← Back to Archive · Homepage - Newsletter

Web archive · this issue of MCC AI Weekly was sent by email on Thursday, September 10, 2026.

MCC AI Weekly · Thursday, 10 September 2026

The cost of AI is falling faster than its returns are rising

MCC AI Weekly — issue cover illustration

Token prices keep falling and consumption keeps rising, but the economics behind the headlines are more uneven than they look — for hyperscalers, for enterprises, and for the workers whose jobs sit on the other side of the ledger. This issue tracks that gap: between cost and value, between strategy and integration, and between what leaders say about AI and what their metrics actually reward. Start with Gates's case for taxing the machines, then see why most companies still can't turn cheaper tokens into measurable returns.

Insights

White-collar AI disruption is already here, with no plan to manage it

Bill Gates reports white-collar disruption is already visible. Employment has fallen among young workers in occupations especially exposed to generative AI, but not among older peers. He expects sales, customer support, software engineering and paralegal work to be hit first, followed by loan assessment, data analysis and patient triage. Gates argues no one has a real plan for this transition.

So what

Gates recommends taxing AI tokens and robots much as payroll taxes apply to human workers. The revenue would fund retraining and a stronger safety net, since he sees the current tax code as favoring machines over people.

Gates Notes →

Falling token costs set up margin inflection, not just a spending story

Goldman Sachs Research forecasts token consumption will multiply 24 times by 2030, reaching 120 quadrillion tokens monthly, as consumers and enterprises adopt AI agents. Analyst Jim Schneider notes inference costs are falling 60-70% per year, positioning hyperscalers for a "gross margin inflection" within 3-12 months. Enterprise adoption trails consumer use: Goldman forecasts only 12% of knowledge workers using agentic AI by 2030, rising to 37% by 2040, with a chip shortage likely over the next 12-18 months.

So what

Goldman's Schneider argues the prevailing narrative of AI usage as an unsustainable cost burden should be reset, since falling per-token costs are outpacing prices. Expect differentiation among hyperscalers on operating cash flow as the margin inflection arrives unevenly, not simultaneously.

Goldman Sachs Research →

Most workers outpace the systems meant to support them

Microsoft's Work Trend Index, analyzing trillions of Microsoft 365 signals and surveying 20,000 workers across 10 countries, finds just 19% of AI users sit in the 'Frontier' zone where individual skill and organizational readiness reinforce each other. Microsoft calls this the Transformation Paradox: 65% fear falling behind without AI, yet 45% say it feels safer to stick with current goals, and only 13% feel rewarded for reinventing work.

So what

Leaders should set strategy and redesign metrics, incentives, and expectations so people are rewarded for changing how they work, not just for hitting current goals. Microsoft's own study found managers who openly model AI use and create space for experimentation drive measurably higher trust and AI value among employees.

Microsoft Work Trend Index 2026 →

Fewer than 10% of AI users drive most of the token bill

Accenture finds that in nearly every deployment it has examined, including its own, fewer than 10% of users and workflows generate the vast majority of AI token spend. It reports that only 23% of C-suite leaders surveyed see widespread, sustained business value from AI. Accenture argues the gap comes from treating tokens like a utility bill instead of matching each task to the model it actually needs.

So what

Have the CFO and CIO jointly own one enterprise view of AI cost, usage and return, as Accenture does internally. Build routing and access controls into deployment before scale hardens, rather than adding them once the bills expose the gap.

Accenture →

Only 6% of firms see real AI value despite CEO optimism

BCG's 2026 AI Radar survey finds 82% of CEOs are more optimistic about AI ROI than a year ago. Yet only 6% of companies report meaningful cost or revenue gains from AI. Pilots automate existing tasks rather than redesigning end-to-end processes, so cost structures stay unchanged even as work feels faster. The successful 6% prioritize a few initiatives and redesign processes around AI-enabled decision making.

So what

Prioritize a small number of high-impact initiatives and redesign the end-to-end process around AI-enabled decision making, rather than automating existing workflows. Measure cost and revenue outcomes, not pilot-level activity.

BCG →

Weekly Short News

KPMG: German firms embrace AI strategy, but integration lags

KPMG surveyed 480 decision-makers across German industries for its 2026 study on generative AI. It finds 97% rate AI as relevant to their business model, and 79% see AI already strongly affecting their industry. Fewer than 1% report full, systematic AI integration across all areas reviewed. Only 33% steer trustworthy AI use strategically company-wide, while 82% cite regulation such as the EU AI Act as a strong influence.

KPMG →

Token prices halved, but consumption grew 450%, Bain finds

Bain & Co. reports token costs fell by half between December 2024 and December 2025, while token consumption rose 450% in the same period. Apollo chief economist Torsten Slok called it a case of Jevons paradox, where cheaper AI use leads companies to run more agents and automate more workflows. Bain projects company operating expenses eventually splitting 70% headcount and 30% tokens.

Fortune →

Foreign investors pay to tour Chinese AI and robotics factories

Foreign investors and entrepreneurs are paying to tour Chinese factories in AI, robotics and EVs, Reuters reports. Beijing has designated over 140 official industrial-tourism demonstration sites, and Xiaomi's EV factory alone has welcomed more than 250,000 visitors since March 2024. Lottery-awarded entry slots are being resold online for up to 2,000 yuan, or about 300 dollars.

Reuters →

TrendForce: Chinese AI chips to reach 90% domestic share in 2026

TrendForce projects domestic AI accelerators will capture nearly 90% of China's AI chip sales in 2026, up from 45% in 2025, leaving AMD and Nvidia roughly 10%. Huawei and Cambricon are expected to be the biggest beneficiaries. The forecast assumes China's industry increases high-end accelerator output by more than 83% year-over-year, per TrendForce.

Tom's Hardware (citing TrendForce and DigiTimes) →

OpenAI publishes first Jalapeño chip benchmarks

OpenAI released its first benchmark results for Jalapeño, the inference chip it co-designed with Broadcom. The company reports the chip serves more AI work per unit of power while returning responses faster. Existing hardware systems often must trade one for the other. OpenAI evaluated performance at a matched user experience across multiple models, and plans to ramp Jalapeño deployment in the coming months.

OpenAI →

AI Player of the Week

Make

Make is a visual, no-code platform for wiring business applications into automated workflows, connecting thousands of pre-built apps through a drag-and-drop scenario builder. Process-mining firm Celonis has owned Make since acquiring it in 2020. It added AI Agents in April 2025, letting workflows make context-aware decisions instead of following fixed rules. Operations teams without engineering staff get a shortlist candidate for assembling integrations themselves; teams needing custom logic beyond drag-and-drop connectors should look elsewhere.

Make →

Tip of the Week

Turn your repeat prompt into a saved skill

Take a prompt you reuse weekly, such as a report format or a review checklist. Save it once as a skill instead of retyping or copy-pasting it from an old chat. In Claude, open Settings, then Customize, then Skills, and store the instructions there. Invoke the skill by name next time, and it runs exactly as written, with no drift between uses.

Found this useful? Forward it to a colleague — they can subscribe here.

You are receiving this email because you confirmed your subscription to MCC AI Weekly. Questions? Just reply to this email — it reaches us directly.

Imprint · Privacy Policy