{"slug": "man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill", "title": "Man Group Is Betting Big on AI Tokens While Wall Street Debates the Bill", "summary": "Man Group, the world's largest listed hedge fund manager, is aggressively integrating large language models into its trading operations, including a partnership with Anthropic to deploy Claude across its strategies. The firm has built internal tools like AlphaGPT and ManGPT that have already generated dozens of trading signals slated for live trading, even as industry-wide AI inference costs decline. This bet on AI infrastructure reflects a broader trend among quant shops like Citadel Securities and Hudson River Trading, which are increasing AI spending despite falling per-token prices, treating rising token bills as a necessary cost to stay competitive.", "body_md": "*Man Group is pouring real money into AI infrastructure just as fresh data shows the price of that same computing power is falling. The world's largest listed hedge fund manager doesn't seem to care.*\n\nMan Group closed the first quarter of 2026 with $228.7 billion under management, according to its own trading update, roughly flat from the prior quarter after a client pulled $6.1 billion from one long-only equity strategy. Withdrawals like that would normally dominate the headline. Instead, the more interesting story inside the firm is how aggressively it has wired large language models into the actual business of finding trades, not just answering emails faster.\n\nIn February, Man Group announced a partnership with Anthropic to run Claude across its public and private market strategies, including Claude Skills and Claude Code. Gary Collier, the firm's chief technology officer, framed the ambition without much hedging: \"We're aiming to fundamentally disrupt the creative quant process,\" he said, adding that \"it's early days, but we're showing some significant promise already.\" That's not the language you use for a pilot program you might kill in six months.\n\nThe tools behind that ambition have names. AlphaGPT, built internally by Man Numeric, mimics how a human quant researcher works: it digs through data and academic research for ideas, writes the code to test them, then runs the backtests before sending results into the firm's normal investment committee review. Ziang Fang, a senior portfolio manager at the firm, has said several dozen signals generated by the system have already cleared that committee and are slated for live trading. A second internal tool, ManGPT, handles coding, document drafting and workflow customization for portfolio managers, and roughly half the firm now uses it monthly.\n\nNone of that comes free. And that's where the story gets interesting.\n\nBloomberg reported this month that the Silicon Data LLM Token Expenditure Index, which tracks what buyers actually pay for AI inference, is down nearly 20% from its May peak after nearly doubling since the index launched in December. Cheaper tokens should mean cheaper AI. Instead, total industry spend keeps climbing, because falling prices are pulling in far more usage than they're removing in cost per query. Bloomberg has also reported that OpenAI spends close to two dollars for every dollar of inference revenue it earns, effectively subsidizing the market it's trying to build.\n\nThat's the tension quant shops are navigating in real time, and they're not all reading it the same way. Jim Esposito, president of Citadel Securities, told the Bloomberg Markets and Banking Summit in April that \"our spend has been increasing,\" and added that \"at the moment we're getting a healthy return on that spend.\" It's the same trade Man Group is making: treat a rising token bill as the cost of staying ahead, not a line item to shrink.\n\nNot everyone frames it so comfortably. Hudson River Trading's head of AI, Iain Dunning, told Bloomberg's Odd Lots podcast last month that compute has become a genuine bottleneck at the market maker, with employee token consumption growing fast enough that the firm is weighing whether to build its own chips. HRT isn't a discretionary quant fund the way Man Group is, but the underlying pressure reads the same: usage is scaling faster than anyone's budget assumed, regardless of which direction the per-token price is moving.\n\nWhat separates Man Group's spending from a chatbot experiment is where the tokens actually go. They aren't funding customer support. They're funding the trial-and-error grind of research that used to require a room of PhDs testing hypotheses by hand against decades of market data. If AlphaGPT's signals keep clearing the live-trading bar, the token bill isn't overhead sitting on the cost side of the ledger. It's research spend that pays for itself in basis points, assuming the signals hold up outside a backtest.\n\nThe risk sits on the other side. According to pricing data compiled by research group Epoch AI, inference costs have fallen roughly 600-fold since the GPT-3 API launched in 2020, and further declines are widely expected as competition among model providers intensifies. If you're looking for a real test of whether the AI capex story turns into returns rather than just efficiency gains, this is closer to it than any consumer chatbot usage figure: a $228 billion quant shop paying real money for tokens because they generate alpha, not because a vendor talked them into a trial. Man Group is betting those returns outrun the bill long before falling prices make today's build look overdone.\n\n**Also read:** [CXMT's $4.3 Billion IPO Puts Beijing's Chip Ambitions to a Real Test](https://startupfortune.com/cxmts-43-billion-ipo-puts-beijings-chip-ambitions-to-a-real-test/) • [Zhipu raises $4 billion in Hong Kong days after its stock surged nearly 1,500 percent](https://startupfortune.com/zhipu-raises-4-billion-in-hong-kong-days-after-its-stock-surged-nearly-1500-percent/) • [Lovable is in talks to double its valuation to $13.2 billion in six months](https://startupfortune.com/lovable-is-in-talks-to-double-its-valuation-to-132-billion-in-six-months/)", "url": "https://wpnews.pro/news/man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill", "canonical_source": "https://startupfortune.com/man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill/", "published_at": "2026-07-09 08:52:54+00:00", "updated_at": "2026-07-09 08:58:42.100655+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-infrastructure", "ai-products", "ai-tools"], "entities": ["Man Group", "Anthropic", "Claude", "AlphaGPT", "ManGPT", "Citadel Securities", "Hudson River Trading", "Bloomberg"], "alternates": {"html": "https://wpnews.pro/news/man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill", "markdown": "https://wpnews.pro/news/man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill.md", "text": "https://wpnews.pro/news/man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill.txt", "jsonld": "https://wpnews.pro/news/man-group-is-betting-big-on-ai-tokens-while-wall-street-debates-the-bill.jsonld"}}