AI Is Getting Cheaper — Fast. Here's the Data.

Every few months I get asked the same question: "Is AI actually getting cheaper, or is that just hype?" The answer is: yes, dramatically, and faster than almost any technology in modern history. Here is the data, plotted two ways.

Chart 1 — How many tokens you get for $1

In 2020, one US dollar bought you about 17,000 tokens (roughly 12,000 words) of the best AI model available (GPT-3). Today, one dollar buys you 800,000 tokens on GPT-5. That is ~48× more AI per dollar in six years.

In 2020, one dollar bought you about 17,000 tokens of the best AI (roughly 12,000 words). Today, one dollar buys 800,000 tokens — ~48× more AI per dollar in 6 years.

Green bars get taller each year because you are getting more output for the same money. Growth charts feel intuitive in a way that price-declining charts do not — so this is usually the version I lead with.

Chart 2 — How much 1 million tokens costs

The flip side of the same coin. A million tokens of the best AI cost $60 in 2020. Today the same workload costs $1.25 — roughly 50× cheaper.

Flip side of the same coin: a million tokens cost $60 in 2020, now costs $1.25 — ~50× cheaper. Faster price decline than computers, electricity, or solar ever managed.

For context, that rate of price decline is faster than:

• Computers (Moore's Law doubling cost-performance every ~2 years → ~8× per 6 years)
• Solar panels (~10× cheaper per decade)
• Electricity, cars, steel, aluminum — pick any major industrial technology, AI is beating it.

The data

Year	Best AI model	Price / 1M tokens	Tokens per $1	vs. 2020
2020	GPT-3	$60.00	~17,000	—
2023	GPT-4	$30.00	~33,000	2×
2024	GPT-4o	$5.00	200,000	12×
2025	GPT-5	$1.25	800,000	48×
2026 (today)	GPT-5 / Claude Opus 4.7	$1.25 – $15	67,000 – 800,000	4× – 48×

Why this matters for builders

The "cost per token" line does not feel revolutionary until you realise what it unlocks. At $60/1M you think twice about letting a user ask a question. At $1.25/1M you stop thinking about cost entirely and start asking "what if I ran 100 AI calls per user action?" — which is exactly how modern AI agents work.

The product patterns of 2026 (autonomous agents, document-heavy pipelines, real-time reasoning loops) were simply uneconomic in 2023. They are routine now because the floor fell out from under the price. And it is still falling.

The Cloud Ate the Robot

Physical Intelligence is two years old. It has not built a robot. It has built a model that controls other people's robots, hosted in the cloud, sending action commands over an API, with no model code running on the robot itself. Co-founder Quan Vang, on Y Combinator's Lightcone podcast last week, mentioned casually that "almost all of the robot evaluations we run at Pi today - including the complicated demos, making coffee, folding laundry, mobile robots navigating around - the model is actually hosted in the cloud. A real cloud. A data center somewhere." This single architectural choice has more implications for the next decade of robotics than any of the cooler demo videos they've shown.

Why this was supposed to be impossible. For twenty years, the first question any robotics customer asked was: what compute unit goes on the robot? It mattered because real-time control loops demand millisecond-level latency, the compute hardware you pick gets obsoleted every 18 months, and it bloats your BOM. The classical answer was "everything on-device," which meant robots were powerful, expensive, heavy computers with arms attached. Pi's answer is a systems-engineering trick called real-time chunking. The robot executes actions in chunks - say 100ms at a time. At the 50ms mark, before the current chunk is done, it requests the next one from the cloud with a continuity constraint so the transition is smooth. Inference happens in parallel with execution. Network latency is buried. The robot doesn't need onboard compute. Vang goes further: he has never seen the robots his model controls, and intentionally avoids knowing how they work internally. The layers are decoupled.

This is the unbundling of robotics. Here is what robotics used to require to build a company: your own customer relationships, your own hardware platform, your own autonomy stack, your own safety certification, your own data collection infrastructure, your own everything. Vertical integration wasn't a choice - it was table stakes, because the intelligence layer didn't exist as a component you could buy. Pi has explicitly externalized the intelligence. They open-sourced the Pi 0 and Pi 0.5 model weights - the same weights they use internally. The result is that a new founder can now walk into an industry with:

• Off-the-shelf robot hardware
• Pi's model handling perception, planning, and control as a cloud API
• A workflow they understand better than anyone else
• Scrappy data collection for their specific deployment

This is the playbook Vang actually walks through in the interview, and it's worth copying verbatim. Starting a vertical robotics company now looks like:

1. Understand an existing workflow deeply. Not conceptually - operationally. Where does labor bottleneck?
2. Identify the single insert point where a robot saves the most cost or unblocks the most capacity.
3. Use cheap hardware. The model is reactive; it compensates for hardware imprecision. You do not need a $100k precision arm.
4. Set up data collection and evaluation in the real deployment - not in a lab demo.
5. Get to mixed autonomy. Humans take over when the robot fails. This is okay. The point isn't perfect autonomy; it's economic break-even.
6. Once you're break-even per robot, scale the fleet. That's when the flywheel spins.

Two YC companies are already running this exact playbook. Weve folds diverse laundry in a real laundromat (not a demo) with clothes it's never seen, while people walk by outside. Ultra packs Amazon-style soft pouches in an actual e-commerce warehouse, running the full workday - same video starts bright outside and ends after sunset. Both built their autonomy on top of Pi's model. Weve reportedly got to a deployable laundry-folding system in two weeks.

What makes this the Cambrian explosion moment. Vang is careful academically but personally confident: he believes thousands of vertical robotics companies are about to exist, one for every workflow that currently has a labor shortage. The reason this is credible and not just vibes is that the startup recipe no longer requires a 20-year robotics PhD. It requires someone scrappy who can do system integration, understand a specific customer workflow, and collect data for that workflow. These are operator skills, not ML-researcher skills. Pi's role in this is not to win every vertical. It's to be the intelligence layer that lets a thousand other companies start. Their success is defined as their model performing useful work on somebody else's robot, in a warehouse they've never seen, for a customer they don't know.

The broader pattern. Every major technology wave gets unbundled at some point. Compute unbundled from hardware into the cloud. Payments unbundled from banks into Stripe. Distribution unbundled from publishers into the app stores. When the intelligence layer of robotics unbundles - and Pi has pretty much just made that happen - the sector moves from a capital-intensive, vertically-integrated, enterprise-only business into something that looks a lot more like the normal startup economy. If you've been waiting for the right moment to build something in the world of atoms, this is the setup Vang is handing you. The hard part is no longer the robotics. The hard part is the workflow, the customer, and the discipline to get to economic break-even before you scale.

Source: The GPT Moment for Robotics is Here - Lightcone, Y Combinator

Your Job Is Not Your Task

Jensen Huang told a story at Stanford last week that should be required listening for anyone planning their career or running a company through the AI transition. It's about radiologists, and it's the cleanest mental model I've heard for thinking about which jobs AI eliminates and which it doesn't.

Ten years ago, one of the most influential computer scientists of his generation - and one of the actual founders of modern AI - told the world that radiology was the worst career a young doctor could pick. AI was about to read scans better than humans within a decade. He was completely right. AI now permeates every aspect of radiology. Almost every scan is assisted by AI. The volume of scans being read has gone through the roof.

The number of radiologists also went up. Not down. Up.

Huang's framing for why is the line worth tattooing on the inside of every founder's eyelid: the purpose of your job and the tasks that you do in your job are related, but they are not the same thing.

The radiologist's task is to read scans. That got automated. The radiologist's purpose is to diagnose disease, work with patients, and partner with doctors. That demand only grew - more patients can be admitted, more conditions caught, more revenue per department, so hospitals hire more radiologists. The flywheel only collapses if the people who confused the task with the purpose start steering young doctors away from the field. Which is exactly what happened. There is now a shortage of radiologists in the United States, caused largely by the warning that the field would die.

This same trap is being set right now in software, design, marketing, sales, and law.

Huang volunteered the second example on himself. "What I do for a living is typing and talking. Both have been automated to superhuman level by AI. And I'm busier than ever." His engineers tell the same story. NVIDIA's coders all use agentic AI. The good ones - the ones being promoted and poached - are the ones who are best at working with the agents. The bottleneck used to be writing the code. Now the bottleneck is having the next idea, because the agents have already finished what you asked them to do and they're "perpetually harassing you in text" asking what's next.

Then Huang says something that explains why the productivity gain doesn't compress headcount the way most pundits assume. Pundits assume NVIDIA needs to ship a fixed amount of code per year - say a billion lines - and if AI lets a thousand engineers do what ten thousand used to, then nine thousand are out. But that's not how it works. A billion lines of code was the most they could do with that many people in that much time. The cap was always human bandwidth, not ambition. Huang wants to write a trillion lines of code. He'd hire more people to write a trillion lines, not fewer to write a billion.

This is the practical version of the same point: task automation doesn't shrink the org if the org's purpose is bottlenecked by ambition rather than by hours. The companies that contract are the ones whose purpose really was just to do the task at fixed throughput. The companies that grow are the ones whose ambition was always being constrained by the throughput, and now isn't.

The single quote founders should put above their desk:
"It is unlikely that most people will lose a job to AI. It is most likely that most people will lose their job to somebody who uses AI."

Two practical reads of that:

1. If you're hiring, the test isn't "have they used AI?" It's "are they faster than the humans who don't?" Treat AI fluency the way you treated Excel fluency in 2002 or English fluency in 1995 - non-negotiable for anyone in a role where the agents are now reachable.

2. If you're working, separate your task from your purpose. Then ruthlessly delegate the task to the agents and reinvest the saved hours into the purpose. The radiologist who learned to use AI now reads more scans, catches more disease, and is the most valuable hire in the department. The radiologist who refused is being told the job is being restructured.

Congressman Ro Khanna's contribution at the same panel sat alongside this and is worth taking seriously: the productivity gains will not be evenly distributed unless someone makes them so. Past industrial revolutions ended with more jobs but spent twenty miserable years getting there. Workers' bargaining position during the adoption phase determines whether the gains end up only with capital. That's a policy question, but it's also a culture-of-the-company question for any founder reading this.

The radiologist parable doesn't close the gap.

Source: U.S. Leadership in AI with Jensen Huang and Congressman Ro Khanna - Stanford GSB

When Latency Becomes Oxygen

Will Bodis runs Phoneley, a voice AI company that just raised a $16M Series A from Bessemer. The company handles millions of calls a month across hundreds of verticals, and 80% of the people on the other end don't know they're talking to a machine. By the end of this year, Bodis predicts that number will be close to 100%.

The interesting part isn't that voice AI works now. It's where Bodis says the bottleneck has moved.

For two years - back when "voice AI" wasn't even a phrase - latency was the thing everyone obsessed over. Phoneley was an early customer of Groq's fast-inference chips precisely because of it. Today Bodis calls latency "oxygen": you need it, but nobody talks about it anymore because everyone has enough. Companies that can't deliver low latency just aren't in the conversation.

The new game is statistical optimization of outcomes. That's a different sentence than "build a chatbot that talks to your customers." Phoneley's pitch is that they don't just answer your phone - they continuously surface what's working and what isn't. He gives one concrete example from earlier in the week: a customer changed one question in their voice AI's script and outcomes improved 5%. Phoneley told them which question to change, and could prove the lift statistically.

That's the position Bodis is staking out, and it's worth attention because it generalizes far beyond voice. In any AI-driven product, the layer above the model is usually where the durable business is built.

Bodis started with small businesses - the pest control company, the pizza shop. Not because the economics were great, but because they gave him constant, frequent, brutal product feedback he couldn't have gotten chasing one enterprise deal that took a year to close. Four or five months of that compressed iteration, and his first call center customer paid more than every small business combined. Then he moved upmarket.

Most founders flip this order - they chase the enterprise logo first because the deal size is bigger, and end up shipping based on their own assumptions because they can't get enough at-bats. Bodis's discipline: find the customer who will tell you what's broken every week, even if they can't pay much, and use them as the iteration substrate for the customer who eventually can.

The investor side of the same lesson

Bessemer didn't come from a deck or a banker. Caroline from Bessemer reached out after a LinkedIn post Bodis had written about doing 300-mile ultra-endurance bike races. The post was about commitment and being a founder. The conversation grew from there. A few months later, Bessemer preemptively offered the round - Bodis didn't shop it. He picked the investor the same way he picks employees: would I want to hire this person? If yes, take the offer.

Closing thought

There's a temptation in AI right now to think that the only game is to build the model. Bodis's career so far is an argument for the opposite: the model is becoming a commodity, and the durable position is the thing built on top of the model that knows how to measure, optimize, and improve a specific outcome. Voice AI is barely past the "books on the internet" stage. The companies that figure out the optimization layer for each vertical will own a lot more value than the ones competing on whose voice sounds most human.

Source: This Startup Built AI That 80% of Callers Think Is Human - Phoneley founder Will Bodis

Set Goals That Break Your Brain

Jon McNeill spent years running Tesla under Elon Musk. Five other startups bracketed it on either side. He's now telling the story in a book called The Algorithm, and the cleanest piece of it is also the one most founders get wrong: the goal you set determines how you think.

He puts it simply. Set a 5-7% improvement goal and you'll get 3-5%. Set a 100% goal and you have to reorganize how you think about the problem - because no amount of optimization within the current frame can get you there. Order-of-magnitude goals - 10x, 100x - force a different brain.

The example he gives is the moment Elon walked over and told him to 20x Tesla's digital sales. At the time Tesla had a 64-click checkout flow. McNeill's instinct, like most operators, was to start optimizing the 64. Elon pulled out a Domino's app and counted 10 thumb taps to buy a pizza. That's the new target.

Twenty-x meant the entire build-to-order religion at Tesla had to die. The company offered 360,000 possible car configurations. The data, when McNeill's team actually ran it, showed customers were buying two - call them Standard and Performance. So they killed the 360,000-option menu. Manufacturing simplified. Engineering simplified. The supply chain simplified. The clicks collapsed. None of that is reachable from a 7% goal. All of it is forced by a 20x goal.

This is the part founders flinch at. We pick incremental targets because they feel responsible. McNeill's argument is that incremental targets are the least responsible thing you can do - they lock you into the current architecture and waste a year tweaking it.

But the goal alone doesn't find the answer. Walking the floor does.

The second pillar of his algorithm is the part most founders skip entirely. Before McNeill formally joined Tesla, Elon asked him to look at the topline problem. Promised the street 12,000 cars; sold 3,000. McNeill didn't open dashboards. He went and mystery-shopped eight Tesla stores, used different email addresses for each test drive, and waited.

Zero callbacks. He called the head of sales ops and asked how many test drives in the past 30 days hadn't been called back. The answer was 9,000. That's the quarter. He shipped a fix in two hours - block all new leads to any salesperson who hadn't called their previous test drives back - and called Elon to confess he'd just made a CEO-level decision in a company he didn't yet work for. Elon paused for a long, uncomfortable minute, then said you'll fit in here just fine.

McNeill calls this the most powerful analytics tool a leader has: two eyes and two ears. Sit in the warehouse for eight hours. Stand on the factory line where the inventory is piling up. Watch a real customer try to buy something on your site and feel the friction yourself. He tells the Falcon Wing door story - Elon dragged him to the production line, they stood and watched workers thread bolts blind, and within ten minutes they could see the fix was a jig. No data could have told them that as fast as their eyes did.

The bank executives he met recently illustrated the inverse. He asked the room to raise their hand if they'd used their own bank's consumer app in the last week. No hands. He told them: I've used two of your apps; they're terrible; if you used them you wouldn't survive another day. The dashboards never told them. They never went and looked.

Combine the two.

The shape of McNeill's algorithm, stripped down, is two-step:

1. Set a goal so large that the existing frame can't reach it. The goal is the forcing function on your thinking.
2. Find the answer by going to where the work happens - store floor, warehouse, customer's screen - not by sitting in the dashboard.

Most founders do the opposite. They set incremental goals and analyze them from a distance. The result is a year of well-executed motion that doesn't move anything.

The other rule worth stealing from his career: hire salespeople from the company with the bad product, not the great one. Anyone closing inbound at Apple is an order taker. The salesperson at the Microsoft store two doors down - still beating quota with the worse offering - is the one who can actually sell. The same logic that finds 9,000 ignored test drives applies here: don't be impressed by the surface; find the friction that made the work real.

Source: Ex-Tesla President Reveals Everything Elon Does to Win - My First Million

The One Decision That Cost InVideo a Year

In January 2025, Sanket Shah and his co-founder were sitting in San Francisco, talking through a fork they both already knew the answer to. Their insight - the one that had taken InVideo from $10M to $70M in revenue - was expiring. The market was shifting under them. The honest move was to drop the existing strategy and bet the company on the next insight. They didn't take it.

Twelve months later, on the Z47 Moments podcast, Sanket called that single non-decision the most important thing that happened to InVideo all year. Not the shipping pace. Not the marketing spend. Not the AI roadmap. The decision they didn't make.

That story sits inside a worldview about how companies actually grow that's worth taking seriously - especially right now, when AI is collapsing and re-creating product-market fit on a six-month clock.

Insights, not optimizations.

Sanket's frame is simple: every business has a single load-bearing insight. The product, the architecture, the culture, the people you hire - all of it organizes around that insight. Without one, you don't have a company; you have a feature competing on price.

He distinguishes insight depth. L1 insights are surface-level - "creator economy" was his - and they get you to maybe $10M. To break through, you need an L3 insight that's actually hard to copy. For InVideo, the L3 insight that drove the $10M-to-$70M run was video editing, not just generation. Specific, not generic.

The corollary is the part most founders flinch from: optimization cannot break a ceiling. If your business has hit one, no amount of A/B testing the funnel or shaving CAC is going to move you. Only a drastic re-orientation around a new insight can. And drastic, by definition, means the existing status quo gets blown up.

Insights expire.

This is the part that should keep AI founders awake. Sanket's view of product-market fit is that it's a spectrum, not a binary, and the score moves on you. A 9-out-of-10 PMF today can be a 6-out-of-10 in six months because someone else found the next insight while you were still defending the current one.

In an AI market where new capabilities open up every quarter, a new entrant has a structural advantage: they're building from what's possible today, while incumbents are stuck inside an architecture and a worldview optimized for last year's possible. That's why the January meeting mattered. They could see the expiry coming. They chose to keep shipping inside the old frame instead.

How to read customers without being lied to.

Most of the conversation founders have with customers is wishful theater - leading questions, hypothetical futures, customers offering solutions back. Sanket's discipline is brutal:

• Only ask about the past. What do they actually do today? Never "would you" or "what if."
• Never accept their solution. The solution is the founder's job; the problem is the customer's.
• The framework is from The Mom Test - and he treats it less as advice and more as a hard rule.

This is how he separates signal from noise. Speculation, including the customer's own speculation, is noise. Behavior in the past is signal. Everything else gets filtered out.

When the bet isn't working, stop watching it.

This was the most counterintuitive operating principle in the conversation. When momentum dies, most founders stare harder at the dashboards - refreshing retention, revenue, signups, hoping for a turn. Sanket does the opposite. He stops looking entirely. Two to four people get assigned to keep the current business running. He goes all-in on the next bet.

He paired this with a cash discipline most founders don't have the stomach for: when his views-to-signup ratio drops below his bar, he kills marketing. Not throttles - kills. The cash he conserves becomes the runway for the next bet. "I have one more bet in me. If this one doesn't work, I have one more. I have to make one of them work."

The takeaway.

The pattern is a loop, not a ladder. Find an insight. Build the company around it. Ride it until it expires - and it will expire. Recognize the expiry early. Take the painful pivot before the ceiling becomes a floor. Conserve enough cash to survive the gap. Repeat.

The cost of getting this wrong is rarely a crash. It's a year of hard, well-executed, perfectly-shipped work that didn't matter. Which is exactly what Sanket says happened to InVideo last year - and why he's telling the story now.

Source: The Insight of Differentiation: Sanket Shah's InVideo Journey - Z47 Moments

The Information Mover Is Dead. The Builder Just Got the Keys.

There are more open product manager roles globally right now than there have been in three years. The last time the number was this high was peak COVID. So why is half of your LinkedIn feed full of laid-off PMs?

Because the job split in two and nobody told the loser.

Nikhyl Singhal — ex-Meta, ex-Google, ex-Credit Karma CPO, and the guy who runs the Skip community of 125 sitting heads of product — calls this a "complete renaissance for the product industry." But it's a renaissance with strings. The strings are: if you built your career as an information mover, you are the dinosaur. If you built it as a builder, you just got handed a vault.

Here is the bet he's making out loud: in the next twelve to twenty-four months, large companies will shed thirty thousand and rehire eight thousand. The eight thousand will be AI-first. The thirty thousand will be the people whose entire job was to take a deck from one boss, reframe it for the next boss, run the standup, write the status report, manage the backlog, and drive “alignment” through theatrics. That entire job description has been quietly automated. Nobody put up a sign.

The split. About half the product industry got into product because they liked moving information through organizations. The other half got into it because they liked making things. The first group is being shed. The second group is having the time of their lives — they have more offers than they've ever had, comp is at all-time highs, and the wall between PM, founder, CEO, and even non-product C-suite roles is dissolving. Singhal has fourteen sitting founders inside his community of 125 today. Twelve months ago there was one. He recently watched a senior member interview for a Chief HR Officer job because the company decided they wanted a product builder running HR.

That last detail is the real signal. Forward-leaning companies have started believing the obsolescence skill — the instinct to look at any manual workflow and write software around it — is more valuable than the function it's being applied to. The function is easier to learn than the skill.

The shadow superpower. Here is the cruel part. The people who are best at the old game are the slowest to adopt the new one. Their entire identity says what I do is working. Their employer agrees. Their bonus agrees. So they don't reinvent. Meanwhile the weaker performer — the one who was already struggling — has nothing to defend, opens Claude Code on a Saturday, ships something silly, catches the bug, and a year later is the most valuable person on the floor.

The block isn't intelligence. It isn't even taste. It's time. The mid-career person in their thirties — the one with the most leverage on paper — is in their power years for everyone else too. Aging parents. Small kids. A spouse. The first body aches. A career that demands constant relearning at exactly the moment when no one has a free hour. Singhal's framing for what your daily prioritization actually is: “I am going to equally disappoint everyone.” That is the honest math. And on top of that math, the industry is now asking you to spend your nights “feeding the LLM.” That is why the most predictable casualty of this era is going to be the diversity progress of the last five years. The Bay Area pace tax falls hardest on whoever already has the least slack.

What the new job actually looks like. Less moving information. More judgment. Judgment meaning: when ten ideas can be tested for the cost it used to take to test one, somebody has to decide which of them deserve to ship — what's good for the brand, the system, the long-run product. That somebody used to be in a meeting. They are now in their IDE.

The PMs Singhal sees thriving have one thing in common: they obsoleted a part of their own job. They built a chief-of-staff app. They wrote an agent that does the matching their community needs. They replaced their status reports with software. They got rid of the meetings they hated by automating the work the meetings were supposed to coordinate. That single moment — the first thing they ship that makes them go wait, this works — is the conversion event. Joy is the antidote to burnout, and most product work has been joyless for a decade. Building is joyful. Once a PM crosses that line, the energy comes back, the time appears, and the rest is mechanical.

Three things that are about to be true.

One: brand on your resume is depreciating fast. Six years at a name-brand company that builds the old way is now a liability in the room. Nobody asks “where did you ship?” They ask “what tools, what judgment, what would you build right now?” If your answer is a 2021 answer, you don't get the job.

Two: this is not a thirty-year treadmill. Singhal is firm on this. The next two years are chaos because every operating system of building software is being rewritten in real time. After that, things settle. There will be a new normal. There will be training programs and predictable tracks again. The activation energy is now. The merry-go-round does not spin forever.

Three: PMs are about to invade every other industry that runs on legacy software — which is most of them. The HVAC company a private-equity firm just bought. The school district. The mid-market manufacturer. They are all going to need someone who can walk in, look at how things are done, and obsolete the manual parts with software. The supply of those people right now is roughly equal to the population of the Skip community plus a few thousand others. Demand is the entire economy.

The one decision. If you love building, stay current — relentlessly, even at the cost of every other comfort, for two years. If you don't love building, be honest with yourself early enough to do something about it. The middle path — wait and see, hope it stabilizes, lean on the brand on your resume — is the path that gets shed in the round of thirty thousand.

Smiling exhaustion is the new floor. Exhausted-but-not-smiling is what comes next if you don't move.

Source: Nikhyl Singhal on Lenny’s Podcast — Why half of product managers are in trouble