AI Models, EVs, and Biotech: What's Actually Moving Technology Right Now

The State of AI: Smaller, Cheaper, and Everywhere

Artificial intelligence in 2025 looks very different from the frontier demonstrations of 2023 and 2024. The narrative has shifted from "who has the biggest model" to "who can deploy a capable model cheaply and reliably." That shift is reshaping the entire AI stack, from hardware demands to deployment strategies to the business models of the labs that build these systems.

Open-Source Models Close the Gap

Meta's LLaMA family has continued its march toward broader capability. LLaMA 4 introduced mixture-of-experts (MoE) architecture at a scale that made it competitive with many closed-source models on coding, reasoning, and multilingual tasks. Several community fine-tunes—particularly from platforms relying on LLaMA base weights—have beaten larger pure-dense models on specific benchmarks, showing that architecture and data curation now matter more than raw parameter count.

Mistral has also been active, releasing models that aggressively target on-device and edge deployments. The company's small-footprint models—ones that run on consumer laptops or even smartphones—are becoming strong enough to replace cloud calls for many enterprise use cases. That matters because it means AI inference is no longer a centralized cloud bill; it's becoming a device-side capability with privacy and latency benefits.

Multimodal Becomes the Default

Models that handle text, images, audio, and structured data in one forward pass are now the standard. GPT-4o, Claude 4, and Gemini 2.x can process and generate across modalities without switching context representation. This isn't just a feature; it's the foundation for agents that can read a screenshot, click the right button, hear a user's voice, and carry a task to completion across multiple apps.

At the same time, video generation models are maturing. Way beyond two-second novelty clips, providers are now generating multi-minute coherent video with consistent characters and scene logic. The commercial applications in advertising, prototyping, and media production are already significant.

The Provider Landscape Consolidates

OpenAI remains the default choice for enterprises, largely because of the maturity of its API, tooling ecosystem, and compliance posture. Anthropic has carved out a strong position for long-context, code-heavy, and agentic workflows, with Claude 4 models excelling at extended reasoning chains. Google DeepMind has pushed model-first product integration through Gemini inside Workspace and Android, making AI a background capability rather than an explicit feature. Google has also released several open-weight models, including Gemma 2 and Gemma 3, contributing to the open ecosystem while keeping its most capable frontier research internal.

Microsoft's Copilot has evolved from a product wrapper to a platform strategy, embedding AI across Windows, Office 365, GitHub, and Azure. The company is clearly betting on being the operating system layer for AI-assisted productivity, with mixed results so far but enormous distribution advantage.

Finally, xAI's Grok continues to chase differentiation through integration with the X platform and a more unfiltered style, while Chinese providers—Alibaba, Baidu, and Zhipu AI—have shipped competitive open models and are investing aggressively in domestic infrastructure to reduce dependency on Western chips.

The EV and Autonomous Driving Landscape

The electric vehicle market in 2025 is no longer a niche enthusiasm; it's a mainstream competitive landscape with margins under pressure, technology rapidly improving, and autonomous driving progressing from demos to licensed operations in multiple geographies.

Tesla and the Full Self-Driving Debate

Tesla continues to iterate its Full Self-Driving (Supervised) system, with version updates pushing the boundaries of what a camera-only stack can achieve on public roads. The approach remains controversial—relying on video-based perception rather than lidar—but the system has reportedly reached a point where supervised drives in favorable conditions are routine for many users in North America.

More importantly, Tesla is building its xAI partnership deeper, incorporating more sophisticated neural architectures into the vehicle stack. The company is also pushing its Optimus humanoid robot program, using manufacturing-scale production to drive costs down. The bet is clear: Tesla sees the car as one node in a broader autonomous-intelligence platform, not the final product.

BYD and the Global EV Price War

Chinese manufacturer BYD has overtaken Tesla in global EV sales and is applying pressure across every price segment. The company's vertically integrated model—batteries, semiconductors, vehicle assembly all under one roof—gives it cost advantages that Western OEMs struggle to match. BYD's vehicles are now competitive not just on price but on build quality, range, and feature sets, challenging assumptions about Chinese manufacturing quality.

The price war BYD helped trigger is squeezing margins industry-wide. Legacy automakers like Volkswagen, Ford, and GM are accelerating EV transitions while absorbing losses on electric models to maintain market position. The next 18 to 24 months will determine which legacy manufacturers survive the transition without government bailouts or dramatic restructuring.

Level 4 Autonomy in Limited Geographies

Several companies have moved beyond robotaxi pilots to licensed commercial operations. Waymo operates in multiple U.S. cities, Baidu's Apollo Go runs large fleets in Chinese cities, and Cruise—despite its 2023 setbacks—is working to regain licenses in controlled conditions. The pattern is becoming clear: Level 4 autonomy works in mapped, well-maintained geofenced areas with good weather and clear road markings, but generalizing to unstructured environments remains a hard problem.

Charging Infrastructure and Interoperability

The North American Charging Standard (NACS) adoption by most legacy OEMs has simplified one major barrier: connector compatibility. Combined with EU mandates for standardized charging ports and the continued expansion of Tesla's Supercharger network, EV ownership is becoming logistically simpler. The remaining gaps are rural coverage, charger uptime, and the pace of high-power fast-charging installations along major transit corridors.

Biotech's Quiet Revolution: Gene Editing, mRNA, and AI Drug Discovery

While artificial intelligence and electric vehicles grab most headlines, biotechnology is undergoing its own profound transformation. Gene editing tools are entering clinical practice, AI is compressing drug discovery timelines, and the line between software and biology is blurring in ways that will remake healthcare over the next decade.

CRISPR Goes Clinical

CRISPR-Cas9 and its successor base-editing tools have moved from promising research to approved treatments. The first CRISPR-based drugs for sickle cell disease and beta thalassemia have received regulatory approval in multiple jurisdictions, with some patients achieving what was previously impossible: a functional cure rather than lifelong management. Base editing—a more precise variant that doesn't cut DNA but swaps one base for another—is advancing through trials for conditions including high cholesterol and certain inherited blindnesses.

Economics remains a challenge; current treatments cost millions per patient. But the trajectory of gene therapies follows a familiar pattern: ultra-rare diseases first, then broader indications, then cost reductions through manufacturing scale and platform approaches. The next wave of CRISPR therapies targeting more common conditions is already in early trials.

AI in Drug Discovery

Machine learning models trained on protein structures, chemical libraries, and clinical data are accelerating drug discovery in measurable ways. AlphaFold's protein structure predictions have become a standard tool in pharmaceutical research, giving scientists structural information that previously required months of laboratory work. Newer AI models are designing novel molecules with desired properties from scratch, predicting binding affinities, and identifying promising drug candidates in days instead of the months or years typical of traditional approaches.

Several AI-discovered molecules have already entered clinical trials, and at least one AI-designed drug has reached Phase 2 testing. The ROI case is compelling: if AI can reliably reduce the failure rate of early-stage drug candidates, it will pay for itself many times over at the scale the pharmaceutical industry operates.

mRNA Beyond Vaccines

The success of mRNA COVID-19 vaccines created a platform technology that researchers are now applying to cancer, autoimmune diseases, and rare genetic conditions. Personalized cancer vaccines—tailored mRNA treatments trained on an individual patient's tumor mutations—are in late-stage trials with promising early results. The technology could transform oncology from a game of chemistry to a game of information: sequencing a tumor, designing the right immune response, and manufacturing a custom vaccine in weeks.

Longevity and the Biology of Aging

Research into the biology of aging—senescent cell clearance, mitochondrial function, epigenetic reprogramming—is attracting serious investment and moving toward human trials. Companies developing senolytics (drugs that clear senescent "zombie cells" that drive age-related tissue dysfunction) have shown promise in treating osteoarthritis, pulmonary fibrosis, and kidney disease. The line between "anti-aging" and disease treatment is becoming legitimate scientific territory, even if consumer marketing often gets ahead of the evidence.

Connecting the Dots: Shared Patterns Across These Sectors

Looking at AI, EVs, and biotech together reveals some common dynamics. In all three, the cost of core capability is falling rapidly—whether that's the compute to run a good language model, the battery capacity to drive 300 miles, or the sequencing and synthesis costs to engineer a biological molecule.

Software-defined behavior is the thread connecting them. Cars are becoming platforms that receive feature updates over the air. Biology is becoming a code-like substrate that can be read, edited, and programmed. AI models resemble software components that can be combined, chained, and deployed at scale. In each domain, the organizations that treat their field as an information problem rather than purely a physical one are gaining ground.

What to Watch in the Coming Months

For AI, the critical moves will be around smaller specialized models, agent frameworks that reliably execute multi-step tasks, and regulatory clarity on how frontier models are tested and deployed. For EVs and autonomous driving, the pressure points are legacy OEM transition timelines, battery cost trajectories, and whether Level 4 deployment expands beyond geofenced urban environments. For biotech, the milestones will be clinical trial results for AI-designed drugs and CRISPR therapies for common conditions, plus any regulatory shifts that accelerate or constrain the field.

What unites these sectors is that they're all in the messy middle between science fiction and settled industrial practice. That's exactly where most real progress happens—and where most of the interesting opportunities lie for the next few years.