Treasury Is 65% Wrong, AI Is a Dead End, and We're All Paying for the Build-Out

Jim Cooper is a political scientist, innovation policy expert and venture capital advisor with decades of experience across frontier technology, defence and commercialisation.James Yang is the Co-founder of KYZON Solutions, building AI-native workspace and sovereign AI infrastructure products for SMBs and regulated industries across APAC.Hours before the consultation window closed on the biggest tax change in Australia in decades, Jim lodged a submission of his own. His verdict is not what you would expect from someone who opens by disclosing he has been a Labor supporter for a very long time: the country spent months arguing about the wrong lever, and nobody who lobbied made their case in Treasury's own language.Then what is actually moving the economy. MIT priced a decade of AI productivity gains at 0.7 per cent against a trillion dollars of capex. They work through what Salesforce really spends on tokens, why Jim thinks LLMs are a dead end, where the senior engineers of 2035 come from once AI has eaten the bottom rung, and a supply side where Australia imports AI as a service while its exports get eaten, the US has locked Anthropic out of its newer models, and three companies control the world's memory as RAM prices double in a quarter.Jim spends the episode refusing to call it a bubble. The RAM shortage is what finally moves him, and it ends with his device running out of storage mid-sentence in a conversation about a global memory shortage.Timestamps:[00:00] "Treasury is 65 to 70% wrong on this" [00:28] The submission that beat the midnight deadline [03:41] 85% of new revenue comes from negative gearing [05:51] "The Australian startup starvation diet" [08:20] "Working hard is not a fairness argument" [10:06] July 2027, and the biggest asset sell-off in Australian history [11:13] MIT ran the numbers. 0.7% against a trillion in capex [11:57] "Are we going to reach that speculative bubble?" [13:56] What Salesforce actually spends on tokens [18:26] "LLM is a dead end" [19:01] 17,000 tasks. AI failed half the legal and management work [20:39] Deloitte shipped a government report full of AI errors [23:32] "Is it slop if it makes the best banana bread recipe?" [25:57] Junior dev jobs down 20%. Where do 2035's seniors come from? [29:41] Services, resources, property, no sovereign AI model [34:26] The US lock on Anthropic. "Effectively a kill switch" [35:26] The worst day for chip stocks since 2020 [39:43] GPUs process the intelligence. RAM holds the context [41:26] "Not even indirectly. We are directly subsidising it" [42:10] A cartel of oil, and now a cartel of memory [45:08] "I'm now leaning more towards it's a bubble" [47:22] The shortage runs to 2028 [47:55] A house that was 1.1 million is now 835,000 [49:56] The cheap smart home we were promised [50:14] "You've completely run out of memory, Jim"Subscribe on Spotify: https://open.spotify.com/show/16jy2Th6ES9uyfge4saltn?si=950d93992dec443e Subscribe on Apple Podcasts: https://podcasts.apple.com/us/podcast/making-the-cut/id1895846511Follow Jim Cooper on LinkedIn: https://www.linkedin.com/in/spyfor/ Follow James Yang on LinkedIn: https://www.linkedin.com/in/jamesyang8/#makingthecut #jimcooper #jamesyang #cgt #australianstartups #ai #aibubble #sovereignai #semiconductors #ram