Every AI story published this year seems to arrive with an opposite one attached. The same week a DeepMind spinout announces it's compressing drug discovery timelines from decades to months, another headline reports a six-figure round of AI-linked layoffs. Both stories are true. Neither cancels the other out. What follows is what the 2026 data actually shows on both sides, with the source attached to every number.
What's Actually Working
Drug discovery has moved from theory to Phase II. Insilico Medicine's INS018_055, a candidate for idiopathic pulmonary fibrosis, is the first drug where both the biological target and the molecule itself were identified entirely through AI, and it's now in Phase II trials. The company took the candidate from target discovery to Phase I in about 30 months, against an industry average closer to three to five years for a small-molecule drug. Isomorphic Labs, the DeepMind spinout built on AlphaFold, raised a $2.1 billion Series B in May 2026, backed in part by the UK's Sovereign AI Fund, and released an upgraded design engine called IsoDDE in February that the company says roughly doubles AlphaFold 3's accuracy on difficult drug-binding cases. Isomorphic is targeting its first human trial for an AI-designed candidate by the end of 2026, working with Novartis, Eli Lilly, and Johnson & Johnson. As of mid-2026, independent trackers counted more than 170 AI-designed or AI-enabled drug programs somewhere in clinical development industry-wide.
AI is also creating jobs, not just eliminating them, at some of the same companies cutting headcount. IBM cut roughly 200 HR roles to AI agents in 2026 while tripling its entry-level hiring elsewhere in the company. Meta laid off 8,000 employees this year while separately moving about 7,000 employees into new AI-focused roles internally. Company-level survey data backs this pattern up more broadly: 47 percent of companies told Resume.org in a February 2026 survey they're hiring more AI-focused staff, and 48 percent said they're hiring more workers who can use AI tools effectively, even as many of the same companies freeze entry-level hiring elsewhere.
What's Actually Breaking
The job displacement is real, and it's landing hardest on entry-level work. AI-attributed layoffs in the US reached roughly 205,000 workers through August 2026, according to a ResumePulse tracker, already matching the full 2025 total in under eight months. Challenger, Gray & Christmas, the outplacement firm that's tracked layoff reasons for decades, reported that AI was the single most-cited reason for corporate job cuts in March and April 2026, the first time a technology factor topped its monthly rankings. Entry-level professional job postings have dropped 29 percent since January 2024, according to tracking cited by MIT's Andrew McAfee, who has warned that automating away entry-level roles risks cutting the apprenticeship pipeline that produces senior talent later. Oracle's 30,000-person cut, tied to AI data center costs, was the largest single layoff event of 2026. Resume.org's February 2026 survey of business leaders found 51 percent expect their own company to cut existing staff in 2026 specifically because AI is consolidating roles.
Deepfake fraud has become a line item, not an edge case. The FBI's Internet Crime Complaint Center logged its first standalone AI-fraud category in its report published in April 2026, recording $893.35 million in losses across 22,364 complaints for 2025 alone. Entrust's 2026 Identity Fraud Report, drawing on more than a billion identity-verification events across 195 countries, found deepfakes now account for roughly one in five biometric fraud attempts worldwide. Gartner's 2025 survey of 302 security leaders found 62 percent of organizations had experienced at least one deepfake incident in the prior twelve months, and more than a third had encountered one on a live video call. The largest single verified case remains the $25.6 million loss at engineering firm Arup in Hong Kong, where finance staff wired funds after a video call with deepfaked senior executives.
Data centers are consuming electricity and water at a pace utilities weren't built for. The International Energy Agency's 2025 Energy and AI report put global data center electricity demand at roughly 415 terawatt-hours in 2024, projected to nearly double to about 945 terawatt-hours by 2030, with AI workloads driving most of that growth; electricity use at AI-focused data centers alone surged 50 percent in 2025. In Ireland, data centers now consume more than 20 percent of the country's metered electricity, prompting the national grid operator to pause approvals for new facilities. On the water side, a 100-megawatt AI data center typically uses between 1.5 and 3 million cubic meters of water a year for evaporative cooling, and Google's most recent environmental disclosure reported a 37 percent year-over-year increase in data center electricity use paired with a 34 percent increase in water use. A United Nations University report published in 2026, the first UN-commissioned study to quantify AI's water and land footprint alongside its carbon impact, found that data center water consumption still doesn't appear in most hyperscalers' sustainability disclosures at the resolution needed to assess local watershed impact.
The Part Both Sides Share
Look closely at the four data points above and a pattern shows up that's easy to miss when the stories run separately: every one of them is a scaling problem, not a capability problem. AI didn't fail at drug discovery, it worked well enough to reach Phase II in record time. It didn't fail at automating work, it worked well enough that companies are restructuring around it faster than the labor market can absorb the change. It didn't fail at generating convincing video and audio, it worked well enough that federal law enforcement needed a new fraud category. The technology is doing roughly what it was built to do in each case. What's lagging is the layer around it, the regulation, the labor market transition support, the fraud verification standards, and the water-use disclosure requirements that would normally develop alongside a technology moving this fast.
Congress has started responding on at least one front. The National Defense Authorization Act for Fiscal Year 2026, signed into law and reviewed by the Congressional Research Service in July 2026, now requires the Department of Defense to estimate the electricity and water impact of any new data center built on a military installation, the first federal water-reporting requirement tied specifically to data centers. Separate bills aimed at broader public water-use disclosure and water-recycling tax credits remained in committee as of late July 2026. On the fraud side, 47 US states have passed some form of deepfake legislation since 2022, totaling 169 separate laws, though standards vary widely by state and there's no federal equivalent yet.
What We Still Don't Know
Nobody has published a credible number for how many of the 205,000 AI-attributed layoffs in 2026 reflect AI actually doing the work faster, versus companies using AI as a convenient explanation for cuts they'd have made anyway. OpenAI's Sam Altman made exactly this argument in a June 2026 interview, and it's impossible to verify from the outside without access to internal productivity data no company has disclosed.
Nobody has settled whether Isomorphic Labs or a competitor will actually get an AI-designed drug through full FDA approval, since a Phase II trial or even a first human dose is not the same as market authorization, and drug approval timelines haven't historically compressed at the same rate as drug design timelines.
And nobody outside a handful of hyperscalers knows the true water and electricity footprint of the AI industry as a whole, because the disclosure standards that would make the numbers comparable across companies don't exist yet. Every figure in the section above comes from a different company reporting on a different basis, which is itself the story: the industry using the most water and electricity in a generation is also the industry least required to say exactly how much.
Written by Muntazir Mahdi, founder of ANFA Technology, for AI Future Insights. AI Future Insights covers artificial intelligence, automation, and future tech for readers who want signal over hype, built and maintained by a Karachi-based team working on privacy-first software including Canvas Convert Pro. This piece is sourced from Isomorphic Labs and Insilico Medicine's public disclosures, the IEA's Energy and AI 2025 report, the FBI Internet Crime Complaint Center's 2025 annual report, Entrust's 2026 Identity Fraud Report, Gartner's 2025 AI Risk Management Survey, Challenger Gray & Christmas layoff tracking, Resume.org's February 2026 business leader survey, and the Congressional Research Service's July 2026 review of the FY2026 NDAA.
Frequently Asked Questions
Has an AI-designed drug actually been approved for patients yet?
Not yet. Insilico Medicine's INS018_055 is in Phase II trials and Isomorphic Labs is targeting its first human trial by the end of 2026. Both are meaningful milestones, but full regulatory approval is a separate, later step that hasn't happened for a fully AI-designed drug as of this writing.
How many jobs has AI actually eliminated in 2026?
Trackers differ by methodology, but ResumePulse counted roughly 205,000 AI-attributed layoffs in the US through August 2026, already matching the full 2025 total. Separately, entry-level professional job postings have dropped 29 percent since January 2024. Some of this reflects AI directly replacing tasks; some reflects companies restructuring and citing AI as the reason.
How much money has deepfake fraud actually cost people?
The most conservative, audited figure comes from the FBI's IC3 report: $893.35 million in AI-related fraud losses across 22,364 complaints in 2025. Other private-sector estimates that include unreported or indirect losses run several billion dollars higher, but those aggregate multiple data sources with different definitions of "AI-related."
Is AI actually making data centers less sustainable?
The IEA projects global data center electricity demand will nearly double by 2030, driven mostly by AI. Water use is rising alongside it, though exact comparisons between companies are difficult because disclosure standards aren't consistent yet. Some hyperscalers have also improved efficiency metrics like water usage effectiveness even as total consumption rises with scale.
Are the "AI took my job" stories overblown?
It depends who you ask. Some economists, including Federal Reserve Bank of Dallas researchers, found wages in AI-exposed jobs haven't uniformly declined through early 2026. Others, including MIT's Andrew McAfee, warn that the damage to entry-level hiring specifically could take years to show up in senior-level talent shortages.