Brian Armstrong on Bitcoin, Anthropic Drops Fable 5 & Mythos 5, NewLimit's $435M Age-Reversal | EP #264

Welcome back to Moonshots, the show for AI and exponential tech. We’ve got Brian Armstrong from Coinbase and NewLimit with us, and we’re diving into Bitcoin, AI agents, government stakes in frontier labs, the OpenAI IPO window, SpaceX’s compute play, and we’ll close later with longevity and epigenetic reprogramming—no politics or doom, just the science, tech, and investing that make this the best time to be alive. Quick vibe check from our creator community: fresh intro music, then straight into the news.

Citi floated a one hundred eighty nine thousand Bitcoin by 2026 and the big banks flipped from skeptics to cheerleaders. Are we near a bottom and what’s your read on Bitcoin right now?

Zoom out and take a long view. AI soaked up risk capital and stablecoins got hot after regulatory clarity, so Bitcoin looked less like the inflation hedge of the moment, but I still see it as internet‑era gold and expect higher prices by 2030; sixty thousand felt like a likely local bottom, though no one can be sure.

If autonomous agents are going to transact, they’ll want programmable money and instant settlement. That world doesn’t run on checking accounts.

Wasn’t Bitcoin supposed to rise when the world shakes—wars, market stress, inflation? Is that still the thesis?

It’s playing out slowly because many still trade it like a volatile tech asset; over time more capital will treat it as a store of value, and that counter‑cyclical behavior should strengthen.

On agents, I expect stablecoin payments to be the default. I’m in the many‑models camp: swarms of specialized AIs coordinating work, payroll, and value transfer at scales that could surpass the human economy.

Did the GPU and energy rush for AI cannibalize crypto?

Energy and cutting‑edge chips are scarce, so yes there’s overlap in the hunt for power and fab output; Bitcoin miners use ASICs so they can’t train LLMs, and Ethereum’s shift to proof‑of‑stake cut energy use dramatically.

Risk capital worldwide is piling into U.S. IPOs and data centers, and that flow can pull money from elsewhere. In places with broken currencies, Bitcoin still fills the gap, but dollar gravity is strong when everyone chases U.S. AI.

Citi’s one hundred eighty nine thousand—did that shock you?

I hadn’t seen that report, but if past cycles rhyme, a turn by fall is plausible; something around one hundred to two hundred thousand by year‑end wouldn’t surprise me, though it’s inherently uncertain.

Bitcoin’s traded share each day far exceeds gold’s, which shows how much easier it is to access and move.

Kudos on agent wallets—without them, most AI agents would be economically sidelined.

Quantum risk keeps coming up. Is Bitcoin really safe, and what’s the plan?

Quantum threatens internet cryptography broadly, so we should get ahead of it; Bitcoin core has a proposal, Ethereum has a roadmap that’s maybe a fifth done, Solana’s working too, and our Coinbase council brings top cryptographers together to push post‑quantum upgrades.

The thorniest debate is what to do with older coins vulnerable to quantum, including Satoshi’s; options range from freezing un‑upgraded coins to preserving absolute property rights or a hybrid appeal route, and Satoshi‑era holdings are roughly five to ten percent and likely lost anyway.

You said the agent economy is live. Are you becoming the payment rails for AI?

That’s the strategy, and usage is exploding—around one hundred million on‑chain transactions and roughly fifty million dollars moved so far. First, let users link LLMs to their Coinbase accounts, then add an agentic advisor in‑app, and finally let AIs open self‑custody wallets instantly to transact on their own.

If regulators let AIs open fiat accounts, would you bank them on legacy rails?

Even if rules changed, the old pipes are slow and expensive; on‑chain stablecoins are near‑instant, ultra‑cheap, and global, so that’s where scale and micro‑payments work.

If an agent scams or gets scammed, who’s liable?

For now it likely traces back to a human or company, but true autonomous agents will force new law; either way, we can cut fraud with on‑chain reputation, a graph of trusted payments akin to PageRank.

Once agents can buy, sell, and negotiate on their own, coordination costs fall again—this is the big unlock.

The U.S. is weighing equity stakes in frontier labs; there’s precedent in other strategic firms. Smart move or dangerous?

In a crisis the urge to invest feels right, but handing the state a stock‑picking role invites politicization and future dump risk; taxes already give government all the revenue power it needs.

Could this fund some flavor of UBI?

If a tiny duopoly grows bigger than the rest of the economy, golden shares or quasi‑nationalization become hard to avoid, and a universal dividend suddenly looks timely.

Nationalization tends to slow progress—remember nuclear regulation’s history—even if AI investment feels urgent.

AI started in the private sector and the government passed on early equity, so I’m less worried about a near‑term regulatory choke unless labs themselves want a brake.

Brian, how do you see it from a forty‑billion‑dollar operator’s chair?

Pay taxes, help shape policy, and sell to government as a customer—we already serve about one hundred forty agencies; capital allocation should stay private.

Should the U.S. have a sovereign wealth fund?

I love the shared‑ownership cohesion but worry about politicized bets; a rules‑based index approach could blunt that, yet execution quality would vary by administration.

Senator Sanders pitched taking half of leading AI equity, while others talk about a ten percent public slice. Where does that land?

Fifty percent is a non‑starter; some proposals function like a tax in shares, not an investment, and they risk tilting the field toward chosen giants.

A one‑time ten percent donation at IPO could politically inoculate a lab and seed a public fund, then later be rebalanced into a broad index to support dividends.

OpenAI filed to go public; SpaceX and Anthropic are lining up too. Is it smart to go third in a liquidity crunch?

You can’t miss the window—visibility is months not years now, and waiting for the market to recharge could take forever while rivals lap you.

OpenAI tightened its story by shifting to code, trimming expensive bets, and leasing compute instead of owning mega‑data centers; that makes a steadier prospectus.

Healthy tension between a founder and CFO is a feature; go public even if it’s choppy, but beware retail pain as open source catches up fast and model costs crater beyond Moore’s law.

Google will pay SpaceX AI about eleven billion dollars a year for one hundred ten thousand NVIDIA GPUs—compute is scarce. Does that pull Elon out of the frontier‑model race?

Short term he looks like a hyperscaler, but that revenue can bankroll a return to the frontier once he amasses overwhelming compute; many customers also want CUDA, not TPUs, so renting GPUs makes sense now.

SpaceX unveiled the AI‑1 satellite—huge solar, massive radiators, designed as a Dyson Swarm node. Reaction?

It’s gorgeous and mostly power and heat rejection, which shows how bulky first‑gen space compute is; breakthroughs in power and thermal tech could shrink these radically in the 2030s.

They’ll need redundancy against micrometeoroids and clever origami to fit inside Starship, which is the enabler here.

SpaceX is building a giga‑sat factory in Texas to vertically integrate everything from wafers to finished satellites.

Vertical integration also sets up off‑Earth manufacturing; build on the Moon and rail‑gun units into orbit with far better launch economics.

Texas keeps winning these projects and the economic lift is massive; more states should compete this hard.

Morgan Stanley sketches trillions in SpaceX revenue by 2040 and hints at a first hundred‑trillion company. Too wild?

If a Dyson Swarm powers a big slice of civilization, multi‑trillion revenue is conceivable in that world.

In that space‑compute era, what money do AIs use?

Bitcoin as the reserve asset and on‑chain rails for payments, credit, and markets; even interplanetary transfers can clear with latency and laser links.

Quick pivot to longevity: congrats on NewLimit’s four hundred thirty five million raise. What’s the origin and what’s working?

We’re focused on partial reprogramming to restore cell function without changing identity, using AI to search an enormous space, high‑throughput pooled screens to find hits, then functional assays, and now our first drug candidates head to human trials next year.

Life Biosciences just dosed an OSK patient, and LEV is on the table—what’s your timeline?

I haven’t picked a year; we’re heads‑down shipping biology and let the data drive us, but I’m optimistic about the decade.

We may reach LEV in spikes within subpopulations, with GLP‑1s already showing double‑digit clock shifts in early studies; it might sneak past like the Turing test, noticed only in hindsight.

Clocks are noisy, so I care most about function—liver, cognition, immunity, muscle—and GLP‑1s look like early longevity drugs.

I agree it could arrive with little fanfare and the FDA’s pacing will matter; I’ll fly for treatment when it’s safe and real.

United Therapeutics is working on thymus restoration; is that on your map?

Not yet, but we’ll eventually hit all major tissues, and muscle rejuvenation pairs nicely with weight‑loss agents.

What does four hundred thirty five million actually buy in biotech?

People, materials, automation, and expensive clinical trials—phase ones can run in the tens of millions—so we now have enough shots on goal.

I want to back someone who uses advanced compute to digitally resurrect everyone who has died; it’s time to try.

Last one for now: Columbia edited cholesterol and hemoglobin genes in embryos in vitro. When this is safe and reliable, do we start upgrading our kids’ biology the way we already upgrade everything else?

This goes far past the Gattaca idea of picking the best embryo; base editing lets you swap a single DNA letter with very low error, and I don’t see why it wouldn’t become common.

If it erases inherited disease or even boosts height, norms will shift; where do you land on that?

We have been selecting traits through breeding for ages, and now biology is becoming programmable; disease prevention is compelling, but the governance challenge will be massive.

Local bans will not stick because people can travel to permissive places and bring children home.

This is coming and likely net positive; polls show big support for disease prevention and little for enhancement, but that line blurs fast, like stronger bones to avoid osteoporosis and eventually even cognitive traits.

Remember when IVF was taboo and now it is routine.

Designing a seven-foot kid for economic upside shows why firm guardrails are needed.

Companies already let you fertilize a batch, sequence them, and pick within your gene pool today; base editing is the step beyond that.

With editing, you are no longer limited to the parents’ best roll of the dice.

Brian, thanks for making the time while running a giant crypto company.

Appreciate it; I love talking with people who want to build the future.

On to Anthropic’s new models, Fable 5 and Mythos 5, same core system with extra safeguards in Fable and big promises on testing, code, and research.

Anthropic looks ahead again, with Fable as the guarded twin of Mythos; it one‑shotted my cyberpunk FPS benchmark and shows strong visual game play wins, which hints at heavy long‑horizon training with verifiable rewards.

I have been on it since launch and it is stunning and costly; the output is so dense it races ahead while you are still reading, and it felt like Anthropic waited for OpenAI, then vaulted past.

We have said they have been holding back.

They just split safety guardrails from performance guardrails in public, which says a lot.

The names make it clear, with Fable refusing cyber or nuclear while Mythos is less inhibited.

Fable High even drops to an older model if a biology or chemistry query looks remotely risky.

I hit refusals trying to reach remote agents from a plane and had to work around it.

If IPOs approach, expect faster leapfrogging to win the news cycle.

At least we have real rivalry instead of one lab running the table.

Final story: Apple is remaking Siri as an agent with personal context and is partnering with Google’s Gemini, which is Apple renting brains and admitting the real moat lives in your data.

Glass half empty, Apple outsourced a core layer; glass half full, foundation models behave like search and can be localized by region, and fast cost declines mean today’s cloud‑heavy features could soon distill onto iPhones.

Android has nailed personal context for years, and I have been tempted to switch.

If Apple stalls, it could lose the interface to Google and see TSMC capacity flow to higher‑paying AI chips, while Samsung’s stack and Android pull users away.

On Siri, I will believe it when I see it; we met the founders long ago and the promise sat idle for too long.

That is a wrap; great having Brian from Bitcoin to longevity, and thanks for being with us as we bring the news and keep the optimism high—it is an extraordinary time, so during the singularity stay awake and do not miss a beat.

Live long and thrive.

Thanks, moonshot mates—take care, travel safe, and see you next time.