TL;DR
- The vast majority of AI companies are thin software layers built on top of someone else's technology, and their valuations don't reflect that fragility.
- Compute infrastructure and energy access are the real competitive moats in AI, not flashy consumer apps.
- AI talent wars have pushed top-researcher compensation past $1 million, crushing profitability for most players.
- Many AI products still lose money on every user served, subsidized by venture capital that won't last forever.
- The biggest long-term winners in the AI boom are likely to be the "boring" infrastructure companies most retail investors overlook.
Everyone wants a piece of the AI boom. And who can blame them?
AI companies attracted over $80 billion in venture capital in 2025, capturing more than a third of all global VC funding despite representing a fraction of funded startups. OpenAI reached a $300 billion valuation after its March 2025 funding round.¹ The four largest hyperscalers told investors they will spend roughly $700 billion on capital expenditures in 2026, nearly double what they spent the prior year.⁹ The gold rush is real.
But here's the problem: most investors are buying into a narrative, not a business model. They're chasing ticker symbols with "AI" in the pitch deck without understanding the brutal economic realities underneath. And those realities are about to catch up.
Why it matters: The gap between AI hype and AI economics has never been wider. The investors who understand these seven truths will be positioned to profit. The ones who don't will be left holding the bag when the market stops rewarding "AI" as a magic word and starts demanding proof of monetization.
Let's break down the seven truths that most investors still get wrong about AI companies, and what you should actually do about it.
1. Most "AI Companies" Are Just Wrappers Around Existing APIs
Here's a dirty little secret that the AI startup ecosystem doesn't want you to know: the vast majority of companies branding themselves as "AI-powered" are doing little more than making API calls to models built by someone else.
Their "proprietary AI system" is an OpenAI or Anthropic API call. Their "custom engine" is a carefully designed prompt. Their "intelligence layer" is a large language model with a pretty user interface sitting on top. Strip away the marketing, and you're looking at a thin software wrapper with zero technical defensibility.
Why it matters: When the platform you're built on decides to offer your core feature for free (and they will), your entire business model evaporates overnight. We've already watched this happen. Jasper, which peaked at around $80 million in annual recurring revenue, saw growth stall after ChatGPT started doing the same thing for free. Dozens of "chat with your PDF" apps launched and died within months of each other. Meeting summarizer companies watched their value proposition collapse when Zoom and Microsoft Teams added native AI summaries.
The companies that survive the wrapper apocalypse share one trait: they have something the foundation model providers can't replicate quickly. Cursor built deeply integrated IDE features that go far beyond simple code generation. ElevenLabs developed proprietary voice models with quality that foundation models can't match natively. Harvey AI has access to proprietary legal datasets that no general-purpose model can replace.
The investor takeaway: Before you invest in any AI company, ask one question: if OpenAI or Google shut off their API access tomorrow, would this company still have a product? If the answer is no, you're not investing in an AI company. You're investing in a reseller with a countdown clock.
2. Compute and Energy Are the Real Moats
Forget algorithms. Forget clever chatbot interfaces. The true competitive advantage in AI comes down to something far more mundane: raw computational power and the electricity to run it.
The numbers are staggering. Microsoft, Meta, Amazon, and Alphabet collectively told investors during their Q1 2026 earnings calls that they will spend roughly $700 billion on capital expenditures in 2026.⁹ Three of the four raised capex guidance during that reporting period.⁹ An estimated 75% of aggregate hyperscaler capex in 2026 is going directly to AI-related infrastructure, representing approximately $450 billion in AI-specific spending.¹⁰
Why it matters: This level of spending creates a barrier to entry that is effectively insurmountable for new competitors. You can't build a frontier AI model without access to tens of thousands of high-end GPUs, and you can't run those GPUs without enormous amounts of reliable power. In the first half of 2025, AI-related capex contributed more to U.S. GDP growth than consumer spending.⁷
Power is the bottleneck, not capital and not demand. Hyperscaler regions are becoming saturated. Utilities favor predictable, large-scale power consumers. New capacity faces lead times measured in years. Community resistance to data center construction has flared up across multiple states.
A single NVIDIA B200 GPU server system runs about $500,000. NVIDIA is running approximately 80% gross margins on these chips, meaning the barrier to entry in compute hardware is also sky-high.
The investor takeaway: If you're bullish on AI, the safest bet is the companies that own the physical infrastructure everyone else depends on, not the app layer.
3. Talent Costs Have Become Completely Unsustainable
The AI talent war has created one of the most distorted labor markets in modern history.
AI professionals command a median salary of roughly $160,000 annually, with specialized skills adding 25 to 45% premiums on top of base compensation.⁸ Those are the averages. The top 1% of AI researchers now command total compensation packages exceeding $1 million, including stock grants between $2 and $4 million at late-stage startups.⁸
AI and ML hiring grew 88% year-over-year in 2025.⁵ At the same time, administrative hiring dropped 35.5% and entry-level hiring plummeted over 73%.⁵ The industry is hoovering up senior talent at premium prices while the rest of the workforce gets squeezed.
Why it matters: These compensation levels are mathematically unsustainable for most AI companies, especially the ones that aren't yet generating meaningful revenue. When you're paying your top five researchers $5 million combined and your product is still losing money on every user, the runway math gets ugly fast.
Companies that delayed hiring AI talent in early 2024 ended up paying 15 to 20% premiums for the same skills just months later. The bidding war shows no signs of cooling, and every dollar spent on talent acquisition is a dollar not spent on infrastructure, marketing, or achieving profitability.
The investor takeaway: When evaluating AI companies, pay close attention to headcount growth, compensation costs as a percentage of revenue, and whether the company is building a sustainable team or simply outbidding competitors in an unsustainable auction.
4. Most AI Products Are Losing Money on Every User
Here's a number that should make every AI investor uncomfortable: in 2025, OpenAI anticipated total spending of roughly $22 billion against $13 billion in sales, resulting in a net loss of approximately $9 billion.¹ That means the company behind the most successful consumer AI product in history was spending approximately $1.69 for every dollar of revenue it generated.¹
Those losses were driven heavily by the cost of serving billions of inference requests per day. OpenAI's inference costs reached an estimated $8.4 billion in 2025.¹ CEO Sam Altman publicly stated on X in January 2025 that OpenAI was actively losing money on its $200-per-month ChatGPT Pro subscriptions.
Why it matters: The current API pricing that enterprises budget around is subsidized by venture capital and hyperscaler cross-subsidies. When that subsidy ends, prices go up or companies go under.
Gartner projects that by 2030, performing inference on a large language model with one trillion parameters will cost GenAI providers over 90% less than in 2025.² That sounds encouraging until you read the fine print: token consumption is rising faster than token costs are falling, which means total inference spending keeps going up even as per-token prices come down.² Gartner's own analysts warn that companies shouldn't confuse cheaper commodity tokens with the democratization of frontier reasoning.²
The investor takeaway: Unit economics matter more than user growth. Ask every AI company you're considering: what does it cost to serve one user, and when do you expect that number to be less than what that user pays you? If they can't answer clearly, proceed with extreme caution.
5. High-Quality Training Data Is Running Out
Every frontier AI model needs enormous quantities of high-quality text, code, and multimedia data to train on. The problem? The supply of genuinely useful training data is finite, and we're approaching the wall.
The most commonly cited high-quality text sources (books, academic papers, curated web pages, Wikipedia, and high-quality code repositories) have largely been consumed by existing models. What remains is increasingly noisy, redundant, or legally contested. Copyright lawsuits over AI training data are multiplying, with publishers, artists, and content creators pushing back hard.
Synthetic data, the practice of training AI models on outputs generated by other AI models, has been proposed as a solution, but it carries well-documented quality risks. Studies have shown that training on synthetic data can introduce compounding errors, a problem researchers sometimes call "model collapse." Each generation of synthetic training degrades signal quality in subtle ways that become catastrophic over time.
Why it matters: Data scarcity creates a massive advantage for companies that already have access to proprietary, high-quality datasets. It also means that the next wave of meaningful model improvements may come more slowly than the market expects. If you're investing based on the assumption that AI capabilities will keep improving at the pace we've seen over the past three years, you may be disappointed.
The EU AI Act, which becomes broadly applicable in August 2026, requires providers of general-purpose AI models to publish summaries of their training data and adopt copyright compliance policies.³ That regulatory pressure further constrains the data supply for companies that have been relying on loosely defined "fair use" arguments.
The investor takeaway: Proprietary data is the new oil. Companies with exclusive access to valuable, domain-specific datasets (medical records, financial filings, legal documents, industrial sensor data) have a structural advantage that API wrappers never will. Invest accordingly.
6. Regulation and Government Intervention Are Coming Faster Than Expected
If you think AI regulation is a distant concern, you haven't been paying attention.
The EU AI Act entered into force in August 2024 and is rolling out in phases. Prohibited AI practices have been banned since February 2025. Rules for general-purpose AI models took effect in August 2025. The high-risk AI system rules become fully applicable in August 2026.³ Penalties for non-compliance range from €7.5 million or 1.5% of worldwide annual turnover up to €35 million or 7% of worldwide annual turnover, depending on the severity of the infraction.³
In the United States, the Trump administration published its AI Action Plan in July 2025, covering everything from export controls and chip restrictions to data center permitting and national security safeguards. The plan includes over 90 federal actions aimed at securing U.S. leadership in AI while addressing security risks.
Geopolitically, the landscape is shifting fast. Export controls on advanced AI chips have already reshaped supply chains. Companies operating in regulated industries like healthcare, finance, and legal face additional layers of compliance complexity that most AI startups are simply not prepared for.
Why it matters: Regulation doubles as a competitive moat for companies that get compliance right early. The companies that invest in governance, data lineage, and transparency infrastructure now will have a massive advantage when enforcement ramps up. The ones that treat compliance as an afterthought will face fines, lawsuits, and market access restrictions that could destroy them.
The investor takeaway: Regulatory readiness is an underappreciated differentiator. When evaluating AI investments, ask whether the company has a compliance strategy that goes beyond "we'll deal with it later." In a tightening regulatory environment, "later" is already too late.
7. The Biggest Winners Will Be Boring Infrastructure Companies
Every gold rush has its winners, and they're rarely the miners. They're the ones selling shovels, building railroads, and running the general store.
The AI boom is no different. While investors chase flashy consumer AI apps and next-generation chatbots, the real wealth is being built by the companies providing the infrastructure that makes all of it possible.
Consider the evidence. NVIDIA posted 114.2% revenue growth and a 67.5% EBITDA margin in fiscal year 2025.⁴ Seagate CEO Dave Mosley confirmed during the company's Q2 FY2026 earnings call that nearline HDD capacity is fully allocated through calendar year 2026, with long-term agreements extending into 2027 and customer discussions already underway for 2028.⁶ Hyperscaler capital intensity has surged to 45 to 57% of revenue, levels that would have been unthinkable just a few years ago for technology companies.¹⁰
The infrastructure layer includes chip manufacturers (NVIDIA, AMD, Broadcom), cloud providers (AWS, Azure, Google Cloud), energy companies powering the data centers, cooling technology providers, storage and networking companies, and enterprise infrastructure software. These businesses have pricing power, recurring revenue, and the kind of customer lock-in that consumer AI apps can only dream of.
Meanwhile, the "boring" infrastructure companies benefit from a structural advantage: they win regardless of which AI company or model comes out on top. Whether the dominant consumer AI product in 2028 is ChatGPT, Claude, Gemini, or something that doesn't exist yet, all of them will need chips, servers, storage, electricity, and data center space.
Why it matters: The infrastructure layer captures value with far less risk than the application layer. AI apps face winner-take-all dynamics, rapid commoditization, and the ever-present threat of the underlying platform eating their lunch. Infrastructure companies face growing demand, pricing power, and structural scarcity.
The investor takeaway: If you want AI exposure without betting on which app will dominate, look downstream. The picks-and-shovels play is where the durable returns are.
So What Does a Smart AI Investment Strategy Actually Look Like?
Let's pull it all together. Here's how to think about AI investing once you strip away the hype:
Focus on infrastructure over applications. The companies building and powering the AI stack (chipmakers, cloud providers, energy companies, storage manufacturers) have more durable competitive advantages and less binary risk than any individual AI app.
Demand proof of unit economics. Revenue growth means nothing if every user costs more to serve than they pay. Ask for gross margins, cost per inference, and a credible path to profitability. If a company can't articulate these numbers, they're still in the "hope" phase of their business plan.
Look for proprietary data moats. The AI companies with the best long-term prospects are the ones sitting on unique, high-value datasets that can't be replicated by competitors. Vertical AI companies targeting specific industries like legal, healthcare, and financial services with domain-specific data are better positioned than horizontal players building on commodity models.
Factor in regulatory risk. The regulatory environment is tightening globally. Companies without a clear compliance strategy are carrying hidden risk that their valuations don't yet reflect. Invest in companies that are ahead of the regulatory curve, not behind it.
Be skeptical of valuation multiples. AI startups currently trade at 10 to 50x revenue multiples, with some outliers clearing 100x. Those multiples are supported by investor FOMO and market narrative, not fundamentals. When discipline returns (and it always does), the companies without real margins will be the first to correct.
The AI revolution is real. The economic transformation it enables will be profound. But the biggest mistake an investor can make is confusing the revolution itself with every company claiming to be part of it.
Stay informed. Stay skeptical. And invest in the infrastructure, not the hype.
Disclaimer: This blog post is for informational purposes only and does not constitute investment advice. Equity Sesame does not recommend specific securities. All investments carry risk, including the potential loss of principal. Always conduct your own research or consult a licensed financial advisor before making investment decisions.
References
¹ Fortune, "OpenAI says it plans to report stunning annual losses through 2028," November 12, 2025. Based on financial documents obtained by The Wall Street Journal. OpenAI anticipated ~$22B in total spending against ~$13B in sales for 2025, resulting in an approximately $9B net loss. https://fortune.com/2025/11/12/openai-cash-burn-rate-annual-losses-2028-profitable-2030-financial-documents/
² Gartner, Inc., "Gartner Predicts That by 2030, Performing Inference on an LLM With 1 Trillion Parameters Will Cost GenAI Providers Over 90% Less Than in 2025," press release, March 25, 2026. https://www.gartner.com/en/newsroom/press-releases/2026-03-25-gartner-predicts-that-by-2030-performing-inference-on-an-llm-with-1-trillion-parameters-will-cost-genai-providers-over-90-percent-less-than-in-2025
³ European Commission, "AI Act: Shaping Europe's Digital Future," official regulatory framework page. Phased implementation timeline: prohibited practices (Feb 2025), GPAI rules (Aug 2025), high-risk systems (Aug 2026). https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
⁴ PitchBook, "Q1 2026 AI Public Comp Sheet and Valuation Guide," April 2026. NVIDIA FY2025 data: 114.2% revenue growth, 67.5% EBITDA margin. https://pitchbook.com/news/reports/q1-2026-ai-public-comp-sheet-and-valuation-guide
⁵ Ravio, "The AI Compensation and Talent Trends Shaping the Job Market in 2026," 2026 Compensation Trends Report. AI/ML hiring grew 88% YoY in 2025; administrative hiring fell 35.5%; entry-level (P1/P2) hiring fell 73.4%. https://ravio.com/blog/ai-compensation-and-talent-trends
⁶ Seagate Technology, Q2 FY2026 Earnings Call, January 28, 2026. CEO Dave Mosley: "Our nearline capacity is fully allocated through calendar year 2026." Long-term agreements in place through CY2027; CY2028 discussions underway. Reported via TrendForce and Blocks & Files.
⁷ KKR, "Beyond the Bubble: Why AI Infrastructure Will Compound Long After the Hype," February 2026. AI-related capex contributed more to U.S. GDP growth than consumer spending in H1 2025; top four hyperscalers expected to spend $350B+ in 2025 capex. https://www.kkr.com/insights/ai-infrastructure
⁸ Rise, "AI Talent Salary Report 2026," February 2026. Median AI professional salary of $160,000; specialized skill premiums of 25-45%; top 1% total compensation exceeding $1M including $2-4M stock grants. https://www.riseworks.io/blog/ai-talent-salary-report-2025
⁹ Om Malik, "What I Learned about Hyperscalers' AI Spend," April 30, 2026. Based on Q1 2026 earnings reports: Microsoft, Meta, Amazon, and Alphabet collectively guided for ~$700B in 2026 capex; three of four raised guidance during reporting week. https://om.co/2026/04/30/what-i-learned-about-hyperscalers-ai-spend/
¹⁰ IEEE Communications Society Technology Blog, citing CreditSights research, "Hyperscaler Capex > $600B in 2026," December 22, 2025. Approximately 75% of aggregate capex tied to AI infrastructure (~$450B); capital intensity reached 45-57% of revenue. https://techblog.comsoc.org/2025/12/22/hyperscaler-capex-600-bn-in-2026-a-36-increase-over-2025-while-global-spending-on-cloud-infrastructure-services-skyrockets/