[Sequoia Capital, Sand Hill Road — January 2015, 10:00 AM]
The pitch deck had seventeen slides now — eight more than the nine-slide version Sarah had stripped down in a café eleven months ago, back when the company was two people and twelve thousand dollars and the phrase "AI that writes" drew blank stares from every investor in the Bay Area.
The extra slides held numbers. Real numbers. Revenue: $200,000 in annual recurring contracts, across six enterprise clients, with a pipeline of fourteen more in various stages of negotiation. Customer retention: one hundred percent — nobody who'd tried the documentation product had cancelled. Net Promoter Score: 72, which Diana had described as "insanely good for enterprise software" and Sarah had described as "a number that doesn't measure what matters but investors love it."
Ethan presented to Sequoia's AI-focused partner, a man named Andrew Lau, who'd spent twelve years at Google Research before crossing to venture and who asked questions that demonstrated he understood the difference between attention mechanisms and autocomplete. The first VC in fourteen months who hadn't needed the distinction explained.
"Your Transformer paper has twelve citations in two months," Andrew said, scrolling through the deck on his tablet. "That's unusual velocity for a pre-print from a startup. The Google Brain group referenced it in a workshop paper last week."
"The architecture is gaining academic traction. We're ahead of the research community, and the research community is starting to notice."
"Your training infrastructure. The cloud provider — ChronoCloud." Andrew looked up from the tablet. His expression was analytical, not suspicious — the face of a man evaluating a variable rather than questioning a story. "Our due diligence team flagged it. Provider doesn't appear in any commercial registry."
Monica intercepted from her seat beside Ethan — the same smooth deflection she'd perfected over dozens of meetings. "Proprietary research arrangement with a hardware manufacturer under NDA. Pre-release silicon, not commercially available. We've provided benchmarking data that verifies the hardware performance independently."
"The benchmarks are impressive. The provenance is unusual." Andrew set the tablet down. "I'm less concerned about the provider's identity than about dependency risk. If this provider disappears, what happens to your training pipeline?"
The question was new. Not "who is your provider?" but "what if your provider vanishes?" A practical concern from a practical investor — the kind of question that came from someone who'd seen startups die because a critical vendor went bankrupt.
"We're building redundancy," Ethan said. The statement was technically true — Sarah and Marcus had been developing a secondary training pipeline on standard AWS infrastructure, limited to smaller models but functional as a backup. "Our current provider is the only source for our largest-scale training runs, but we can sustain product-level operations on commercial hardware."
Andrew nodded. Wrote something on his tablet. The cloud question was filed — not dismissed, not resolved, but categorized as a manageable risk rather than a disqualifying one.
The meeting ran seventy minutes. Andrew asked about competitive moat, team composition, market timing, and the specific technical advantages of the GPT architecture over what Google and Facebook were building internally. Ethan answered each one with the fluency of someone who'd been pitching for a year and had calibrated every response to the particular frequency of investor comprehension.
The difference from the Basecamp meeting — from Alan Rao's "like Siri?" and the bar afterward — was categorical. Alan had been a generalist evaluating a category he didn't understand. Andrew was a specialist evaluating technology he'd helped build in a previous career. The conversation operated at a level of technical depth that made the pitch feel like a collaboration rather than a performance.
"I want to bring this to our Monday partnership meeting," Andrew said as the session closed. "I'll need a full financial model and access to the product for our technical review team."
"We'll have both to you by Friday."
Walking to the parking lot — where the Honda Civic sat between a Porsche Cayenne and a Mercedes S-Class, maintaining its tradition of economic dissonance — Monica allowed herself the pen-tap. Once. The tell of suppressed excitement.
"He didn't compare it to Siri," Ethan said.
"He asked about training loss. Gradient stability. Attention head specialization." Monica pulled out her phone, already composing follow-up emails. "That's the first VC who's asked about training loss instead of total addressable market. The world is changing, Ethan."
"We changed it."
"Don't get philosophical. Get me the financial model by Thursday so I can review it before Andrew's Friday deadline."
The Honda started on the first try. The AC rattled — Monica was right, he should lease something — but the engine was steady. He drove north toward the city while Monica made calls from the passenger seat, her voice cycling between professional warmth and professional authority depending on whether she was talking to an investor or a lawyer.
The fundraising circuit had taken three weeks and eleven meetings. Sequoia was the eighth. The first seven had ranged from promising (Greylock, who wanted a second meeting) to disqualifying (a fund whose managing partner had asked "what happens when people stop reading?" as if literacy were a fad). The market had shifted since 2014 — AI was no longer a punchline. The arXiv paper, the enterprise revenue, the HooliBot disaster, and the general acceleration of interest in machine learning had moved the investment thesis from "too early" to "early but viable." The VCs who'd rejected him a year ago were now requesting meetings through their associates.
The callback from the early days — Patricia Liang saying "come back in 2018" at Meridian Ventures — surfaced as Ethan merged onto the 101. Patricia had been right about the timing. She'd been wrong about how long it would take for the timing to change. One year, not four. Because Ethan's presence in the timeline had accelerated the very forces that Patricia had predicted would take a decade to mature.
His phone buzzed. Sarah, from the office.
Priya ran the GPT-2 preliminary training at reduced scale. 500M parameters instead of 1.5B. Loss curve is promising. She wants to show you the attention head visualizations tonight.
The work continued. The pitch circuit was a parallel track — money on one side, technology on the other, both racing toward a convergence that would either fund the future or prove that the future cost more than anyone was willing to pay.
Author's Note / Promotion: Your Reviews and Power Stones are the best way to show support. They help me know what you're enjoying and bring in new readers! You don't have to. Get instant access to more content by supporting me on Patreon. I have three options so you can pick how far ahead you want to be: 🪙 Silver Tier ($6): Read 10 chapters ahead of the public site. 👑 Gold Tier ($9): Get 15-20 chapters ahead of the public site. 💎 Platinum Tier ($15): The ultimate experience. Get new chapters the second I finish them . No waiting for weekly drops, just pure, instant access. Your support helps me write more . 👉 Find it all at patreon.com/fanficwriter1