Unique Research · Insight · July 24, 2024
By Stephen DeAngelis, CEO and Founder, Enterra Solutions
Access to verified data and expert knowledge bases is becoming decisive in the AI race.
The global AI arms race is accelerating in real time. In May, Google and Microsoft-backed OpenAI each shipped new multimodal models with striking voice and vision analysis — releases within 24 hours of each other that underscored how fierce H1 2024 competition already was. Meta has outlined aggressive AI investment plans; Amazon completed a $4B commitment to Anthropic in March — its largest external investment in thirty years.
What is unfolding is a true big-tech contest: hyperscalers racing to build the most advanced systems, break old operating constraints, and open new commercial ground. It starts with data centers that can host advanced AI — Microsoft and OpenAI have discussed a $100B facility, on the order of 100× the cost of today’s largest existing data centers.
Building models is only one layer. With AI investment expected to exceed $1.8T over coming years, who wins will depend on embedding AI in business applications that actually make money. That requires deep expert-grade databases — far more valuable than raw scrapes from the open web — to power next-generation GenAI. Systems trained only on unverified mass data cannot tell signal from noise, and are poorly suited to high-accuracy autonomous apps that amplify human capability. In short: smarter, more precise systems — not just more data.
But the “ocean” of quality text is drying up. A recent Wall Street Journal report warned that within two years demand for high-quality text may exceed supply, constraining model progress. Leading labs already need training sources broader than what the public internet offers. Some turn to synthetic data — innovative, but any inaccuracy can poison the stack. Others pursue content licenses: on May 22 OpenAI struck a major deal with News Corp for current and archive content from titles such as the Wall Street Journal, Barron’s, and MarketWatch.
Neither synthetic data nor licensing alone is a durable moat. Lasting advantage goes to systematic verification, integrity, and reliability of GenAI data — with certification or a “trusted” mark — plus expert knowledge from trusted external sources. Those two factors — AI trust and the compute of emerging data centers — may decide which big-tech brands lead.
Domain expertise
GenAI is only a slice of the arms race. Next-generation systems will fuse multiple AI techniques for fine-grained business needs, then stack large commercial apps for consumers and enterprises. Major developers may acquire smaller tech companies to lock in deep vertical data for training. Think of future Gemini and ChatGPT models as carrier-class platforms — the author compares them to the Gerald R. Ford–class carrier (100,000 tons, $13B, 12 years) — while acquisitions supply a more agile “fleet” of industry applications riding that base.
Unconventional partnerships
Academia can become a primary source of foundational knowledge. Princeton’s libraries alone hold more than 7 million volumes. Imagine Meta or Amazon digitizing Ivy League print collections under exclusive rights to train systems for complex industries such as manufacturing and aerospace — and, if scaled across higher education, a vast new knowledge base that helps challengers close on Google and OpenAI. Such deals could also open revenue for universities under enrollment and student-debt pressure: U.S. four-year private for-profit college freshman enrollment has fallen more than 55% since 2010; student loan debt exceeds $1.7T; federal loan rates hit 6.53%, the highest since 2012. A large new content-license income stream could help schools cut tuition and broaden access — a win-win framed in the source.
— END —