AI has real problems. The smart money is investing in the companies solving them now.

MW AI has real problems. The smart money is investing in the companies solving them now.

By Charlie Garcia

This overlooked area of tech is making money for investors

The same AI that aced the genius test can't count how many times the letter "R" appears in "strawberry."

OpenAI's o3 just cleared artificial general intelligence $(AGI)$ benchmarks. Eighty-seven percent on ARC-AGI, the test that's supposed to measure whether machines can actually think.

Silicon Valley popped the champagne. OpenAI's Sam Altman took a victory lap. The headlines screamed "AGI is here!"

Except there's one tiny problem. The same AI that aced the genius test can't count how many times the letter "R" appears in "strawberry."

It's three, by the way. S-T-R-A-W-B-E-R-R-Y. Three R's. But advanced language models (the ones that can write legal briefs and debug code) insist there are two.

My 4-year-old grandson can count the three Rs in strawberry. I could reward him with Skittles. OpenAI's o3 costs up to $30,000 per task and still can't figure it out.

AI fails at character-level reasoning because of how it processes language as "tokens" rather than individual letters. When you can write a sonnet but can't spell "strawberry," you haven't achieved intelligence. You've achieved an extremely expensive parlor trick.

It's like teaching a dog to play poker, then discovering it can't count the cards. Impressive? Sure. Useful? Not if you're playing for money.

This isn't a bug. It's a metaphor for America's entire AI strategy. We're winning a race to build artificial gods that can't count. China is winning the race to deploy AI that actually works.

But some middle-layer companies are bridging this gap. They're about to print money. And Wall Street is missing it entirely.

Read: Sam Altman was finally asked how OpenAI can target trillions in spending on very little revenue

The wrong holy grail

China isn't trying to build artificial deities. It's embedding good-enough AI into manufacturing, logistics, and infrastructure.

Former Alphabet $(GOOG)$ $(GOOGL)$ CEO Eric Schmidt calls artificial superintelligence "tech's holy grail." In congressional testimony this year, Schmidt warned that America's AI sector would need power equivalent to 90 nuclear plants by 2030.

Meanwhile, a report in Foreign Affairs, entitled "The Cost of the AGI Delusion," argued that, by chasing superintelligence, America is falling behind. China isn't trying to build artificial deities. It's embedding good-enough AI into manufacturing, logistics and infrastructure - with 70% adoption targets by 2027.

Yes, American companies are deploying AI aggressively. Microsoft's $(MSFT)$ Copilot, Salesforce's (CRM) Einstein - every major enterprise is racing to integrate LLMs. U.S. tech companies lead in cutting-edge AI research.

But here's the uncomfortable truth: More than 80% of U.S. AI projects fail to deliver results. Eighty-eight percent of pilots never reach production. The issue isn't whether America is using AI - it's whether the U.S. can deploy it at scale. China's treating AI adoption like high-speed rail: centrally coordinated, massive infrastructure investment and mandatory targets. America is letting each company figure it out independently.

U.S. companies are trying to build superintelligence that can't count to three. China is building infrastructure that works right now.

As President Donald Trump's AI czar, David Sacks, admitted: "China is not years and years behind us in AI. Maybe they're three to six months." Real impressive lead we've got there.

A recent analysis coauthored by Schmidt; Alexandr Wang, now chief AI officer at Meta Platforms (META); and Dan Hendrycks, director of the nonprofit Center for AI Safety, warned that if either the U.S. or China approaches superintelligence first, the other would view it as a national-security emergency, potentially triggering preventive cyberattacks or even kinetic strikes on AI datacenters.

America is betting civilization on a sprint to build something it doesn't understand while China wins the race to deploy AI that works.

The missing ingredients nobody's fixing

Three fundamental problems stand between automation and actual reasoning:

1. Autonomous self-learning: Current AI needs massive training runs consuming millions of dollars in compute. It can't learn continuously like humans do.

2. Structural reasoning: AI excels at pattern matching. It fails at causal logic, understanding why something happens, not just that it happens.

3. Asymmetric ethics: How do you align values across vastly different cognitive scales? How does an AI that thinks a million times faster than humans adopt human values?

These aren't engineering problems. There are fundamental gaps in how AI works. And while America obsesses over building smarter models, China simply deploys imperfect AI anyway.

The Foreign Affairs report identified the problem: America is trying to deploy 2025 AI on 1990s enterprise infrastructure. China built its platforms this decade. The U.S. is retrofitting.

Read: The AI boom is over - here's your bubble survival guide

The middle-layer gold rush

Invest in the companies solving AI's actual problems right now.

Here's where this gets interesting for your portfolio: You should absolutely own Nvidia and Microsoft. They're the picks-and-shovels play on the AI boom.

But there's a second layer most investors are missing entirely - the companies solving AI's actual problems right now. Not the chips. Not the models. The middle layer: verification, governance, data pipelines, security and alignment tools.

Think about it: What good is AI that writes legal contracts if you can't verify it's correct? What good is AI that analyzes financial data if it hallucinates numbers? What good is superintelligence if it can't count to three?

Solving verification

Fair Isaac (FICO) is building explainability frameworks that let humans actually understand what AI is doing. In regulated industries - finance, healthcare, defense - you can't deploy AI systems you can't explain. Fair Isaac built its reputation on making black-box credit decisions transparent. Analysts project 28% upside potential as AI explainability demand grows, and the company's 71% institutional ownership includes Vanguard, BlackRock and State Street.

Solving governance

ServiceNow (NOW) integrates AI governance into enterprise platforms. As AI proliferates, companies need systems to track which models are deployed where, monitor their decisions and ensure compliance. ServiceNow reported revenue up 21% year-over-year to $11 billion, with 29 analysts rating it a consensus "buy."

Palantir Technologies (PLTR) is already embedded in defense and intelligence, building the infrastructure to manage AI systems in high-stakes national security applications. Bank of America analysts projected that Palantir's government AI contracts could reach $8 billion by 2030, and put a $215 price target on the shares. Palantir recently secured a $10 billion, 10-year contract with the U.S. Army and a $795 million expansion for its Maven Smart System.

Solving security

CrowdStrike Holdings (CRWD) and Palo Alto Networks (PANW) become prime targets as AI systems integrate into critical infrastructure. AI-native security isn't optional; it's existential. Major institutional investors - Vanguard, BlackRock and State Street - have increased positions, with CrowdStrike seeing $17.8 billion in institutional inflows.

Solving infrastructure

Eric Schmidt is right: AI's natural limit is electricity, not chips. In a recent column, I discussed the companies solving the AI power crisis.

But there's one more layer for investors: the operational infrastructure that connects AI to physical industrial systems.

Siemens (XE:SIE) builds the industrial automation and digital twin platforms that let AI actually control manufacturing lines, logistics networks and infrastructure systems. This isn't just software or electricity, but the bridge between AI models and physical operations. China's AI Plus Initiative aims to embed AI into 70% of manufacturing by 2027 and 90% by 2030. Siemens operates extensively in China with manufacturing subsidiaries across platforms spanning finance, procurement and automation.

Your strategic play now

You're not betting on who makes AI smarter. You're betting on who makes it work.

Here's the investment thesis: Two races are happening simultaneously.

America's race: Build the smartest AI possible. This means high-risk, massive capital requirements, unclear timelines and unsolved fundamental technical problems. It means leading in research, but failing at deployment.

China's race: Deploy good-enough AI everywhere. It's happening now, with measurable results and actual economic impact.

These companies don't need Chinese customers to profit from China's AI lead. They profit from the West's panicked response. When Washington realizes it's falling behind - as happened with electricity infrastructure - the scramble begins. The U.S. government will need to rapidly deploy the verification, governance and security systems that make AI trustworthy at scale.

Whether America builds superintelligence first, or China embeds AI everywhere first, or both happen at once: Investors in these companies win. You're not betting on who makes AI smarter. You're betting on who makes it work.

Charlie Garcia is founder and a managing partner of R360, a peer-to-peer organization for individuals and families with a net worth of $100 million or more.

Agree? Disagree? Share your comments with Charlie Garcia at charlie@R360Global.com. Your letter may be published anonymously in the weekly "Dear Charlie" reader mailbag. By emailing your comments to Charlie Garcia, you agree to have them published on MarketWatch anonymously or with your first name if you give permission.

MW AI has real problems. The smart money is investing in the companies solving them now.

By Charlie Garcia

This overlooked area of tech is making money for investors

The same AI that aced the genius test can't count how many times the letter "R" appears in "strawberry."

OpenAI's o3 just cleared artificial general intelligence (AGI) benchmarks. Eighty-seven percent on ARC-AGI, the test that's supposed to measure whether machines can actually think.