MIT Technology Review: AI Is Rapidly Changing the World, but Its Future Remains Uncertain
In his speech at SXSW London, a senior editor at MIT Technology Review highlighted five key themes: AI has become commonplace, real fears (deepfakes, dangerous chatbots) are materializing, anti-AI protests are growing, AI's enormous potential for science, and the technology's pervasive reach. He urged preparing for a marathon, not a sprint.
Marathon, Not Sprint: Why the AI Euphoria Is Ending and the Real Game Is Just Beginning
If you read Will Douglas Heaven's column in MIT Technology Review following his talk at SXSW London, you might have thought it was just another trend overview. Five points—from fears about jobs to protests against data centers. But behind these seemingly obvious theses lies a much deeper and more troubling picture for insiders. We are moving from the "gold rush" phase to the "survival of the fittest" phase, where what's at stake is no longer just market share but the very model of the industry's existence.
The Core: What's Really Happening
The main non-obvious insight that is being overlooked is that the AI industry is simultaneously experiencing peak penetration and a crisis of trust, measured not only in words but in dollars and even Molotov cocktails. On one hand, we see phenomenal numbers: ChatGPT surpassed 1.1 billion monthly active users, becoming the fastest-growing app in history. On the other hand, for the first time ever, OpenAI's market share fell below 50% (to 46.4% at the end of May), and capital spending on data centers in 2026 will exceed $1 trillion. This is not just competition—it's a structural shift.
The most striking marker of this shift is the politicization of AI and loss of trust as a key factor in choice. When OpenAI signed a contract with the Pentagon in February 2026, the number of ChatGPT app deletions in the US jumped 295% in one day, and its App Store rating saw a 775% increase in negative reviews. People vote with their wallets (and uninstalls) against what they consider ethically unacceptable. Notably, announced layoffs at Block (40% of staff) and Atlassian (1,600 people) coincided with the start of public protests, and management didn't even hide that AI was a formal pretext for cost-cutting.
Players are trying to outmaneuver each other not in technology but in geography. By embedding Gemini into Android, Chrome, and Gmail, Google captured 27.7% of the market—a structural advantage that cannot be caught up with just a smarter model. Meanwhile, China, represented by DeepSeek, has effectively dropped out of the market share race (less than 5%) but ranks third in usage time—users linger there because it's very cheap and free.
Timeline and Context
To understand how we got here, we need to look at the calendar of the last two years. 2024–2025 was the "Wild West" era: launch of GPT-4, mass corporate adoption, regulators racing to catch up. In 2024, the European AI Act was passed—the world's first strict law, which in 2025 banned "unacceptable risks." 2025: Trump returned to the White House and announced private investments in AI infrastructure of $500 billion, while revoking Biden's safety executive orders. This created a unique situation: the US ran toward deregulation, the EU enforced strict rules, and the UK tried to become a "third force" (securing promises from Microsoft, Nvidia, and Google for investments totaling over £150 billion).
The first half of 2026 is the moment of truth. May 2026: ChatGPT lost its absolute majority for the first time. June 2026: IEA data showed data center energy consumption reached 565 TWh, 26% more than in 2025. This is no longer an abstract threat; it's a direct conflict with the public. In March, protesters in London held the largest march against AI, and in Texas, someone threw a Molotov cocktail at Sam Altman's home. Regulators are no longer "good cops"—California and New York have passed laws on chatbot oversight, and at the federal level in the US, AI is required to be "impartial."
Who Wins and Who Loses
If before the winner was the one with more parameters, now the winner is the one with better distribution and workforce stability. Google wins in distribution—every sold Android smartphone automatically adds a Gemini user. This is a hard channel monopoly that OpenAI cannot overcome with any marketing budget. Anthropic with Claude wins in "loyalty of paying customers": Claude's paying user share is about 13%, the highest conversion rate, and its audience grew 452% year over year.
All those who cannot afford data centers and talent lose. The cost of entry is becoming astronomical. Capital expenditure per 1 kW of capacity for an AI data center is $40,000–50,000, and total investment through 2030 is estimated at $3.9 trillion. Transformers for power grids have 2–3 year lead times, and queues for grid connection in Texas reach 160 GW—twice the state's peak demand.
All small startups not in the orbit of hyperscalers lose. The market is consolidating. 46.4% market share for OpenAI, 27.7% for Google, 10.3% for Anthropic. The rest—including Grok, Perplexity, and even DeepSeek—fight for 5%, and most will either die or be acquired in the next 18 months.
What the Media Isn't Saying
The media write about "explosive growth" and "the race," but omit the main point: we are facing an "accelerating stagnation" effect. The number of users is growing, but the quality difference between models is rapidly shrinking. If a year ago ChatGPT was technologically unreachable, today most users see no difference between responses from GPT-4o, Gemini Ultra, and Claude. This means competition is shifting to a "flat" plane: price, convenience, trust, and integration.
The second hidden factor is the real cost of errors. In the military domain, AI already provides recommendations for target selection. An error in that context means lives, not just a wrong dinner recipe. Yet companies cannot control their own models: if a military chatbot errs, liability falls on the company in court. Lawsuits already exist claiming chatbots pushed teenagers to suicide. The media report this tepidly because it's hard to monetize with clicks, but it's a ticking time bomb.
The third point is the real scale of "green" resistance. The MIT article mentions protests, but the numbers are shocking: in Q2 2025, activists stalled data center construction worth $98 billion. This is not "neighborly discontent"; it's a systemic risk for the entire industry, as any delay in bringing capacity online means losing market share for players who have already sold cloud capacity forward.
Forecast: Next 30 and 90 Days
Next 30 days (July 2026): Expect the first wave of "regulatory lawsuits" in the EU. On August 2, 2026, full requirements for "high-risk" AI systems under the European AI Act come into force. This means companies will start massively disabling features for European users or hastily rewriting code to meet transparency and audit requirements. This will cause a local stock crash for those unprepared.
Next 90 days (August–September 2026): OpenAI will hold an IPO with a potential valuation of up to $1 trillion. This will be the climax of the "bubble." Importantly, if the valuation falls below $800 billion, it will signal that the market no longer believes in extraordinary growth, and we will see a correction across all analog AI stocks. Simultaneously, due to transformer shortages and rising memory prices (memory prices have risen, and inventories have dropped from 4 months to 3 weeks), we will see the first serious delays in Nvidia Rubin server deliveries, cooling hyperscalers' enthusiasm and forcing them to revise 2027 forecasts.
We are indeed in a marathon. Only those who survive the next 90 days of regulatory and infrastructure storm will live to see 2027. The rest will be written off by history, just as we once wrote off that very "conference speech" generated by AI that cost someone their job.
— Editorial Team
No comments yet.