The Shape of Things: Agentic AI, Self-Driving Cars, and AI-Discovered Drugs

May 2026: The Technology Backlog Is Being Processed

If there is one thing that mid-2026 makes clear, it is that every board-of-directors conversation in Silicon Valley, Detroit, and Cambridge has the same underlying question: What happens if we bet everything on AI? The answer is no longer theoretical. It is measurable in quarterly revenue, enrolments of patients in AI-designed drug trials, and passenger kilometres logged by fleets of vehicles with no human at the wheel. The technology press has cycled between extremes in recent years: hysteria about employment apocalypse and then breathless dismissal of AI as the new metaverse-style bubble. The more useful framing, increasingly, is not either extreme but the precise mechanism of substitution—the speed at which AI is being embedded at the hardware, software, and domain-specific layers of three industries, and what that embedding means for the economics of each one over the next twenty-four months. Over the past twelve weeks alone, companies across three unrelated sectors have made irreversible technological and financial decisions that position AI—in its agentic, embodied, and scientific forms—not as a beta feature but as core infrastructure.

This update covers three of the clearest stories in technology right now, tracked at ground level rather than through venture-capital press releases: the infrastructure boom making AI workloads tangible, the second generation of self-driving cars that crossed the adoption threshold, and the biotech revolution quietly advancing from the AI lab to the patient side.

1. AI Infrastructure: Nvidia’s $81.6 Billion Quarter

On May 20, 2026, Nvidia reported first-quarter fiscal 2027 revenue of $81.6 billion—up 85 percent year-over-year and ahead of every Wall Street consensus estimate. Data center revenue hit a record $75.2 billion in the quarter, up 92 percent from a year earlier. The headline alone is stratospheric; the more consequential story is what Nvidia is naming the era: the buildout of AI factories, the largest infrastructure expansion in human history.

Nvidia positions itself as the only platform that runs in every cloud, powers every frontier and open-source model, and scales everywhere AI is produced—from hyperscale data centers to the edge. Second-quarter guidance for fiscal 2027 is $91 billion, ±2 percent, with Gross margins expected to remain above 74 percent.

The Vera Rubin Platform: Agentic AI Gets Its Own Silicon

The standout product at the earnings was the launch of the Nvidia Vera Rubin platform, including the Vera CPU—described by Nvidia as the world’s first processor purpose-built specifically for agentic AI workloads. This marks an architectural inflection point: prior GPU generations were optimized for training massive models; this generation is being designed around the inference and orchestration demands of agents—autonomous programs that take action, reason through workflows, and operate continuously rather than in single-turn chat sessions.

The Vera family is paired with the BlueField-4 STX platform, optimized for the demanding memory and IO patterns of agentic workloads. Along with the public launch of Nvidia Dynamo 1.0—open-source software boosting generative and agentic inference on Blackwell GPUs by up to seven times—the software stack is closing the gap between lab benchmarks and real-world production inference at scale.

Nvidia debuted NemoClaw for the agent platform, Nvidia OpenShell for privacy-gated autonomous agent deployment, and the open-source Agent Toolkit for building enterprise AI agents. Taken together, these tools mark the moment agentic AI moved from academic interest to platform-grade productization.

Cloud Partnerships and the Silicon Photonics Layer

The earnings also revealed an expanded collaboration with Google Cloud, covering agentic and physical AI with Vera Rubin-powered A5X instances and a preview of Google Gemini models on Google Distributed Cloud running on Nvidia Blackwell and Blackwell Ultra GPUs. Strategic partnerships with Corning, Coherent, and Lumentum on silicon photonics were announced in parallel—a signal that the optical interconnect tier of the datacenter is now an AI story every bit as important as the GPU itself.

Google’s own AI Studio “vibe coding” Android app, which lets developers prompt their way into mobile applications, shipped in the same surge. The AI coding assistant ecosystem continues to expand across Google, GitHub Copilot, Cursor, and others—though seasoned developers continue to note that the rote portions of a creative technology workflow remain deeply human in their friction points.

Trending AI Providers Beyond Nvidia

While Nvidia dominates the hardware story, the model-provider landscape has quietly matured into a genuinely competitive market without a single dominant vendor. OpenAI’s pursuit of the agentic execution layer—rather than just the language surface—now sits alongside Anthropic’s Claude, which remains the developer-favorite for creative and programming-heavy tasks. Meta’s open-source Llama lineage normalises fine-tuning frontier models at organisations without enterprise budgets.

The industry’s honest accounting question is whether the economic model of running foundation models at scale is sustainable absent a direct monetisation path from the end-user conversation. Infrastructure costs are immense and growing; hyperscale capex cycles are pressuring pricing structures; and model differentiation—the ability of one provider’s output to be meaningfully better than a competitor’s in practical use—is narrowing as open-weight frontier models iterate closer to closed-source performance. Providers that can sustain low-margin API pricing while controlling GPU utilisation across their infrastructure will ultimately win; providers whose pricing depends on margin-from-hype rather than margin-from-structure will face a reckoning far more severe than the 2022 downturn. Unlike prior infrastructure booms, this one is demonstrably driven by product demand, not speculative capital alone.

2. AI-Powered Design: Figma’s Native AI Agent

Among the tool launches most worth watching this quarter is Figma’s native AI design agent, announced in May 2026 and largely under the radar of the wider tech press—a surprising given Figma underpins the design workflow of everyone from solo freelancers to Fortune 50 engineering teams.

What makes the Figma agent structurally different from everything that came before is its location: it lives on the canvas itself, inside the left rail, tied directly to each file’s component system, design tokens, and best-practice library. Unlike third-party AI integrations that require exporting, prompting, and re-importing, the Figma agent produces outputs inside the native vector layer—not a raster sample image—because it operates alongside Figma’s own syntax and context model rather than a separately trained image model.

The agent is designed around three production patterns: exploring multiple design directions in parallel, automating repetitive bulk edits across components and tokens, and working through stakeholder feedback without hand-holding every change from a designer. It can iterate three different information architectures simultaneously, batch-rename hundreds of variables for consistency, swap components across many screens at once, and seed frames with realistic domain copy. Figma frames this as a collaborative layer: designers prompt, the agent performs, and they take over for precision work. The broader signal is that design tools, like databases and data-centers, are becoming active participants rather than passive hosts.

Google’s parallel launch of Flow Music’s Android app—an AI-generated music tool that edits, masters, and produces music videos—and CapCut’s Gemini integration complete the picture: AI creative tools are moving from sample-museum to embedded, iterative, context-aware production environments.

3. AI Goes to Work: The 3,000-Person Footnote

One of the least celebrated but most consequential moves of the quarter was Intuit’s layoff of approximately 3,000 employees—roughly 17 percent of the workforce—to fund its AI transformation. CEO Sasan Goodarzi confirmed to Reuters that the cuts would free resources to accelerate AI integration across the company’s products. Note what is being said implicitly: a profitable, publicly traded company is choosing to shrink rather than protect roles because its leadership calculates that AI can perform that work more efficiently over some time horizon.

The Intuit cuts are not a commentary on the technology of AI itself; they are the leading edge of a pattern that will accelerate through 2026 and into 2027. Organisations facing both competitive and investor pressure to demonstrate AI-driven operating leverage will, over time, substitute higher-cost knowledge work rather than solely augment it. AI can summarise legal documents, synthesise competitive intelligence briefs, generate structurally coherent UI components, and automate the routine portions of a financial analyst’s weekly report—faster and more consistently than a salaried professional at any cost point. The industry has not fully answered where human value concentrates once the predictable, rule-and-pattern portion of knowledge work is automated away.

4. Autonomous Vehicles: The Car That Finally Learned to Drive

The San Francisco Signal

For years the autonomous-vehicle story was a graveyard of broken promises and pilot-program announcements that never graduated to full production. Mid-2026 has shifted that dynamic. Waymo’s fully driverless fleet continues to expand in San Francisco under Level 4 accreditation, serving hundreds of thousands of passenger trips per month in the most demanding urban conditions imaginable, navigating dense traffic patterns, construction zones, and pedestrian-intersection governance that no passenger vehicle has managed reliably in previous form factors. Alert response times, disengagement frequency, and passenger complaint rates from Waymo’s fleet are being tracked as leading indicators by regulators and competitive operators alike. Tesla’s Full Self-Driving (FSD) software is running at scale in select markets outside California with consumer oversight and a growing dataset of real-miles. Cruise’s return to passenger service in 2025-2026, following its 2023 safety incident, is the strongest signal yet that the regulatory posture on autonomous passenger vehicles is finally catching up to where the technology is.

The Hardware Breakthrough

The sensor making this wave practical is Lidar—a pulsed-laser system that measures distance robustly under conditions where cameras fail. Lidar used to cost tens of thousands of dollars and was confined to robotaxi fleets. Lidar startup Luminar brought its Iris platform to sub-$1,000 pricing and production-ready timing for consumer vehicles, making mass deployment viable for mainstream OEMs. Volvo was among the first major manufacturers to commit; within a few model-year cycles Lidar-assisted driving will be standard on most premium EVs. The competitive dynamics are worth tracking closely. Tesla’s FSD relies almost entirely on camera-only perception with neural net predictions—an aggressively scaled bet that vision alone is sufficient at the kilometres-and-years scale of consumer training data. The Lidar-first OEM strategy championed by Luminar and followed by Volvo, Mercedes, and a growing cohort of manufacturers takes a different bet: that capacitor-grade distance certainty addressed through laser pulses is a necessary complement to vision at the safety integrity levels required by regulators for hands-off operation. The outcome of this engineering debate—which is ultimately a debate about deployment timeline versus safety margin rather than pure performance—will shape which technology philosophy wins consumer and fleet operator trust in the critical 2026-2028 window.

Nvidia’s DRIVE AGX Orin platform is the dominant compute substrate running the autonomous logic on most advanced ADAS and Level 3/4 vehicles in production today, providing 200 trillion operations per second. The silicon running a 2026 autonomous car is orders of magnitude more capable than the laptops commonly used for creative work in the mid-2010s.

The Remaining Bottlenecks

Autonomous driving still faces hard edge-case constraints: dense urban environments require laser data comprehension at real time, sophisticated intersection-level mapping, and cybersecurity hardening specific to a connected vehicle stack. Regulatory conversations across the US, EU, and China are racing but uneven: San Francisco, Phoenix, and Beijing are aggressive adopters; jurisdictions with different liability frameworks are moving slower. The secular story for 2026 and beyond is whether the economics of EVs, Lidar, and autonomous stacks continue to improve until driver-assisted shared mobility is the primary transport mode in dense metropolitan markets. The data points that way. The total cost of ownership argument for a robotaxi fleet—where a vehicle logging 16-20 hours of daily service amortises high per-unit hardware costs over many thousands of rides, eliminating driver pay and insurance substitution costs from the P&L—is structurally compelling for fleet operators in high-density markets where per-passenger costs are already elevated. The remaining question is whether liability frameworks in individual jurisdictions will adjust quickly enough to allow fleet operators to realise that economics at scale without the burden of insurance premiums reflecting residual human-on-the-loop requirements.

5. Biotech: AI Discovers What the Lab Cannot Yet See

AlphaFold and the Protein Revolution

DeepMind’s AlphaFold—the AI system that predicts protein 3D structure from raw genetic sequence—has graduated from landmark research paper to production-grade infrastructure underpinning drug discovery across multiple pharmaceutical companies, biotech startups, and research consortia. AlphaFold operates in the dimension of opportunity cost: protein crystallography is a multi-year wet-lab process requiring beam time at national synchrotrons and years of experimental refinement; structure prediction with AlphaFold arrives in seconds, eliminating the waiting game that has historically throttled pharmaceutical progress. The 2024 release of AlphaFold 3 further extended the model’s reach to ligand-protein interactions and RNA folding, broadening its utility beyond pure protein structure into the full molecular topology relevant to most drug targets. For computational biologists, the capability shift is comparable to moving from maps to GPS in real time—the difference in velocity of hypothesis generation and in silico experimentation is not applicable in percentage terms alone. For every protein whose structure is now accessible to computational biologists rather than laboratory scientists waiting for diffraction time, the pharmaceutical research pipeline moves materially faster.

AI Drug Discovery at Machine Speed

AI-inferred structure has already translated into a pipeline of AI-designed therapeutics reaching clinical trials faster than could historically be justified. The comparison is worth stating directly: a traditional drug from hit-to-lead to Phase 1 candidate takes, on average, four to six years; biotech startups powered by AI-inferred protein models have crossed this threshold in one to two years without sacrificing compound quality. Partnerships between DeepMind, pharmaceutical houses, and specialized AI biotechs are racing this pace.

One of the most exciting emerging threads is the use of AI to optimise CAR-T receptor design in oncology, improving antigen-specificity and reducing off-target toxicity rates in ways that exceed manually-designed equivalent candidates in early clinical data.

CRISPR Gene Therapy: From Indication to Standard

Separate from the molecule-design story is the CRISPR gene-editing platform, which passed a landmark therapeutic milestone in late 2025 and early 2026. The FDA has now approved CRISPR-based therapies for multiple rare-disease indications, representing an addressable population of roughly 300 million people worldwide—one of the most underserved segments in pharma research history. The regulatory pathway forged through these approvals is likely to set the template for future indications in more prevalent conditions; cardiovascular disease is currently viewed by biotech insiders as the next major frontier for CRISPR utilities.

Where Everything Meets: The Agentic Stack Is No Longer Theoretical

The intersection of these three stories is also a convergence of risk. The massive datacenter infrastructure buildout powering the AI revolution consumes electricity at a scale that directly conflicts with national net-zero climate targets in nearly every jurisdiction in which major datacenters are being sited. Water consumption for AI cooling in northern Virginia, Texas, and Nevada is straining local infrastructure capacity. The convergence of agentic AI, autonomous transport, and AI-driven biotech is also a story about regulatory velocity: data privacy frameworks, autonomous-vehicle liability regimes, and pharmaceutical-approval pathways are all under structural pressure from technologies moving faster than the governance institutions designed to oversee them. These frictions are real and unresolved; they will not disappear because the technology is impressive., and the AI drug discovery pipeline is a single architecture: a new generative AI computation stack—hardware layer, foundation models, agent orchestration, application tooling—that is productively substituting automation for work at scale. Nvidia is building the silicon. Google DeepMind, OpenAI, Anthropic, and Meta are building the foundation models. Figma, Intuit, and a hundred other vendors are building the applications. Autonomous vehicle companies and biotechs are the industries now running production workloads on this stack.

This is not hype. The Nvidia earnings confirm the demand is real. The autonomous miles logged in dense urban environments confirm the technology is working at a level that satisfies most consumer workloads. The proteins predicted in seconds and now advancing through Phase 2 trials confirm the scientific case. What does not yet have a satisfying answer in mid-2026 are the distribution and equity questions: who benefits from AI-enabled productivity when AI-augmented wages compress demand faster than they expand it, and how do societies decide which AI development priorities serve collective interest rather than the narrow margins of a handful of private companies? These are not hypothetical questions—they are already surfacing in city council hearings over AI data centre siting, in regulatory filings over AI deepfake liability under the US Take It Down Act, and in boardrooms everywhere a CFO is recalculating labour costs against a new automation baseline.

What is worth watching with sustained attention is the infrastructure cost—financial and environmental. Datacentres are consuming electricity at rates that conflict with national net-zero commitments. AI regulatory conversations around liability and data right are keeping uneven pace with systems being deployed. The 3,000-person Intuit layoffs are one visible face of a broader economic transition; the structural shift in how knowledge work is compensated and organised is the slower-moving pattern that will define the 2030s.

These tensions are real and unresolved in mid-2026. What makes May 2026 a moment worth documenting is that the industry has crossed a threshold: it is now delivering outcomes at scale rather than only describing future promise.

Looking Forward

What to Watch

Nvidia’s second-quarter guidance of $91 billion will test whether enterprise conversion of pilot AI projects to production infrastructure is sustaining its hyperscale cloud partners and maintaining momentum in the AI Clouds, Industrial, and Enterprise segment. Watch the ACIE contrast: hyperscale and edge have materially different unit economics.

Waymo’s continuous geographic expansion, particularly through Europe and South Korea as regulatory frameworks settle, will provide the second-wave data point for autonomous vehicle commercial viability.

On biotech, the first cohort of AI-designed oncology therapeutics entering Phase 2 during this year will likely accelerate capital flows toward AI-native biotech companies and reorient pharmaceutical investment strategy toward computational-design-first pipelines. The probability-weighted opportunity set for AI-designed drugs in cardiovascular and neurodegenerative indications is particularly high—these are indications where target structure has historically been intractable to crystallography and where AI models are making meaningful structural predictions for the first time. The biotech venture capital cycle is already reacting to this dynamic; AI-native biotech rounds are commanding premium valuations relative to traditional biotech counterparts with comparable preclinical pipelines.

If there is one thread uniting all three of these tracks, it is this: every technology race in 2026 is being accelerated by agentic AI—systems that reason, act, scale, and instrument reality rather than merely narrate it. The organisations that understand how to align the agent with the domain—whether a design canvas, a vehicle control stack, or a protein structure—are the ones coming away with durable competitive advantage. The difference between a generic prompt-response and a domain-specific agentic system is, in every currently measurable metric, economically transformative. Engineering teams that embed AI in IDE and code-review workflows report faster iteration loops and lower review error rates. Autonomous-vehicle operators that log more miles accumulate more training data, compounding the advantage with every passenger journey. Biotechs that deploy AI-designed molecules in Phase 1 trials learn faster on every patient cohort. The combined velocity effect across all three tracks is the central logic underpinning the $81.6 billion quarter and the autonomous miles now being logged daily.