Andreessen & Altman’s Story About Deflation Is Dangerously Incomplete
I was reading an article about Sam Altman last week and then listening to the a16z podcast and I realized I was hearing the same story from two different angles but with the same convenient omissions. Altman told a Morgan Stanley conference in March 2025 that AI will have a deflationary impact on the global economy and that investors don't appreciate how significant this will be. Then today (January 29th) Marc Andreessen went on the a16z show and laid out what he called the "straightforward extrapolation of very basic economics" that explains why AI will make everyone richer.
The prices of goods and services will collapse, Andreessen said. The thing that costs a hundred dollars today will cost ten dollars and then one dollar. Healthcare will get cheaper. Housing will get cheaper. Education will get cheaper. "There is no scenario in which everybody's just poor," he insisted. "In fact, it's quite the opposite, which is everybody gets a lot richer because prices collapse."
It is a beautiful vision. It is also a vision being articulated by two men who have billions of dollars riding on whether you believe it and who are now in position to write the policies that will determine whether their predictions come true.
Andreessen's firm just raised over $15 billion in new funds in January 2026. That brings Andreessen Horowitz to more than $90 billion in assets under management, neck and neck with Sequoia as the largest venture capital firm in the world. They raised over 18 percent of all venture capital dollars allocated in the United States in 2025. Their portfolio includes investments in OpenAI going back years, in Elon Musk's xAI, and in dozens of other AI companies across healthcare, legal tech, financial services, and infrastructure. In July 2025 they led the largest seed round in history for AI startup Thinking Machines Lab, a $2 billion investment valuing the company at $12 billion before it had a product.
But here is where it gets interesting. Andreessen has been spending what he himself describes as roughly half his time since the election advising the Trump administration. Bloomberg reports he has allies embedded across the government. David Sacks, a fellow venture capitalist with extensive AI and crypto investments, serves as the White House AI and crypto czar. The New York Times investigated Sacks' vast AI holdings and the potential conflicts they create. Sriram Krishnan, who was a general partner at Andreessen Horowitz until joining the administration, serves as senior White House policy advisor for artificial intelligence. Scott Kupor, Andreessen Horowitz's first employee back in 2009, was sworn in as director of the Office of Personnel Management this past summer. According to The Washington Post, Andreessen has been a key networker for talent recruitment at DOGE, Elon Musk's government efficiency operation.
German journalist Claus Kleber made a documentary about this arrangement called "Trump and Silicon Valley: Coup d'état of the Tech Oligarchs." Claus argues that Andreessen's power is only outmatched by that of Peter Thiel. Spanish writer Evgeny Morozov, writing for El País, calls these men "oligarch intellectuals" who have the power to turn their "interpretive gift into legislative mandate." They are not just predicting the future. They are writing the rules that will create it.
Andreessen has been explicit about his motives. He told Joe Rogan in late 2024 that the Biden administration's attitude toward cryptocurrency and AI was why he switched from donating to Democrats to supporting Trump. He described attending White House meetings in May 2024 where Biden officials laid out plans he found alarming. "They said that this is already over," Andreessen recounted. "It's going to be two or three companies and we're just going to control them and that's that." He walked out and endorsed Trump. His firm donated $33.5 million to pro cryptocurrency political groups, more than six times what they gave directly to Trump's campaign. ProPublica reports that eight companies Andreessen Horowitz invested in since 2016 landed in the crosshairs of the Consumer Financial Protection Bureau. The Trump administration has since hollowed out the CFPB.
I do not think Andreessen is lying about deflation. I think he genuinely believes what he is saying. But I also think he is describing a world that will exist for people like him while leaving out what will happen to everyone else. And I think it matters that the person telling you not to worry about AI's economic effects is the person who stands to make tens of billions of dollars if AI deployment accelerates without restraint.
Let me give Andreessen's argument the full hearing it deserves because it is intellectually serious even where I think it is dangerously incomplete.
On today’s podcast he started with history. He said we have been in a regime of very slow technological change for fifty years, running at half the rate of the previous era and a third the rate of a hundred years ago. He said people worried about job loss are being "very reductive" and using "an overly simplistic model." Even if AI triples productivity growth in the economy, which would be "a massively big deal," it would only take us back to the same level of job churn that existed between 1870 and 1930. "And if you go back and you read accounts of 1870 and 1930," he said, "people just thought the world was awash with opportunity."
He then made a demographic argument. Without AI, he said, we would be facing a crisis of depopulation. The economy would shrink. Opportunity would diminish. There would be no new jobs, no new fields, no new source of consumer demand. "You would be very worried about going into a period of severe decline of stagnation," he said. "You'd be very worried about these very dystopian scenarios of an economy kind of self euthanizing itself over time." AI arrives precisely when we need it. "The timing has worked out miraculously well," he said, "in the sense of we're going to have AI and robots precisely when we actually need them to keep the economy from actually shrinking."
Then he described the mechanism of universal prosperity. If AI is as transformative as people think, you get massive productivity growth. That means more output with less input. You substitute AI for human workers. The result is "this massive boom in output with much lower input costs." That creates "gluts of goods and services." Those gluts cause "collapsing prices." The thing that costs a hundred dollars costs ten dollars and then one dollar. "That's the equivalent of giving everybody a giant raise," he said, "because now they have all this additional spending power."
And even if there is unemployment, Andreessen argued, it becomes cheaper to provide a safety net. "The prices of all the goods and services that a welfare program has to pay for, they're all collapsing," he said. "The price of healthcare collapses, the price of housing collapses, the price of education collapses, the price of everything else collapses, because this incredible impact that AI is having. And so in this kind of utopian dystopian scenario that people have, there's no scenario in which everybody's just poor."
He acknowledged that to get mass job loss you would need productivity growth of 10, 20, 30, 50 percent a year, "orders of magnitude higher than we've ever had in any economy in the history of the planet." He thinks that is unlikely. He thinks the more probable outcome is incremental change that is "overwhelmingly going to be a good news process." And even if change is much faster, "it's also going to be a good news process. It'll just be a good news process in the other way that I described."
"Everything I've just described," he concluded, "is just a very straightforward extrapolation of very basic economics. I'm not making any bold predictions."
Here is where I think Andreessen's straightforward extrapolation goes wrong.
Technology does not diffuse evenly. It never has. Everett Rogers, the sociologist who created diffusion of innovations theory in the 1960s, showed that adoption follows a predictable pattern across any population. Innovators make up about 2.5 percent. These are risk takers with resources who adopt first and accept failures. Early Adopters are the next 13.5 percent. They are visionaries who see potential before proof exists and have the means to act on that vision. The Early Majority is 34 percent. These are pragmatists who want to see something work before committing. The Late Majority is another 34 percent. They adopt when not adopting becomes riskier than adopting. Laggards are the final 16 percent. They resist until there is literally no alternative.
Andreessen's economic model treats the world as if everyone adopts simultaneously and benefits equally. That is not how technology works. The Innovators capture first mover advantages. The Early Adopters compound those advantages over years. By the time the Early Majority moves, the infrastructure is owned by others, the talent has been hired by others, the market positions have been established by others. The Late Majority and Laggards do not experience the same transition as the Innovators. They experience a world that has already been reshaped by people who moved first.
This matters enormously for understanding who benefits and who suffers from AI. Let me walk through how this plays out.
Among countries, the Innovators are small wealthy nations with state capacity to mobilize resources at scale. The United Arab Emirates leads global AI adoption at 64 percent of its working age population using AI tools. Singapore follows at 60.9 percent. South Korea surged seven spots in global rankings in 2025 after the government pushed AI into schools, workplaces, and public services. OpenAI opened an office in Seoul because adoption was growing so fast. These nations will experience something like what Andreessen describes. They have the infrastructure, the educated populations, the capital, and the political will to capture AI's benefits broadly.
The Early Adopter countries include the United States, United Kingdom, France, Germany, Japan, and China. These are wealthy nations with significant AI development but uneven deployment. The United States is strange because it leads the world in AI infrastructure and frontier model development but ranks only 24th in actual usage among the working age population. It has the tools. It does not yet have widespread adoption outside of specific industries and demographics. China operates a parallel AI ecosystem. DeepSeek released its model under an open source license and offers a free chatbot, removing financial and technical barriers. It has gained significant traction in China, Russia, Iran, Cuba, and Belarus, countries where Western AI platforms are restricted or too expensive.
The Early Majority countries are watching and waiting. Most of Latin America, much of Eastern Europe, parts of Southeast Asia, and middle income countries around the world fall into this category. They lack the capital to build AI infrastructure domestically. They lack the technical talent to deploy and maintain sophisticated systems. They lack the domestic markets to generate returns on AI investment. Microsoft's AI Economy Institute reports that global adoption reached 16.3 percent by late 2025, up from 15.1 percent earlier in the year. But adoption in the Global North grew almost twice as fast as in the Global South, widening the gap from 9.8 to 10.6 percentage points. Of the ten countries with the largest gains in AI adoption share, all are high income economies.
The Late Majority and Laggards face a harder path. Much of sub Saharan Africa, Central Asia, and parts of South Asia will be adopting AI, if they adopt at all, into a world where the infrastructure advantages have already been captured by others. The UNDP released a report in December 2025 titled "The Next Great Divergence" warning that AI could increase inequality between countries by widening divides in economic performance, capabilities, and governance systems. The IMF warns that growth impacts in advanced economies could be more than double those in low income countries. South Asia alone has nearly 100,000 young people entering the labor market daily, with almost half the region's 1.8 billion population under age 24. That demographic momentum could be a dividend or a disaster depending on whether AI creates opportunities or forecloses them.
Andreessen acknowledged the demographic challenge but framed AI as the solution. He did not address the possibility that the solution might work in some places and not others. He did not address who owns the infrastructure that determines whether a country can participate in AI's benefits. He did not address the geopolitical competition over that infrastructure.
Among companies the pattern is sharper. The Innovators are building the foundation everyone else will depend on. OpenAI, Google DeepMind, Anthropic, Meta AI, Microsoft, Amazon Web Services, and the major cloud providers are investing hundreds of billions in infrastructure. The Stargate Project that Altman announced in January 2025 represents $500 billion in committed AI data center investment through 2029. They have announced sites in Texas, New Mexico, Ohio, Argentina, and Norway. Nearly seven gigawatts of planned capacity. These companies will capture the returns on that infrastructure for decades.
The Early Adopter companies are Fortune 500 firms aggressively deploying AI across operations. This is where Andreessen's deflation story is most accurate. Amazon cut 14,000 managerial roles in October 2025, its largest corporate layoff ever, explicitly to fund AI investment. Beth Galetti, Amazon's HR chief, called AI "the most transformative generation since the Internet" and said the company needs "fewer layers." Salesforce CEO Marc Benioff confirmed in September 2025 that AI had enabled cutting customer support from 9,000 to 5,000 positions. "I need less heads," he said. AI is doing 50 percent of the work. Microsoft cut approximately 15,000 jobs through 2025 while CEO Satya Nadella called for "reimagining" the company's mission as a shift from "software factory to intelligence engine." IBM's CEO Arvind Krishna told the Wall Street Journal that AI chatbots had already replaced hundreds of HR workers.
These companies are experiencing deflation in their labor costs. Their shareholders are getting richer. Their remaining employees, the ones who complement AI rather than compete with it, are getting raises and equity. The workers who lost jobs are not experiencing deflation. They are experiencing unemployment.
Accenture announced 11,000 layoffs in December 2025. CEO Julie Sweet said those who cannot be reskilled "will be exited." Workday cut 1,750 jobs, 8.5 percent of its workforce, in February 2025 because the company needed to "prioritize AI investment." Intel cut 24,000 jobs in July 2025 and now faces a shareholder derivative lawsuit alleging breach of fiduciary duty. Chegg, the education technology company, cut 388 jobs, 45 percent of its entire workforce, because students are using generative AI instead of paying for tutoring. One CEO who went viral in 2025 laid off 80 percent of his staff for refusing to adopt AI fast enough. He said he would do it again.
The Early Majority companies are mid sized firms without the capital to take big AI bets or the technical talent to implement them well. They are watching the Fortune 500 and trying to figure out what works. By the time they figure it out, the Early Adopters will have years of compounded advantages in data, in trained models, in organizational learning, in customer relationships. The same dynamic that made Amazon and Google dominant in their eras will make AI leaders dominant in this one. The gap between leaders and followers will widen, not narrow.
Late Majority and Laggard companies will face a choice between expensive AI adoption that may not work and slow decline as AI enabled competitors undercut them on price and speed. Many will not survive. This is not a theoretical prediction. It is the standard pattern of technological disruption going back centuries.
Among vertical markets, the picture is more complex than Andreessen's model suggests.
Healthcare is furthest along in AI adoption. Andreessen Horowitz has poured money into AI healthcare companies including Ambience Healthcare, Hippocratic AI, Abridge, and others. Their portfolio reflects a thesis that AI will transform clinical documentation, diagnostics, drug discovery, and care delivery. The efficiency gains are real. AI systems can transcribe consultations, suggest diagnoses, flag anomalies in imaging, and automate coding for insurance. But healthcare costs in the United States are not high because delivery is inefficient. They are high because the system is designed to maximize revenue extraction. Hospitals, insurers, pharmaceutical companies, and device manufacturers have pricing power that AI does not eliminate. AI might make delivering care cheaper while the savings flow to shareholders rather than patients. There is no guarantee that efficiency gains become consumer price reductions.
Financial services are similarly advanced. AI is transforming trading, risk assessment, fraud detection, customer service, and back office operations. Goldman Sachs and JPMorgan and the major banks are deploying AI across their businesses. The benefits will flow to institutions with scale and proprietary data. The costs will flow to workers displaced from processing jobs and to consumers who lose human judgment in high stakes financial decisions. When an AI denies your loan application or flags your transaction as fraudulent, there may be no human to appeal to.
Legal services are being disrupted. Harvey, an AI legal assistant, has raised hundreds of millions from Andreessen Horowitz and others. The company targets the research, document review, and contract drafting that junior lawyers traditionally performed. This is entry level work. The traditional path of starting as a junior associate and building skills over years is being disrupted. Law firms may hire fewer associates. Those they hire will need to be productive immediately using AI tools. The ladder is being pulled up for the next generation.
Business process outsourcing faces existential pressure. The IMF estimates 89 percent of Philippine BPO workers face high automation risk. That sector represents 7.4 percent of the country's GDP. India's IT services industry, built on providing software development and support more cheaply than Western workers, faces the same pressure. These are not abstractions. These are millions of families whose livelihoods depend on labor cost arbitrage that AI is eliminating. When Andreessen says prices will collapse, he does not mention that the collapse includes the prices employers are willing to pay for human labor in the Global South.
Let me walk through the stages the economy will actually go through as Altman and Andreessen's vision comes to life. I want to be specific about who benefits and who suffers at each stage.
The first stage is the layoff phase and it is already underway. Challenger, Gray and Christmas reports that over 55,000 U.S. layoffs in 2025 were explicitly AI driven, part of 1.17 million total job cuts, the highest since 2020. White collar unemployment has risen from 3.1 percent to 4.2 percent. Among younger workers aged 22 to 25, employment in high AI exposure roles is already declining. Jobs held by women are nearly twice as exposed to automation as jobs held by men. The layoffs are concentrated in customer service, administrative functions, middle management, content creation, and entry level professional roles.
Within the United States the job losses hit technology, financial services, business services, and retail hardest. The Bay Area, Seattle, New York, and other tech hubs are seeing professional service workers displaced. But these workers have savings, networks, and skills that translate to other roles. They will find new positions, often at lower pay, often after months of searching, but they will land somewhere.
Among countries the layoff phase hits Early Adopter nations first because they have the companies implementing AI at scale. The United States, United Kingdom, and Western Europe will see significant displacement in professional services over the next two to three years. But the layoff phase will eventually hit Late Majority countries hardest because their entire economic model depends on labor cost arbitrage. When AI can do what a call center worker in Manila does or what a coder in Bangalore does or what an accountant in Nairobi does, those workers do not experience deflation. They experience the evaporation of their livelihood with no equivalent opportunity to replace it. There is no safety net. There is no universal basic income. There is no retraining program that makes a displaced BPO worker competitive with an AI system that costs pennies per hour to operate.
The second stage is the productivity boom and this is where Andreessen's model has real merit. Companies that successfully deploy AI will produce more with less. GDP will grow faster than employment. Measured productivity will increase. Corporate profits will rise. Stock markets will climb. The S&P 500 companies that are executing AI strategies will see valuations expand. Their shareholders will benefit. This stage will last for years.
But the productivity boom will not happen everywhere. It requires AI infrastructure, technical talent, capital to invest, and data to train on. The United States, China, and a handful of other wealthy nations will capture most of the gains. PwC estimates AI could contribute $15.7 trillion to the global economy by 2030. But excluding China, only $1.7 trillion of that is expected to flow to the Global South. That is less than eleven percent of the total going to economies that contain more than half the world's population. The productivity boom Andreessen describes is real. It is also geographically concentrated in ways his model does not acknowledge.
The third stage is where the deflation and inflation split becomes visible. Some prices will fall dramatically. Goods and services that can be fully automated will trend toward marginal cost. AI generated content, automated customer service, algorithmic financial advice, basic legal documents, simple diagnostics, language translation, scheduling, data entry. These will become cheap or free. This is real. This will benefit consumers. Andreessen is not wrong about this part.
But other prices will rise. Energy is the obvious example. A single ChatGPT query uses roughly 2.9 watt hours of electricity, nearly ten times a traditional Google search. AI computing racks require 30 to 100 kilowatts each compared to 7 to 10 kilowatts for traditional servers. In Central Ohio, where 130 data centers have clustered, residential electricity bills jumped 60 percent between 2024 and 2025. PJM, the largest U.S. grid operator serving 65 million people across 13 states, saw capacity market prices increase from $28.92 per megawatt in 2024 to $329.17 per megawatt for 2026, more than a tenfold increase driven largely by data center growth. A Carnegie Mellon study projects data centers could increase the average U.S. electricity bill by 8 percent by 2030, exceeding 25 percent in Northern Virginia. The IEA projects AI data centers will consume 90 terawatt hours of electricity annually by 2026 in the base case. That is not deflation for households. That is a regressive tax that hits lower income families hardest.
Water is another constraint no one in Silicon Valley discusses. Data centers require massive quantities for cooling. A typical 100 megawatt facility consumes about 2 million liters daily. Two thirds of new data centers built since 2022 are located in areas already experiencing water stress. In Texas, data centers are projected to consume 49 billion gallons in 2025 and potentially 399 billion gallons by 2030. That is equivalent to drawing Lake Mead down 16 feet in one year. In Arizona, officials have revoked construction permits for homes citing groundwater shortages while simultaneously approving a Google data center that will use the water equivalent of 23,000 residents. Water costs will rise in data center regions. That is not deflation.
Housing will not get cheaper despite Andreessen's claim that prices will collapse. Housing is constrained by land, zoning, construction labor, materials, and local political resistance. AI does not solve any of these constraints. Data centers actually make housing more expensive where they cluster because they attract high paid workers and strain local infrastructure. Northern Virginia's housing costs have risen faster than national averages precisely because of data center employment.
Healthcare will not get cheaper in any meaningful way for American consumers. The United States spends nearly twenty percent of GDP on healthcare not because delivery is inefficient but because the system is designed to extract maximum revenue through market power, regulatory capture, and information asymmetry. AI might make care delivery more efficient while healthcare companies capture those efficiencies as margin expansion. There is nothing in the structure of American healthcare that causes efficiency gains to flow to patients as lower prices. Andreessen's model assumes competitive markets that do not exist.
The fourth stage is wealth concentration. This is where I want to be very specific about who gets richer and who gets poorer because Andreessen's claim that "everybody gets richer" is the weakest part of his argument.
The people who will get substantially richer include owners of AI companies and their early investors. Andreessen Horowitz is positioned to capture billions in returns if OpenAI, xAI, and their portfolio companies succeed. Sam Altman will become one of the wealthiest people on earth if OpenAI achieves its ambitions before it runs out of money. The employees who hold equity in winning AI companies will see life changing wealth. This is perhaps 100,000 people in the United States, maybe 500,000 globally. They will experience exactly the abundance Andreessen describes.
Owners of capital more broadly will get richer. AI increases the return on capital relative to labor. Every dollar invested in AI infrastructure generates returns that previously required human workers. This accelerates a trend that has been running for forty years. Labor's share of national income in the United States has fallen from about 65 percent in 1980 to under 60 percent today. AI will push it lower. People who own stocks, bonds, real estate, and businesses will benefit. People who rely on wages will not benefit proportionally.
Highly skilled workers who complement AI will get richer. Engineers who build AI systems, managers who deploy them effectively, professionals who use AI as leverage for their own expertise. These workers will see higher wages and more opportunities. But this is a small slice of the workforce. McKinsey estimates perhaps 20 to 30 percent of workers will see their productivity enhanced by AI in ways that translate to higher earnings. The rest compete with AI rather than complement it.
Owners of scarce resources will get richer. Land, water rights, mineral rights, spectrum licenses, energy infrastructure. AI increases demand for computation which increases demand for electricity which increases demand for the resources that produce it. Real estate in AI hub cities will appreciate. Energy company stocks will rise. Water rights in the American West will become more valuable.
The people who will get poorer include workers whose skills can be automated. This is not just factory workers and truck drivers. It includes customer service representatives, administrative assistants, paralegals, junior accountants, junior analysts, content writers, graphic designers, translators, and many others. Some will find new roles. Many will face years of declining wages, job insecurity, and periodic unemployment before their occupation disappears entirely. The transition will not be smooth. It will be painful and prolonged.
Workers in the Global South will be hit hardest. Their comparative advantage was cheap labor. AI eliminates that advantage. The call center worker in Manila earning $400 per month cannot compete with an AI system that costs $40 per month and operates 24 hours a day without breaks. There is nowhere for these workers to go. They cannot easily move to the United States or Europe. They cannot easily transition to the jobs AI creates because those jobs require skills and credentials they do not have. They will experience not deflation but immiseration.
Young workers trying to enter the workforce face particular challenges. Entry level jobs are precisely the jobs being automated. The traditional path of starting in a junior role, making mistakes in low stakes environments, and building skills over time is being disrupted. Companies want experienced workers who can use AI productively from day one. They are hiring fewer juniors and expecting more from those they hire. The unemployment rate for recent college graduates has risen. Starting salaries have stagnated. The ladder is being pulled up.
Middle class homeowners outside AI hub regions will see their property values stagnate or decline relative to inflation. The wealth effects of AI concentrate in San Francisco, Seattle, New York, Austin, and a few other cities. People in the industrial Midwest, rural America, and economically declining regions will not participate in the AI boom. They will pay higher electricity bills. They will face labor market pressure as companies automate. They will watch their neighbors' children move away for opportunities that do not exist locally.
The fifth stage is the political reckoning and this is where Andreessen's model breaks down entirely. He treats the economy as a mechanical system that operates according to predictable laws. It is not. It is a human system that responds to political pressure from people who feel left behind.
When enough people experience AI as something that takes rather than gives, they will demand political intervention. This has already begun. Oregon passed legislation requiring data centers to pay for the strain they place on the electrical grid. Texas enacted requirements for large electrical loads including $100,000 screening fees and transmission studies. Politicians from Bernie Sanders on the left to Ron DeSantis on the right have criticized data center expansion. Steve Bannon, Trump's former chief strategist, has emerged as a vocal opponent of the AI acceleration that Andreessen's allies are pushing. Bannon is calling for a pause on AI development until risks are better understood.
The political response will not be rational or efficient. It will be driven by anger, fear, and the desire to punish those perceived as benefiting unfairly. It might include taxes on AI systems, requirements for human workers in certain roles, restrictions on automation in specific industries, tariffs on AI services, privacy regulations that limit data collection, antitrust enforcement against dominant platforms, or outright bans on certain applications. These interventions will reduce the efficiency gains Andreessen predicts. They will also be politically popular with the majority who are not capturing AI's benefits. Democracy gives the losers a vote.
Andreessen has shown he understands this. His firm spent $33.5 million on pro cryptocurrency politics in the 2024 election. They are applying the same playbook to AI, lobbying for federal preemption of state AI laws so that the regulatory framework is set in Washington where they have influence rather than in fifty state capitals where they do not. Rolling Stone reported in December 2025 that the AI industry is using the cryptocurrency model to influence the 2026 midterms. They are playing the political game aggressively. But political games have uncertain outcomes. The backlash they are trying to preempt may come anyway.
The sixth stage is the new equilibrium. History suggests that technological transitions eventually produce broad prosperity but only after painful adjustments and only with active policy intervention. The Industrial Revolution that Andreessen cites as precedent is actually a cautionary tale. Economic historian Robert Allen documented what he calls "Engels' Pause," the period from the mid 18th to mid 19th century when productivity soared while workers' real wages stagnated. Life expectancy in industrial cities fell. Child labor was endemic. Working conditions were brutal. It took generations and massive political struggle before workers began benefiting from industrialization through labor laws, public education, and eventually the welfare state.
We may be entering a new Engels' Pause. Productivity will grow. GDP will rise. Prices of some goods will fall. But wages may stagnate. Economic security may decline. The gap between winners and losers may widen for years or decades before any equilibrium emerges that distributes benefits broadly.
Andreessen is betting that the adjustment will be fast and smooth. That is a convenient belief for someone who will be rich throughout the transition regardless of how painful it is for others.
Let me be specific about what Andreessen is not telling you.
He is not telling you that his firm has $90 billion riding on AI succeeding. He is not telling you that his portfolio companies benefit from every regulatory rollback, every infrastructure investment, every policy that accelerates AI deployment. He is not telling you that his partners and allies now occupy positions throughout the Trump administration where they can write those policies. He is not telling you that the deflationary benefits he describes will flow primarily to capital owners while the costs flow to workers, or that those capital owners include himself and his limited partners, which reportedly include Saudi Arabia's sovereign wealth fund.
He is not telling you that his historical analogy to 1870 through 1930 is selective. That era included the Long Depression of the 1870s, financial panics in 1893 and 1907, the rise of monopolies that required antitrust intervention to break up, and labor conflict that sometimes turned violent. It was not a smooth era of broadly shared prosperity. It was a turbulent era of massive inequality and social conflict that eventually produced political reforms.
He is not telling you that his demographic argument cuts both ways. Yes, declining population growth means fewer workers. But it also means fewer consumers. If workers are displaced faster than population declines, you get unemployment and weak demand despite the demographic shift. The math does not automatically work out in his favor.
He is not telling you that his assumption about the safety net is backwards. He says falling prices will make it cheaper to support displaced workers. But the same political coalition that is accelerating AI deployment is simultaneously cutting social programs. The Trump administration has hollowed out consumer protection, environmental enforcement, and labor standards. There is no evidence that the people pushing fastest for AI deployment are also pushing for stronger safety nets. The opposite is true.
He is not telling you that geopolitical competition could derail his model entirely. The United States and China are racing for AI dominance. The Pentagon has over 800 active AI projects. The Replicator initiative aims to field thousands of autonomous weapons by 2026. All nuclear states are integrating AI into command and control systems. A crisis or conflict could reshape the AI landscape in ways no economic model anticipates.
He is not telling you that he might be wrong. He presents his predictions as "straightforward extrapolation of very basic economics." But the history of technological prediction is littered with confident experts who were wrong. The people who said the internet would democratize information did not anticipate social media disinformation. The people who said smartphones would liberate workers did not anticipate always on work cultures. The people who said globalization would lift all boats did not anticipate the hollowing out of manufacturing regions. Andreessen's confidence is not evidence that he is right.
I do not think Altman and Andreessen are lying. I think they are describing a real phenomenon that will benefit real people. The efficiency gains from AI are genuine. The productivity increases are measurable. The price declines in specific sectors will happen. They are not making things up.
But they are describing the experience of Innovators and Early Adopters. They are describing the experience of capital owners and highly skilled workers in wealthy countries with AI infrastructure. They are describing their own experience and the experience of their investors and the experience of their portfolio companies and the experience of the workers who will thrive in AI enabled roles.
They are not describing the experience of the call center worker in Manila whose job disappears. They are not describing the experience of the young American trying to break into a profession where entry level work is being automated. They are not describing the experience of the homeowner in Ohio whose electricity bill jumped 60 percent. They are not describing the experience of the farmer in Arizona whose water is being diverted to cool data centers. They are not describing the experience of the majority of workers who will face years of uncertainty and downward wage pressure as companies figure out how much human labor they actually need.
When someone tells you AI will make everything cheaper, ask cheaper for whom. When someone tells you AI will create abundance, ask abundance for whom. When someone tells you not to worry about job losses because new jobs will appear, ask where those jobs will be, what skills they require, and who will have access to them.
The people with the most to gain from AI deployment are telling you a story about universal benefit. That story may eventually become true. It has the potential to become true. But the path from here to there runs through layoffs and displacement and rising energy costs and water shortages and political backlash and possibly decades of adjustment during which the benefits concentrate and the costs disperse.
The people telling you the happy story will be rich throughout that adjustment. Marc Andreessen will still have his $90 billion under management. Sam Altman will still have his stake in OpenAI. David Sacks will still have his AI and crypto investments. They are not taking risks. They are asking you to take risks on their behalf, to accept disruption to your life and your community and your economic security so that their investments can compound.
Andreessen says there is no scenario in which everybody is just poor. He is technically right. There are scenarios in which some people become spectacularly wealthy while others become relatively poorer. There are scenarios in which some countries capture AI's benefits while others are permanently left behind. There are scenarios in which some workers thrive while others face decades of declining wages and mounting anxiety. Those are not scenarios of universal poverty. They are scenarios of stratification and divergence and political conflict.
That is the entire story. And that is what Sam Altman and Marc Andreessen are not telling you.