagree.
Echoes of the dot-com era
Despite the optimism surrounding AI, sceptics remain unconvinced of its real-world impact. Some even call it a bluff or a bubble waiting to burst.
Speaking during Cisco’s recent Virtual Media Roundtable — AI Readiness Index 2025: Readiness Leads to Value, Ben Dawson, Senior Vice President and President for Asia Pacific, Japan, and Greater China (APJC), compared the current wave of AI hype to the early days of the internet. He said technological shifts of this scale often follow a familiar pattern — early excitement, heavy investment, and eventual market correction before long-term value takes hold.
Dawson noted that while some AI projects or business models may not last, the overall transformation is real and lasting. He added that, much like the internet revolution, AI will permanently reshape business and society, and organisations that ignore it do so at their own risk.
The role of governments and global policy
Public policy is also shaping how the AI cycle unfolds — and how governments might cushion the risks of a potential AI bubble. As Harvard Business Review pointed out, in the US, government involvement has helped define past technology eras — often through incentives and early investments that encourage private innovation. The same pattern is now visible in AI. Both the Trump and Biden administrations have positioned AI as a matter of economic strength and national security, sending a clear message that speed matters.
China has taken a state-led approach, directing capital toward local AI firms to reduce reliance on US technology. In Europe, efforts have focused more on regulation, though fears of overregulation have led to new programs — such as the AI Continent Action Plan and a €1 billion Apply AI fund — to boost adoption and competitiveness.
Meanwhile, venture capital and sovereign wealth funds are investing heavily, even before widespread AI demand exists. These early bets assume that adoption will eventually justify the buildout. But if that demand slows, some investors could be left with stranded assets, much like the unused fibre networks that followed the dot-com bubble.
For businesses, the challenge is different. Instead of financing the next infrastructure wave, they face the question of how to use AI to strengthen their operations. The companies that survived the dot-com downturn — such as Amazon — succeeded by aligning technology with real business value rather than market hype.
Market warnings over a possible AI bubble
The Bank of England recently warned that markets could suffer a sharp correction if confidence in AI falters, calling the potential impact on the UK’s financial system “material.” The warning reflects growing caution among policymakers about how quickly AI-related valuations have climbed.
This concern is shared by some investors and economists who believe the rapid pace of AI spending may outstrip short-term returns. Others, however, argue that building AI infrastructure now is essential groundwork for future innovation.
Building long-term AI infrastructure amid bubble fears
When asked whether companies are worried about AI infrastructure costs and energy demand, Simon Miceli, Managing Director of Cloud and AI Infrastructure for APJC at Cisco, said he views the issue from the opposite angle.
Rather than fearing overcapacity, he said what’s happening now is a large-scale buildout to support the industrialisation of AI. The question, he said, isn’t whether AI demand exists today, but whether the world is preparing fast enough for what’s coming.
Miceli acknowledged that some correction in the AI market is likely, but he believes the long-term need for AI computing power justifies current investment levels. “There’s a race to develop AI and build the capability behind it,” he said, adding that demand will eventually meet supply as applications mature.
Different shades of caution
Across the industry, opinions vary on whether AI’s momentum represents hype or healthy growth.
According to Reuters, at the Milken Institute Asia Summit 2025, Singapore’s GIC Chief Investment Officer Bryan Yeo said valuations in early-stage AI ventures appear inflated, with many startups commanding “huge multiples” despite modest revenues. He suggested that while some firms may justify their valuations, others are unlikely to deliver returns that match investor expectations.
Jeff Bezos, Amazon’s founder, said that during periods of excitement like this, investors often struggle to separate good ideas from bad ones — though he also noted that innovation-driven bubbles often leave behind real progress once the market settles.
At Goldman Sachs, economist Joseph Briggs argued that the current surge in AI infrastructure spending remains economically sustainable. He said the long-term case for AI investment is strong, but the ultimate winners are still uncertain given how quickly technology changes and how easily companies can switch providers.
Meanwhile, ABB CEO Morten Wierod told Reuters that while he doesn’t see an AI bubble, supply chain and construction limits could slow the rollout of new data centres. IMF Chief Economist Pierre-Olivier Gourinchas added that even if there’s a downturn, it’s unlikely to cause a systemic financial crisis since AI investments aren’t debt-driven.
OpenAI CEO Sam Altman also acknowledged market overexcitement, predicting that some investors will lose large sums while others will profit heavily — an outcome that mirrors past technology bubbles.
Despite growing talk of an AI bubble, many investors remain committed to the sector. UBS equity strategists said that about 90% of investors who think the market is overheated are still holding AI-related assets, suggesting most believe the industry has not yet peaked.
A cycle, not a collapse
While concerns about an AI bubble are valid, most experts agree that the technology’s long-term impact is undeniable. As Cisco’s Ben Dawson put it, every major technological transition goes through a cycle of hype, correction, and consolidation — but what remains afterward reshapes industries for decades.
For now, the question isn’t whether AI will endure, but how well businesses and investors can navigate the growing pains that come with every market bubble.
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.
AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
Login
