The holes in AI’s economic doomsday scenario

September 28, 2026

Welcome back. Qualcomm's latest mobile chip can run a 30-billion-parameter AI model on a phone, roughly seven times the size of the largest phone model we’d previously seen. Sabrina Ortiz explains the clever use of specialized models and memory that makes it possible. But a more capable chip alone won’t make your phone the AI assistant we've been promised for over a decade. In this week’s Deep View Conversations, Qualcomm’s Vinesh Sukumar lays out the other hurdles, from connectivity to battery life. Meanwhile, Anthropic’s economic forecasts raise a different question: Is widespread AI job loss inevitable? Nat Rubio-Licht examines the assumptions behind its bleakest scenario and the choices that could change the outcome. —Jason Hiner

RESEARCH

One of the foremost AI labs predicted three distinct scenarios for AI's future economic impact, and two involve large swathes of the workforce losing out. But what is the reality of that future?

Earlier this month, Anthropic's economics team released research painting a picture of AI's potential modest, substantial, and extreme impact on the economy by 2030. While all three involve increases in the GDP, ranging from a 1.6% increase to a 32.4% increase, the catch is job displacement. The bigger the impact AI has on the economy, the larger the percentage of knowledge workers who are stripped of their jobs, with a large portion unable to find new work in the most extreme scenarios, according to this research.

Additionally, Anthropic predicts that, as the impact of AI grows more severe, despite the growth in the GDP, the distribution of wealth is uneven, with more money made going back to capital than it does to workers' wages. As it stands, 60 cents of every dollar made goes back to the worker, and 40 cents to capital. AI could flip these figures in the most extreme scenarios, stagnating wages and worsening unemployment. That's the picture Anthropic paints in this report.

Here are the three scenarios:

Modest scenario: Anthropic says that AI has roughly the same impact on the economy as the internet, driving significant gains, though within historical norms. The GDP reaches $34.1 trillion, up 1.6%, by 2030, and though 0.3% of knowledge workers are displaced, all of those workers are able to be reallocated into new roles.
Substantial scenario: AI has roughly the same impact as the railroad, causing an 8.3% rise in the GDP to $36.3 trillion. Around 2.5% of knowledge workers are displaced, 1.8% of which are able to find new roles by 2030.
Extreme scenario: GDP sees a 32.4% increase by 2030 to $44.4 trillion as a result of unprecedented growth, likely led by the development of recursive self-improvement. 13.5% of knowledge workers are displaced, and only 5.2% are able to find new work, leaving 8.3% unemployed.

However, there may be a few hitches in the more extreme scenarios that Anthropic laid out, Julius Probst, senior economist at recruitment marketing firm Appcast, told The Deep View. For one, these scenarios assume that all of the GDP value that AI is generating will be consumed. But in the most extreme scenario, if we are heading towards a labor market facing wide-scale unemployment and a great deal of concentrated wealth, "that is probably a bad assumption," said Probst. Many consumers would not be able to afford to buy things, meaning that the GDP won't actually surge.

"Wealth inequality will soar, but these people will not buy additional cars or additional houses," said Probst. "There's only so many more houses a billionaire can have."

The other hitch is that the model largely lumps together all knowledge workers into one bucket. The reality is that, while knowledge workers will be broadly impacted, many companies are still seeking senior, more skilled workers whose roles can be complemented by AI, said Probst. But even that won't be sustainable for long, he said, as many of those senior workers will retire and companies will realize they need to invest in junior talent again. "I don't think this situation can persist for another three to five years. At some point in time, companies will realize we need to hire junior people again."

And given that a large portion of the labor force is involved in physical work that can't be supplanted by AI, of the three scenarios Anthropic painted, the most likely is the modest one, he said. "AI growth is really showing up in two sectors only, and that is the tech sector and professional business services."

There has long been a narrative being pushed by AI stakeholders that the tech is inevitable, and that every worker needs to get on board or be left behind. Anthropic's economic scenarios, especially the most extreme, support that narrative. If you are not able to keep up with AI skills, you could end up one of the 8.3% of knowledge workers that Anthropic predicts will be unable to cross over into a new job by 2030. But there are important caveats to remember in this forecast. The first is that the companies parroting the idea that AI is going to upend our economy and every worker needs to hop on board has a clear incentive to get as many people to adopt its technology as possible. And the second is that the future is not decided. Anthropic, to its credit, notes this in its research. However, in order to prevent the worst outcomes of AI, we need to prepare our economy and workforce for AI now, and stop believing the narrative that AI transformation, and the havoc it can wreak, are inevitable.

TOGETHER WITH DESCOPE

Every B2B company is shipping an MCP server, but how many can answer the question that comes up in enterprise security review: can we manage agent access with our own identity provider?

Cross-App Access (XAA) is the open standard that answers it, and Descope provides the identity infrastructure so you don’t have to build it yourselves:

Validate tokens from your customers' IdPs, so their admins decide which agents reach your MCP server
Issue short-lived tokens so your own agents reach any MCP server
Let tenant admins set up XAA for their agents in a self-service portal

Support SSO-like experiences for your AI agents and MCP servers.

HARDWARE

During Qualcomm's Snapdragon Summit, one number made me snap my head up: support for 30-billion-parameter Mixture-of-Experts (MoE) AI models running locally on mobile devices.

Qualcomm claims its most advanced chipset yet, the Snapdragon 8 Elite Extreme Gen 6, can pull this off. For context, the largest on-device model I'd seen on a phone was around 4 billion parameters, running on top-tier chips like Apple's A20 Pro. To see how a more than sevenfold jump is possible, it helps to understand how an MoE model works, because it's different from a traditional, or dense, model.

"Very smart AI architects here have moved to this mixture of expert models, so you can get something that's sort of equivalent in kind of capability from a single dense model into actually what's composed of multiple small models," Chris Patrick, SVP and General Manager of Handsets at Qualcomm, told The Deep View.

Qualcomm goes a step further by storing some of those experts in flash storage rather than keeping them all in RAM at once. As Patrick put it, "the memory furniture on a phone" isn't big enough to hold 30 billion parameters. Instead, the system predicts which expert it will need for the next token or action, loads it into memory, runs it, and then swaps it out for the next one.

This isn't to say dense models are obsolete. Smaller 4-billion-parameter models are already capable of a lot, from answering questions and summarizing documents to understanding images and carrying out tasks within apps, and their speed and efficiency make them ideal for everyday requests. But when a task calls for more advanced reasoning, users can now tap into far larger models than a phone's hardware would otherwise allow.

"In the end, it is trying to approximate what you'd have for a single big dense model that might require 32GB of RAM on the phone," added Patrick. (Today's top-end phones typically have 12GB to 16GB of RAM.)

So can any model be run in the mixture of expert architecture? Not quite. The model itself has to be built in that specific format, but the good news is that MoE now underpins many of today's most capable open models, from DeepSeek's V4 series and Moonshot AI's Kimi K2.6 to Alibaba's Qwen3.6-35B-A3B, which, at 35 billion total parameters with only about 3 billion active at a time, is close to the scale Qualcomm says its new chip can handle. Google has also explicitly disclosed its use of MoE architectures, and it's worth noting that proprietary frontier models from OpenAI and Anthropic are widely believed to use MoE-style architectures, but they generally don't disclose architectural details.

Raw intelligence is no longer the bottleneck. Models keep getting more capable as more is demanded of them, especially with the rise of agentic AI. The real challenge, as noted above, is running these models on the devices we carry every day, like smartphones, smartwatches, and eventually smart glasses, which are the devices that will truly bring the vision of personal AI to life. That's why the industry is pursuing so many ways to make models not just smaller but more efficient, from model routing, which has surged in popularity, to domain-specific models built for particular tasks. MoE feels like a natural evolution of both: a single model made up of specialized experts, activated only when needed, that together deliver the capability of a much larger model.

Disclaimer: Sabrina Ortiz's travel to Snapdragon Summit was paid for by Qualcomm. The Deep View's coverage is editorially independent from the companies we cover.

TOGETHER WITH BOX

Enterprise AI begins with your content. As the file system for AI, Box securely connects unstructured data to the models, agents, and applications your teams use every day—automating workflows, removing manual bottlenecks, and turning content into meaningful, repeatable business outcomes.

Join Box's Aaron Levie, NVIDIA's Jensen Huang, and Intel's Lip-Bu Tan at BoxWorks 2026 in San Francisco to experience how secure enterprise AI can turn business content into smarter workflows, deeper insights, and faster decisions.

Use code DeepView50 for 50% off your registration.

Claim 50% off →

CONSUMER

Nearly every smartphone launched in the past year features agentic AI capabilities, offering users an early look at what a fully agentic smartphone future could do for them. Of course, the tech powering it is driven by the chipsets.

In this episode of The Deep View Conversations, we talked with Vinesh Sukumar, Qualcomm's VP of AI at the Snapdragon Summit, the company's annual conference where it launches its latest processors. This year, the launch included the mobile platforms Snapdragon 8 Elite Extreme Gen 6 for phones and Snapdragon Sound Elite Gen 2 for wearables.

Vinesh discussed how the chipsets came to be, including the special considerations made during their design such as improving connectivity, on-device support for large models, longer battery life, and other features crucial to smoothly running agentic AI applications. We also discussed what the future of a truly agentic AI phone looks like and what's been holding it back.

Topics covered:

• What an ideal agentic smartphone experience would look like

• The demands agentic AI models make of mobile chipsets

• The obstacles to agentic solutions becoming a game changer

• The crawl, run, walk phases of agentic solutions, and where we are now

• The role of other smart devices in creating agentic experiences

• How support for a 30 billion MoE on-device model was made possible

• Qualcomm's role in working with partners to bring AI experiences to life

If you want to learn more about how the latest chipsets will change the future of Android flagship devices in the next year, including new AI experiences, this conversation will give you a clear idea.

📺 Watch on YouTube

🎧 Listen in your favorite podcast player

Subscribe to Deep View Conversations for interviews with the leaders shaping the future of AI, business, and technology.

And don't forget to sign up for The Deep View daily newsletter. We don’t just cover AI, we decode it. In a world flooded with hype, we deliver sharp, no-nonsense insights to keep you ahead of the curve and help you put AI to work every day: subscribe.thedeepview.com

Disclaimer: Sabrina Ortiz's travel to Snapdragon Summit was paid for by Qualcomm. The Deep View's coverage is editorially independent from the companies we cover.

LINKS

Oracle's New Mexico data center faces power, permitting problems
Bill Gates claims AI could be powerful enough to wipe out humanity
Ex-Anthropic worker Jacob Coxon reportedly worked with safety PR firm
AI finance tech firm Numeral raises $100 million Series C
Google researchers find dark web hackers sell access to AI models
OpenAI pauses training of most powerful AI after model passed restrictions

Microsoft Copilot: Microsoft has launched its Copilot "superapp" that combines three AI capabilities into one interface of chat, coding, and agents.
Exa Agent Ultra: Orchestrates swarms of agents to perform exhaustive research, build comprehensive lists, and answer questions.
Runway MCP: The AI video platform now connects directly into Claude Opus 5.5 to generate polished images and videos with Gen-4.5, Seedance 2.5, GPT Image 2, Kling and others.
LongCat-2.5-Preview: A 1.6 trillion-parameter multimodal model with a 1 million token context window from Chinese developer Meituan.