The Thesis
The AI revolution didn't start with ChatGPT. It started in 2015-2017 with decisions that determined who would win a decade later.
The Foundational Era
2014-2017
When today's winners and losers were determined
This is the deep history. We analyzed 98,000 Techmeme articles to trace how every major dynamic of 2025—the Microsoft hedge, the Anthropic-OpenAI narrative war, the transformer diaspora, Google's product gap—has roots in decisions made nearly a decade ago.
The coverage didn't anticipate any of this. The patterns were visible only in retrospect.
The AlphaGo Moment (March 2016)
On March 14, 2016, Google's AlphaGo defeated Go world champion Lee Sedol 4-1. It was the most heavily-covered AI event before ChatGPT.
Foundational Era Coverage Volume
Go was considered unsolvable by brute-force computation—more possible positions than atoms in the universe. AlphaGo proved that neural networks could develop intuition, not just calculation.
”"Google's AlphaGo wins historic match against Go grandmaster after three consecutive wins in five-game series" — Wired, March 2016
The Pattern That Would Repeat: Google treated AlphaGo as a research achievement, not a product opportunity. This pattern—research dominance without commercial exploitation—would define Google's AI story through 2025. DeepMind would win two Nobel Prizes (Hinton for neural networks, Hassabis for AlphaFold) without shipping a consumer product.
See The Great Power Redistribution for how Google's research dominance failed to translate into startup suppression.
The Three Godfathers
Three researchers shaped the foundational era. Their trajectories from 2016 to 2025 reveal how the industry evolved—and fractured.
Geoffrey Hinton: Research Commentator → Nobel Laureate → Safety Advocate
”"Geoffrey Hinton, the 'godfather of neural networks' on the importance of AlphaGo's use of intuition, why humans shouldn't fear AI, and what AI can take on next" — Maclean's, March 2016
Hinton's first Techmeme mention was as commentary on AlphaGo—"why humans shouldn't fear AI." By 2023, he would leave Google specifically to warn about AI dangers. The arc from reassurance to alarm spans the entire foundational-to-ChatGPT era.
”"The Royal Swedish Academy of Sciences awards the Nobel Prize in Physics to John Hopfield and Geoffrey Hinton for 'foundational discoveries' in machine learning" — Bloomberg, October 2024
First Techmeme mention
"Why humans shouldn't fear AI"
Leaves Google to warn about AI
Reverses 2016 position
Nobel Prize in Physics
For neural network foundations
Yann LeCun: Facebook AI Director → Meta Chief Scientist → Startup Founder
”"Facebook AI Director Yann LeCun on His Quest to Unleash Deep Learning and Make Machines Smarter" — IEEE Spectrum, February 2015
LeCun spent a decade building FAIR (Facebook AI Research), released Llama as open source, and became the industry's most vocal critic of AI safety "fearmongering." By 2025, he would leave Meta to found his own company.
”"Sources: Yann LeCun is in early talks to raise €500M for his startup at a ~€3B valuation" — Financial Times, November 2025
Demis Hassabis: DeepMind CEO → Google DeepMind CEO → Nobel Laureate
Unlike LeCun or Hinton, Hassabis stayed with one company. But he had to absorb Google Brain to consolidate power—a merger that took until 2023.
”"Alphabet's Google combines AI research units Brain and DeepMind under Demis Hassabis" — Wall Street Journal, April 2023
| Feature | Researcher | First Mention | 2016 Role | 2025 Role |
|---|---|---|---|---|
| hinton | Geoffrey Hinton | Mar 2016 | Research commentator | Nobel laureate, safety advocate |
| lecun | Yann LeCun | Feb 2015 | Facebook AI Director | Leaving Meta for €3B startup |
| hassabis | Demis Hassabis | Mar 2016 | DeepMind CEO | Nobel laureate, Google DeepMind CEO |
The divergence: One left to warn. One left to build. One stayed to consolidate. The "three godfathers" of deep learning took radically different paths—and the industry fractured along the same lines.
OpenAI's Founding (December 2015)
Four months before AlphaGo's victory, a nonprofit was founded that would eventually eclipse DeepMind in commercial impact.
”"OpenAI, a new nonprofit for AI research, gets $1B commitment from SV leaders including Reid Hoffman, Elon Musk, Peter Thiel" — OpenAI, December 2015
Related Articles
31
OpenAI founding coverage
The founding coverage focused on:
- The $1B commitment (though only ~$130M was initially deployed)
- Elon Musk's involvement
- The nonprofit structure
- Safety concerns as motivation
No one anticipated that OpenAI would:
- Convert to a capped-profit structure (2019)
- Partner with Microsoft for $13B (2023)
- Launch ChatGPT and trigger the AI boom (2022)
- Achieve a $157B valuation (2025)
- Fire and rehire its CEO in 48 hours (2023)
- Become the subject of an Elon Musk lawsuit (2024)
The nonprofit structure that seemed like an idealistic choice in 2015 became the governance crisis of 2023. See The Microsoft Hedge for the full story.
The Microsoft Seed (November 2016)
The first Microsoft-OpenAI partnership was planted just 11 months after OpenAI's founding:
”"OpenAI partners with Microsoft, will use Azure for majority of its cloud computing needs" — Wired, November 2016
This would grow into a $13B relationship—and then fray into a hedge. The entire arc from partnership to crisis to diversification is documented in The Microsoft Hedge.
TensorFlow: Google Opens the Door (November 2015)
In November 2015, Google made a fateful decision: open-source its deep learning infrastructure.
”"TensorFlow - Google's latest machine learning system, open sourced for everyone" — Google Research, November 2015
Strategic Brilliance and Strategic Error: TensorFlow became the standard framework, spreading Google's influence. But competitors—including OpenAI—could build on Google's work without licensing costs. The infrastructure layer was commoditized before the application layer emerged.
Six months later, Google announced custom AI chips (TPUs). Custom AI chips would become a multi-billion-dollar market. By 2025, Nvidia would be worth $3T+ primarily on AI demand—a market Google helped create but didn't capture.
The Transformer Moment (June 2017)
The "Attention Is All You Need" paper was published in June 2017. It introduced the transformer architecture that would power GPT, BERT, Claude, Gemini, Llama, and every modern LLM.
Techmeme coverage at the time: Zero.
Articles at Publication
0
Transformer paper coverage (2017)
The paper would only be covered retrospectively—seven years later:
”"An in-depth look at the 2017 'Attention Is All You Need' paper, a big breakthrough in AI, and profiles of the eight Google researchers who co-authored the paper" — Wired, March 2024
The Coverage Blind Spot: Technical papers don't make headlines. Product launches do. But the papers determine which products are possible. The transformer paper received zero coverage in 2017 and enabled $500B+ in market value by 2025.
The Transformer Diaspora
Google invented the transformer and lost most of its inventors. This pattern—talent redistribution from Big Tech to startups—accelerated through 2023-2025.
| Feature | Author | 2017 Role | 2025 Role | Company Value |
|---|---|---|---|---|
| shazeer | Noam Shazeer | Google researcher | Returned to Google | $2.7B acquisition |
| gomez | Aidan Gomez | Google researcher | Cohere CEO | $5B+ |
| uszkoreit | Jakob Uszkoreit | Google researcher | Inceptive founder | Private |
| vaswani | Ashish Vaswani | Google researcher | Essential AI founder | Private |
| polosukhin | Illia Polosukhin | Google researcher | NEAR Protocol | $2B+ |
”"Q&A with Cohere CEO Aidan Gomez on co-authoring the 'Attention is all you need' paper at Google, focusing on enterprise, whether there's an AI bubble, and more" — The Verge, June 2024
The pattern: Google invented the transformer, lost five of eight authors to startups, and paid $2.7B to bring back Noam Shazeer—the architect of the technology that powered its competitors.
Nvidia's GPU Pivot (2015-2016)
While attention focused on AI software, Nvidia was building the hardware foundation that would make it a $3T company.
”"Nvidia announces Drive PX 2, a computer for self-driving cars that uses deep neural networks capable of recognizing 120M+ objects" — TechCrunch, January 2016
Tegra X1 unveiled
256-core Maxwell GPU
Drive PX 2 announced
Deep neural networks for self-driving
Tesla P100 GPU
15B+ transistors for deep learning
GTX 1080 launch
Consumer GPU revolution
CANDLE partnership
AI for cancer research with DOE
Nvidia's 2015-2016 strategy: Gaming GPUs → Self-driving partnerships → AI training hardware. The pivot looked speculative at the time. By 2025, Nvidia was worth $3T+, and the AI chip market was the most strategic battleground in tech.
IBM Watson: The Marketing-Without-Capability Lesson
IBM Watson was the most heavily-marketed AI brand of 2015-2016. The coverage was extensive—healthcare, IoT, enterprise partnerships. Watson received 100+ articles in 2015-2016 while the transformer paper received zero.
Watson's problems:
- Proprietary (unlike TensorFlow)
- Expensive (enterprise licensing)
- Overpromised (especially in healthcare—MD Anderson cancelled its Watson oncology project)
- Not based on modern deep learning
By 2025, Watson would be forgotten. OpenAI, founded the same year IBM opened its Watson IoT headquarters, would be worth $157B.
The lesson: Marketing without capability doesn't scale. OpenAI spent years in research obscurity before ChatGPT. IBM spent years marketing before admitting Watson didn't work.
The Coverage Gap Analysis
The foundational era reveals systematic blind spots in tech coverage that persist today.
| Feature | Event | Actual Importance | 2015-2017 Coverage | Verdict |
|---|---|---|---|---|
| alphago | AlphaGo Match | Proved neural net intuition | 50 articles | Appropriate |
| transformer | Transformer Paper | Enabled all modern LLMs | 0 articles | Undercovered |
| openai | OpenAI Founding | $157B company by 2025 | 31 articles | Appropriate |
| watson | IBM Watson | Failed enterprise AI | 100+ articles | Overcovered |
The meta-pattern: Product launches get covered. Technical papers don't. But the papers determine which products are possible. This is why The AI Infiltration Effect matters—the coverage shapes perception, but the research shapes reality.
The Seeds of 2025
The foundational era (2014-2017) established everything that matters today:
TensorFlow open sourced
Google shares infrastructure
OpenAI founded
$1B nonprofit commitment
AlphaGo defeats Lee Sedol
AI masters intuition
OpenAI-Microsoft partnership
Seed of $13B relationship
Transformer paper published
Zero coverage, $500B+ impact
The Causal Chains
The biggest stories of 2025 have roots in decisions made in 2015-2017:
OpenAI's nonprofit structure (2015) → governance crisis (2023) → Microsoft hedge (2024-2025)
DeepMind's research focus (2016) → Google's product gap (2023) → Gemini delays → Anthropic narrative advantage
Transformer paper at Google (2017) → author diaspora → Cohere, Character.AI, Essential AI → startup coverage parity
TensorFlow open-sourcing (2015) → infrastructure commoditization → application layer opportunity
The Meta-Insight
The foundational era created the conditions. The ChatGPT era exploited them.
Every major dynamic analyzed in this research collection—the AI infiltration of all tech coverage, the power redistribution from Big Tech to startups, the chatbot-to-agent terminology shift, the Microsoft hedge, the Anthropic-OpenAI narrative war—traces back to decisions made in 2015-2017.
The coverage didn't anticipate any of this. Neither did the participants. The patterns were visible only in retrospect.
That's why this is the deep history.
See also: The Microsoft Hedge for the OpenAI partnership arc, The AI Infiltration Effect for coverage analysis methodology, The Great Power Redistribution for startup vs. Big Tech dynamics, and The Agentic Category for 2025's category creation.
The Microsoft Hedge: How a $13B Bet Became a Portfolio Strategy
Microsoft's $30B Azure deal with Anthropic wasn't a sudden pivot - it was the culmination of a 20-month hedging strategy that began when the Altman firing revealed Microsoft had no control over its biggest AI bet.
The AI Infiltration Effect: What 77,000 Articles Reveal About Tech's Structural Shift
Tech news feels samey. We quantified why. Analysis of 77,000 Techmeme articles reveals AI didn't just grow—it infiltrated every other beat. The data behind a permanent reorganization.
The Great Power Redistribution: AI Startups vs. Big Tech in the Attention Economy
Everyone says AI concentrates power in Big Tech. The data says the opposite. Startups went from 3% to 86% of Big Tech's AI coverage in five years. What the narrative got wrong.
The Agentic Category: How Enterprise AI Invented a Word and a $100B Market
The word "agentic" didn't exist in tech coverage until January 2025. By December, it appeared in 50 headlines and defined a category that spawned $10B valuations and 139 funded startups. The data on how a word became a market.