Gilead's oncology future depends on molecules, not algorithms

THE BIOLOGICAL tissue of a cancer patient used to be examined under a microscope by a trained pathologist. Increasingly, it is being fed into an algorithm. Gilead Sciences has signed on to the trend, announcing on August 11th a collaboration with Nucleai, a small Israeli-American firm, to use artificial intelligence for tissue analytics in its antibody-drug-conjugate programmes. The deal is presented as a quiet edge in precision oncology. It is better understood as a symptom of a deeper anxiety.

Gilead is a company divided between what it does best and what it needs to do next. HIV remains its economic core: in the second quarter of 2026, HIV product sales reached $5.7 billion, up 12% on the year, and accounted for roughly three-quarters of total product sales. Biktarvy alone brought in $3.8 billion. By contrast, total oncology revenue was $873 million, up a modest 3% to $873 million. That is not diversification. It is a rounding error wearing a growth label.

The company knows it. Over the past year, Gilead has been on an acquisition binge to build oncology scale from scratch. It bought Tubulis, a Munich-based ADC developer, for $3.15 billion in upfront cash, plus up to $1.85 billion in milestones. It acquired Arcellx and Ouro Medicines in separate deals. The accounting consequences have been brutal. In Q2 2026, those transactions generated $11.2 billion in acquired in-process research and development charges, producing a GAAP loss of $8.45 per share and slashing cash reserves from $10.6 billion to $3.2 billion in a single quarter. The company has also pledged $32 billion to strengthen its American research footprint. That is not a portfolio strategy. It is a panic buy.

Antibody-drug conjugates are Gilead's chosen ticket to oncology credibility. An ADC attaches a potent cancer-killing payload to an antibody that home in on specific tumour cells, aiming to deliver chemotherapy with surgical precision rather than a carpet bomb. The trouble is that the ADC space is already one of the most crowded and competitive corners of biopharma. Bristol Myers Squibb, AstraZeneca, Daiichi Sankyo and J&J have larger ADC franchises. Gilead's own first ADC, Trodelvy, grew 26% to $457 million in Q2 but stumbled at the largest breast-cancer trial it mattered most: the Phase 3 ASCENT-07 study for HR-positive/HER2-negative metastatic breast cancer missed its primary endpoint.

Here is where Nucleai enters the picture. The company, founded in 2018 with only $60 million in total funding, makes an AI platform that analyses pathology images — routine haematoxylin and eosin slides and immunohistochemistry stains — to extract spatial and biomarker information that human pathologists cannot reliably quantify. The idea is elegant: rather than looking only at whether a tumour expresses a given target, the algorithm examines how that target sits within the tumour microenvironment, its spatial relationship to immune cells and blood vessels, and whether those patterns correlate with clinical outcomes. For ADC development, where the difference between a hit and a miss often turns on patient selection, the promise is tighter trial design and faster biomarker discovery.

The problem is one of scale. Nucleai is a pre-profit research tool, not a platform that has yet changed a regulatory outcome or a commercial decision. The Gilead partnership covers translational research: the generation of "candidate spatial biomarkers" and "novel biological insights" intended for future publications and presentations. No financial terms were disclosed. No exclusivity was granted. The collaboration sits two or more years from any impact on a product label or revenue line. It is the kind of partnership every large pharma company announces and few ever cite again.

To be sure, AI-powered pathology is not a vanity project. Digital and computational pathology are expanding rapidly, driven by advances in whole-slide imaging, multimodal data integration and improved infrastructure. Several AI tools have demonstrated superior performance over visual evaluation in tasks such as identifying invasive breast cancers. The regulatory pathway for AI/ML-based companion diagnostics is becoming clearer. Gilead's bet is that understanding tissue architecture will become as important as molecular alterations in selecting patients for next-generation therapies. That is not wrong in principle.

The question is whether a small AI collaboration with a startup that has raised less than a tenth of one per cent of Gilead's quarterly revenue changes the arithmetic. It does not. What changes the arithmetic is whether Trodelvy can expand beyond its current label, whether TUB-040 — Tubulis's lead asset for ovarian and lung cancer — actually delivers the durability that preclinical models suggest, and whether the company can stop burning $3 billion a deal to buy its way into a space where it has no track record. The Nucleai deal adds nothing to the answer and obscures the fact that Gilead's oncology franchise is still being assembled from spare parts.

A wider lens reveals the incentive structure driving all of this. HIV drug patents have long half-lives but finite ones. Biktarvy faces eventual generic and biosimilar pressure, even if that pressure is still years away. Descovy is growing fast, up 48% year-over-year in Q2, but from a smaller base of $967 million. Gilead's management has been candid about wanting "a stronger balance across therapeutic areas" to avoid being "overly dependent on any one aging franchise." That is a reasonable ambition. What is less reasonable is the assumption that buying your way into oncology will work as well as building from inside. The history of pharma M&A, from Bristol Myers Squibb's Celgene implosion to Novartis's chronic acquisition hangovers, suggests otherwise.

The stock currently trades at roughly 19 times trailing earnings, with a market capitalisation near $165 billion. AInvest's aggregate signal labels the stock a buy, with a fundamental rating of 7.42 out of 10. The market is clearly still pricing in HIV cash flow, which is durable and predictable. But the forward picture is clouded. Consensus estimates show Q2 2026 EPS of negative $7.26, reflecting acquisition charges that may recur as integration continues. Full-year 2026 guidance calls for non-GAAP losses between $0.30 and $0.65 per share.

The better way for Gilead to build oncology would be to concentrate its firepower on the assets it has already chosen, run the trials that matter, and resist the temptation to announce partnerships that sound innovative but carry no economic weight. The Nucleai collaboration is not a waste of money — it is far too small for that. It is a waste of attention. What investors should watch are Trodelvy's FDA submissions for first-line breast cancer, TUB-040's Phase 1b/2 readouts, and whether the company's cash position stabilises after a quarter that saw it spend $11.3 billion on acquisitions.

AI in pathology is genuine progress. But it does not make up for a strategy built on purchasing ambition rather than earning it. Gilead's oncology future depends on molecules, not algorithms.