The 95% Number That Explains Why Sam Altman Needs a Bodyguard
Analysis | Hard InterruptIt started long before that cocktail (Studio Credits)On April 10, 2026, at 3:40 in the morning, someone threw a Molotov cocktail at Sam Altman’s Pacific Heights home. Within days, gunfire hit the property. A separate incident at Anthropic’s headquarters reportedly ended…
Analysis | Hard InterruptIt started long before that cocktail (Studio Credits)On April 10, 2026, at 3:40 in the morning, someone threw a Molotov cocktail at Sam Altman’s Pacific Heights home. Within days, gunfire hit the property. A separate incident at Anthropic’s headquarters reportedly ended when a security guard stopped a man who told him an executive was “going to be killed.”The Wall Street Journal (Ellis, Elinson, Li) broke the story on how many Silicon Valley executives now travel with armed bodyguards. The headline framing: an AI backlash driven by economic anxiety.That framing is technically true. It also buries one number, three case studies, and one 40-year-old precedent that together explain what is actually happening.MIT published a report earlier this year titled “The GenAI Divide: State of AI in Business 2025.” US enterprises collectively invested between $35 billion and $40 billion in AI. 95 percent of those investments produced zero return or no measurable impact on profits. The remaining 5 percent produced value.The AI industry has been running a $40 billion field experiment. The result: 95 percent of it did not work.Everything that has happened since (the firebombing, the bodyguards, the community protests at 142 sites across 42 states, the collapse of public support for data centre construction) is what happens when a $40 billion investment thesis fails in public.The pattern the Altman fire made visibleThe Altman attacker was reportedly carrying writings about AI causing “impending extinction.” Public coverage focused on that motive. Most readers left the story thinking this was one man’s fringe fear.Read the WSJ’s broader reporting and the picture is different. Silicon Valley-based JPT Security’s Dakota Dominguez told the outlet that tech CEOs did not have security budgets a few years ago. Now many tech companies incorporate personal security into their standard cost lines.Reportedly, another incident followed within days at Altman’s property (gunfire). At Anthropic, a separate intrusion involved a warning to a security guard about an executive being killed. These are the visible edge of a public mood, not one man’s fringe fear.142 protests, 42 states, one demographic in commonData Center Watch counted 142 protests across 42 states on July 18, 2026. Organised opposition groups doubled in Q1 2026 alone, from 396 at the end of 2025 to 833 by the end of March. Blocked or delayed AI data centre projects totalled roughly $130 billion in a single quarter. Reuters and Ipsos polled Americans in June: only 14 percent said they wanted a data centre built in their own community.Box Elder County, north of Utah’s Great Salt Lake, drew thousands of formal protest letters over the Stratos project’s water rights request; the application was withdrawn. Meta’s Project Cosmo reportedly contaminated a municipal reclaimed water system with a bacterium that took months to clear.The protest movement is not partisan. Reuters and Ipsos found Republican support for local data centres at 49 percent and Democratic support at 36 percent. Both groups were minorities. Both groups were angry at the same thing: a facility being built next to them that would raise their electricity prices, consume their water, and produce nothing they could see.Three receipts inside the 95 percentRead the MIT number and it feels abstract. Look at three specific cases and it stops feeling abstract.Klarna, February 2024: the CEO announced AI was doing the work of 700 customer service agents, handling 75 percent of chats, 2.3 million conversations in a single month. The company paused all hiring. Headcount dropped 22 percent to 3,500 employees.Klarna, May 2025: the same CEO, Sebastian Siemiatkowski, reportedly told Bloomberg the company was hiring humans again. His words: “We went too far. We focused too much on cost. The result was lower quality.” The AI stayed as a tool inside a hybrid model with humans reintegrated.IBM Watson Health, 2015 through 2022: over $4 billion invested via acquisitions to build a healthcare AI business. Sold in January 2022 to Francisco Partners reportedly for around $1 billion. Roughly a quarter recovered. MD Anderson’s Oncology Expert Advisor project reportedly spent over $60 million with little benefit before termination. IBM leaders later admitted they were “too optimistic” about Watson’s healthcare potential.Salesforce Agentforce, Q1 2026: approximately 5.3 percent of Salesforce’s 150,000-plus customers had adopted Agentforce. Only 31 percent of Agentforce deployments remained active after six months, per the IBM State of Salesforce 2025–2026 study. Approximately 10 percent of customers had reportedly moved past proof of concept.Three companies. Three industries. One pattern: AI sold to boards as a cost cutter, deployed at scale, walked back or written down when the numbers came in. Klarna, IBM, and Salesforce sit on the trend line the MIT number describes.The number Silicon Valley will not say out loud, and what actually shipped valueMIT’s methodology is not obscure. 150 interviews with AI leaders. 300 AI applications examined. 350 employees surveyed. The 95 percent finding held across all three cohorts. The failure causes are named: brittle workflows, lack of contextual learning, misalignment with day-to-day operations. Over half of the AI spend went to sales and marketing, which still needs humans and produces lower ROI than proponents promised.The 5 percent that succeeded had a shared profile per the MIT researchers. They picked one pain point, executed well, and partnered smartly with tool providers who understood the workflow. They deployed AI to help engineers who already knew the workflow work faster. They kept the engineers.The pattern is testable. Multiple studies of AI coding assistants show developers ship faster when AI runs as a tool inside their editor. SWE-Bench Pro scores from 2025 to 2026 show consecutive AI coding model releases beating each other on complex code tasks. The tool improved. The tool works when engineers use it.An engineer with Claude open in their editor and a specific pull request in front of them is not the target of the backlash. That engineer is producing value. A CFO with an AI vendor deck and a spreadsheet targeting Q3 layoffs is producing the $40 billion category error.The Asian Tigers precedent Silicon Valley will not nameThis is not the first time Western companies bet their operating model on turning a labour cost into a strategic advantage. The last time, the lesson was clear enough to name.Late 1980s through early 1990s: Western manufacturers moved production to the Asian Tigers. South Korea, Taiwan, Singapore, Hong Kong. Later Malaysia, Thailand, Indonesia. The pitch to boards then was the pitch to boards now. Cut labour cost. Keep the output. Multiply shareholder returns.For a decade, the profits landed. Textiles, consumer electronics, semiconductors, hard drives, and displays all shifted east. Western companies posted strong margins for one, two, three quarters running.Then the capability transferred. By the mid-2000s, the same economies that had been the low-cost supplier were building the high-margin brands. Samsung. LG. TSMC. Foxconn as a global powerhouse behind an American logo. By 2020 the West was writing policy documents about how to reduce dependence on Taiwan for advanced chips and on China for consumer electronics manufacturing.The current chip war has one foundation. Western companies traded capability for margin in the 80s and 90s. Thirty years later, the price of embedded development platforms, microcontrollers, multiprocessors, and GPUs sits spiked partly because the manufacturing capability that used to be in Ohio and Bavaria now sits in Hsinchu and Shenzhen.The AI cycle is running that pattern faster. The Asian Tiger arc took three decades from cost arbitrage to strategic competitor. The AI cost-arbitrage arc is running through the same shape in three years. DeepSeek’s V4 landing at Claude Opus 4.8 capability at a fifth of the cost. Chinese open-weight models proliferating while US models sit inside export-control frameworks.The 95% MIT number is one snapshot of the current cycle. The Asian Tigers precedent is the movie the snapshot is a still from.Every board today making a headcount-cutting decision on AI is doing what boards did in 1988. The margin will land. The capability will transfer. The competitor will emerge. The earnings calls three years from now will be defensive.The story the WSJ frame keeps hiddenExecutives get bodyguards because their companies bet on AI as a headcount-cutting mechanism. That bet failed 95 percent of the time in the short term. The Asian Tigers precedent says the surviving 5 percent will face a capability-transfer competitor within three years. The workers cut are angry. The communities absorbing the data centres are angry. The 20-year-old who threw a Molotov at Altman’s gate had a rambling extinction manifesto. The 833 organised opposition groups do not.The security industry can protect executives from individual attackers. It cannot protect them from a public that has looked at the returns and drawn a rational conclusion about who benefited and who paid.The industry that will grow through this cycle is AI as a working tool. AI as a business model has failed the ROI test.Four signals to watch in the next 12 monthsIf the doctrine is confirmed, four specific data points will land inside the next 12 months. Save this list. Check it in October, in January, in April.One. A Klarna-style reversal from a named US Fortune 500 company. Publicly rehired workers replacing AI systems that “did not meet the standard.” The Klarna arc took 15 months from AI announcement to human rehire announcement. Watch for the 15-month clocks running out on companies that announced AI-first pivots in Q4 2024 and Q1 2025.Two. Q4 earnings language shifts. Companies that told analysts in Q1 2026 that AI would produce operating savings will use different language in Q4. Watch for “long-term AI investment” replacing “AI efficiency” in prepared remarks. The language change is the tell.Three. Data centre project cancellations exceeding announcements. Data Center Watch tracks blocked and delayed projects. The threshold to watch: any single quarter where cancellations exceed announcements. That is the signal that community opposition has moved from delay to defeat.Four. Enterprise CIO surveys landing new results. Watch for the “AI is a strategic priority” number in CIO surveys dropping below 50 percent. When it does, the boardroom conversation has already ended.What every enterprise engineering lead should ask their boardThree questions:One. What is our documented ROI to date on AI investment, in dollars, per project, over the last 12 months? If the answer is not a number, the answer sits inside the 95 percent.Two. What percentage of our AI spend went to sales and marketing (in the failure zone) versus engineering tool deployment (in the success zone)?Three. Did any layoff announcement in the last 12 months cite AI capacity as a reason? If yes, what specific AI capacity delivered by what specific measurement replaced the specific work done by the specific people cut?If the board cannot answer those three questions in writing, the company is running the 95 percent play. The Asian Tigers precedent says the capability-transfer bill lands within three years.If you are a board member reading this, forward it to your CFO. If you are a CFO reading this, forward it to your board.The 95 percent number is the story. The Altman firebombing is the visible symptom. The Asian Tigers precedent is the movie. Everything else is aftermath.This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!The 95% Number That Explains Why Sam Altman Needs a Bodyguard was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.Source: Generative AI Pub — Published — Category: Image AI