|
|
From the resurgence of nuclear energy driven by AI’s insatiable demand for power, to the next phase of autonomous defense systems, and the evolution of jobs bridging hardware and software, these are the bold ideas shaping the future.
We asked the a16z partners to share a preview of what’s next across energy, AI, space, biotech, crypto, and more — highlighting the innovations and opportunities poised to drive change in the year ahead.
|
|
|
|
|
David Ulevitch, General Partner, American Dynamism |
@davidu
|
| |
In 2025, I believe we’ll see a surge in demand for nuclear energy. A perfect storm of regulatory reform, public enthusiasm, capital infusions, and insatiable energy needs — particularly from AI data centers — will accelerate orders for new reactors for the first time in decades.
As AI advances, America’s energy demand is skyrocketing. For the first time in decades, electricity consumption is on the rise, rattling our aging grid and reigniting the search for new, reliable power sources. Hyperscale data centers, hungry for clean and consistent energy, are already reviving decommissioned nuclear plants, including Pennsylvania’s once-infamous Three Mile Island, slated to come back online in 2028.
Bipartisan momentum and grassroots support for clean energy have spurred renewed interest in nuclear power. But this is about more than energy: it’s about securing America’s leadership in the global AI race, building a more resilient grid, and future-proofing national prosperity.
|
|
|
|
|
|
|
|
|
Vineeta Agarwala, General Partner, Bio + Health |
@vintweeta
|
| |
In 2025, “big is back” in the biopharma world, as even early stage biotech startups begin to go after big, common diseases again.
What’s driving the shift? GLP-1 drugs for diabetes and obesity are expected to create a $100B+ market by 2030, and have infused new energy into the cardiometabolic disease space. Perhaps more quietly, we are also experiencing a gradual revolution in how we understand and treat many common auto-immune diseases, like lupus and arthritis.
Recently, a German physician in Munich named Dr. Georg Schett hypothesized that engineered CAR-T cell therapy used for certain B-cell cancers might also help patients with B-cell-driven auto-immune diseases — and this year, he published astonishing results. In a cohort of 15 patients for whom no other therapies were offering benefit, every single patient experienced dramatic improvement. As Dr. Schett described it: “The CAR-T therapy is like a reset button on a computer; it basically restarts the system and the immune system works perfectly fine.”
Inspired by powerful results like this, as well as the clinical and commercial success of new obesity medications, we expect a new wave of biotech and startup innovation focused on treating (and potentially even curing) our biggest diseases.
|
|
|
|
|
|
|
|
|
|
|
| |
Imagine being in a band with an AI drummer: every time you switch up the tempo or spontaneously introduce a new riff, the beat changes in unison, matching you note-for-note. It’s seamless and intuitive, meeting your ability, complementing your vibe, and keeping pace better than your usual halting jam sessions. The emergence of real-time AI brings that vision closer to reality.
We saw the first glimmers of near real-time AI with the release of Latent Consistency Models (LCMs) in 2023. As the inference time continues to drop, AI creative tools have become both more productive — yielding more generations per second — and more useful, giving way to new use cases like live video-to-video. In the coming year, exciting new use cases will be unlocked: everything from generated video companions to AI bandmates.
As latency disappears, new possibilities emerge. This technology even has potential in educational settings, giving teachers the ability to shift gears in a lesson or restate a point based on real-time micro-responses, such as students appearing distracted or confused. An immediate feedback loop allows us to prototype, iterate, and refine ideas at breakneck speed. This shift will redefine every creative workflow, where experiencing ideas as they happen unlocks the prospect of true co-creation with machines.
|
|
|
|
|
|
|
|
|
Sam Broner, Investing Partner, Crypto |
@sambroner
|
| |
Stablecoins found product-market fit in the past year — not surprising since they are the cheapest way to send a dollar, enabling fast global payments. Stablecoins also provide more accessible platforms for entrepreneurs building new payments products: no gatekeepers, minimum balances, or proprietary SDKs. But large enterprises have not yet woken up to the substantial cost savings — and new margins — available to them by switching to these payment rails.
While we’re seeing some enterprise interest in stablecoins (and early adoption in peer-to-peer payments), I expect to see a bigger experimentation wave in 2025. Small-/medium-sized businesses with strong brands, captive audiences, and painful payment costs — like restaurants, coffee shops, corner stores — will be the first to switch from credit cards. They don’t benefit from credit card fraud protection (given in-person transactions), and are also the most hurt by transaction fees (30 cents per coffee is a lot of lost margin!).
We should also expect larger enterprises to adopt stablecoins as well. If stablecoins indeed speedrun banking history, then enterprises will attempt to disintermediate payment providers — adding 2% directly to their bottom line. Enterprises will also start seeking new solutions to problems credit card companies currently solve, like fraud protection and identity.
|
|
|
|
|
|
|
|
|
David Haber, General Partner, Fintech |
@dhaber
|
| |
AI has become the ultimate driver of differentiation, transforming software into labor across industries. In 2024, many startups pursued “messy inbox problems” as a wedge to apply LLMs to judgment-intensive tasks. 2025 will be the year of AI companies turning differentiation into lasting defensibility.
The winning startups will focus on building moats around their products. Defensibility still hinges on timeless factors: network effects that grow value with user adoption, high switching costs that make products indispensable, and product virality that drives lower customer acquisition costs.
Successful AI-native companies will transcend narrow use cases, expanding into adjacent workflows and becoming core systems of record. Differentiation — solving a wedge problem 10x (or 100x) better — earns the opportunity to build a moat. But differentiation and defensibility are distinct, and startups that conflate the two risk being outflanked by more strategic competitors.
|
|
|
|
|
|
|
|
|
|
|
| |
Millions of people have downloaded AI companions and interact with them for hours each day. But the experience has limitations: the current generation of companions are passive, only reacting to conversations you initiate. Outside of your interactions, these companions have no friends or external context. In other words, they don’t have inner worlds.
I believe the next generation of companions will become much more engaging and lifelike. They’ll have their own virtual friends, reactions to news, and emotions. They’ll have their own motivations, missions, and desires, and they’ll chat with you about yours. Your friendships will be give-and-take.
The future design of AI companions can learn a lot from what has worked for video game franchises. As with video games, your conversations with companions should have a purpose and be driven by your motivations (akin to “quests” in games, whether you call them that or not). Companions should make reference to other characters, introduce you to friends, and discuss places, topics, and issues in their world. Sometimes they’ll text or call you for a long conversation, other times they’ll simply react. AI companions will feel increasingly real when they, themselves, believe they have a world to live for.
|
|
|
|
|
|
|
|
|
Alex Immerman, Investing Partner, Growth |
@aleximm
|
| |
The search monopoly ends in 2025.
Google controls ~90% of U.S. search, but its grip is slipping. Its recent U.S. antitrust ruling encourages Apple and other phone manufacturers to empower alternative search providers. More than just legal pressure, gen AI is coming for search.
ChatGPT has 250+ million weekly active users. Answer engine Perplexity is gaining share, growing 25%+ month-over-month, and changing the search engagement form; their queries average ~10 words, 3x+ longer than traditional search, and nearly half lead to follow-up questions. Claude, Grok, Meta AI, Poe, and other chatbots are also carving off portions of search. Sixty percent of U.S. consumers used a chatbot to research or decide on a purchase in the past 30 days. For deep work, professionals are leveraging domain-specific providers like Causaly (science), Consensus (academic research), Harvey (law), and Hebbia (financial services).
Ads and links historically aligned with Google’s mission: organize the world’s information and make it universally accessible and useful. But Google has become so cluttered and gamed that users need to dig through the results. Users want answers and depth. Google itself can offer its own AI results, but at the cost of short-term profits. Google as a verb is under siege. The race is on for its replacement.
|
|
|
|
|
|
|
|
|
Anjney Midha, General Partner, Infrastructure |
@AnjneyMidha
|
| |
In the race for AI dominance, compute has become critical national infrastructure. But not every country is equipped to compete. Training and inference of large-scale AI models takes thousands of power-hungry GPUs — you need ample energy and, by extension, land in places that can effectively dissipate hundreds of megawatts of heat. I call places with the ability to develop, train, and host their own state-of-the-art models AI Hypercenters.
Over the next five to 10 years, I believe that a world-class Hypercenter will need to develop about three to six gigawatts of installed capacity to keep a seat at the table of frontier AI. Though that scale doesn’t exist today, there are multiple countries — the U.S., China, Japan, Singapore, and Saudi Arabia among them — racing to reach that target through AI infrastructure build-outs, 100 to 150 megawatts at a time.
Governments have begun to view AI compute infrastructure as a strategic national resource that is critical to maintaining a competitive edge in AI development. In the coming years, I predict the nations that invest in AI through compute capacity, sustainable energy sources, and forward-thinking policy will dictate the future of scientific and economic progress around the world.
|
|
|
|
|
|
|
|
|
|
|
