NVIDIA doesn't make chips. They make files. (So do you.)
Introducing the File Theory of Business Value
Elon Musk said something on the Cheeky Pint podcast a few weeks ago that I can’t stop thinking about.
“The most valuable companies currently by market cap, their output is digital.”
Then he made it concrete: “Nvidia literally just FTPs files to Taiwan. They’re the only ones that can make files that good, but that is literally their output.”
Apple doesn’t make phones. They design files and send them to China for assembly. Microsoft doesn’t build Xboxes. Meta doesn’t manufacture anything physical at all.
Look at the current top five by market cap: NVIDIA ($4.5T), Apple ($3.9T), Alphabet ($3.8T), Microsoft ($3.5T), Amazon ($2.5T). Eight of the top ten are technology companies. Every one of them produces the same thing — digital files.
This isn’t a coincidence. It’s the formula.
I’m calling it the File Theory of Business Value: The closer your business gets to producing purely digital output, the more valuable and scalable it becomes.
Here’s the framework. There are three tiers:
Tier 1: Physical businesses. Value constrained by atoms, labor, geography, and time. A roofing company. A call center. Revenue scales linearly with headcount. You want to grow 2x, you hire 2x.
Tier 2: Hybrid businesses. Digital design, physical delivery. This is NVIDIA and Apple. The file is where the value concentrates, but someone still turns the file into a physical product. Massive margins because the file costs the same to create whether you manufacture one unit or one billion.
Tier 3: Pure digital businesses. Google, Meta, Microsoft’s cloud. The file IS the product. Infinite replication at zero marginal cost. This is where the highest valuations live.
The strategic question: How do you move up the stack?
Now here’s why this matters to everyone reading this newsletter.
The mortgage industry is already a file business. It just doesn’t realize it.
A loan application is a file. A rate lock is a file. A disclosure package is a file. An appraisal is a file. Underwriting is literally file processing — documents in, decision out. The entire origination workflow is files moving between systems and humans.
But we price it like Tier 1 — labor hours, headcount, per-loan costs. The industry average is $7,000-$9,000 to originate a single loan, with most of that cost tied to humans touching files.
Lenders who get this are already moving. AI-powered document verification is cutting processing from 48 hours to under 4. 38% of lenders used AI in 2024, up from 15% in 2023, and Fannie Mae projects 55% by end of 2025. The ones implementing it are reporting 30-50% reductions in operational expenses.
They’re not replacing mortgage professionals. They’re upgrading from Tier 1 to Tier 2. The human provides the judgment, the relationship, the trust. The AI handles the files.
The same logic applies to lead generation. A qualified lead is a file — name, contact info, intent signals, qualification score. Structured data in a row. The entire lead gen industry produces files, but prices them based on the cost of human labor to produce them.
When AI handles content creation, campaign optimization, lead qualification, and nurture sequences, the cost to produce lead #10,000 becomes the same as lead #10. Quality doesn’t degrade with volume because the system — the file — is the same. It actually improves, because each cycle refines the logic.
That’s the jump from Tier 1 to Tier 3. That’s where infinite scalability lives.
Musk’s other big claim: once AI achieves “digital human emulation” — anything a human can do on a computer — it unlocks trillions in value overnight. Harvard Business School research already shows AI users completing tasks 25% faster with 40% higher quality. And that’s with today’s tools, not the digital human equivalent Musk is describing.
The playbook is the same whether you’re a lender, an agency, or a solo loan officer:
Identify your file. What is the actual digital output that creates value? For a loan officer, it’s the complete loan package. For an agency, it’s the campaign. For a lead gen company, it’s the qualified lead record.
Decouple quality from human hours. Use AI to handle the execution while humans provide judgment and relationships. The file should be just as good whether a human spent 6 hours on it or AI produced it in 6 minutes with human review.
Make the file replicable. Once you have a high-quality system — a content engine, a campaign framework, a qualification process — it costs nothing to run it again. This is the NVIDIA principle: the design is the hard part, the deployment is trivial.
Build the feedback loop. The companies that win long-term aren’t just producing files faster. They’re building systems where each file makes the next one better. Content that improves based on performance data. Campaigns that optimize automatically. Lead scoring that refines itself against close rates.
The mortgage industry spent the last decade digitizing paperwork. The next decade is about realizing that the paperwork was always the product — and engineering it like NVIDIA engineers a chip.
THREE LINKS WORTH YOUR TIME
All Aboard the AI Train: A Practical Roadmap for Lenders — STRATMOR Group — STRATMOR’s breakdown of where AI fits in the mortgage workflow today. The document processing and underwriting applications are real and measurable. Good reality check on what’s working versus what’s hype.
The Where and When of AI Making Us More Productive — World Economic Forum — Chief economists weigh in on which sectors see productivity gains first. Financial services and professional services lead the list — exactly the knowledge-intensive, file-heavy industries where the theory applies.
Elon Musk on Space GPUs, AI, Optimus, and His Manufacturing Method — Cheeky Pint — The source material. Two hours and forty-six minutes with Elon, John Collison, and Dwarkesh Patel. The “most valuable companies make files” thread starts about midway through. Worth the full listen if you have the time.
ONE TACTIC TO TRY THIS WEEK
Map your file chain.
Pick one revenue-generating process in your business — loan origination, lead generation, a client deliverable — and list every file that gets created from start to finish. Application. Disclosure. Approval letter. Whatever it is.
Next to each file, write down who or what creates it and how long it takes.
Then ask: which of these files could be produced at the same quality, faster, with AI handling the execution and a human providing the judgment?
That’s your upgrade path from Tier 1 to Tier 2. And it’s probably shorter than you think.
That’s all for this week.
Bill