Forbes: How Did Sam Bankman-Fried’s Alameda Research Lose So Much Money?

ForbesPublicado em 2022-11-21Última atualização em 2022-11-22

Resumo

If Alameda started with bad accounting systems, Pack says it’s “not inconceivable” that they could have ended up with much more debt than they realized, as Bankman-Fried has claimed.

A week after the dramatic collapse of Sam Bankman-Fried’s tangled web of crypto companies, countless unanswered questions remain. One of the biggest: How did his trading firm, Alameda Research, apparently lose billions of dollars? Those losses appear to have prompted someone in Bankman-Fried’s operation to improperly transfer customer funds from trading platform FTX to Alameda, a decision that left FTX vulnerable to a withdrawal run that precipitated the sudden bankruptcy.

Many details remain unknown, but a blurry picture is forming of the possible causes behind Alameda’s steep losses. We spoke with a half-dozen crypto traders and investors familiar with Alameda to understand the leading theories. A spokesperson for Sam Bankman-Fried and Alameda’s former co-CEOs Caroline Ellison and Sam Trabucco didn’t respond to Forbes’ requests for comment. We sent Bankman-Fried questions on messaging app Signal, but he hasn’t yet answered them.

Moving From Arbitrage to High-Risk Bets

The first theory is that the young traders at Alameda, which was once one of the largest crypto trading firms in the world, weren’t as sophisticated as their reputation suggested. Bankman-Fried was regarded as an excellent trader when he started Alameda in 2018, and he focused on arbitraging price differences in cryptocurrencies in different markets. But the next year, he shifted his primary focus to launching his trading platform FTX. He brought with him to FTX his Alameda colleagues Gary Wang and Nishad Singh, who had been some of the most talented people at the trading firm, according to Doug Colkitt, a veteran high frequency stock trader turned crypto trader.

After bitcoin started to rise sharply in the fall of 2020, Alameda moved away from its initial focus on making high-speed, market-neutral bets that didn’t depend on predicting if cryptocurrencies would rise or fall. Some traders believe Alameda changed its strategy because it lost its competitive edge as more experienced firms like Jump Capital ramped up their crypto trading business.

In March 2021, then 26-year-old Caroline Ellison, one of Alameda’s co-CEOs, seemed to acknowledge this pivot when she tweeted, “Also relatable is the point where he realizes he's been wasting time trying to trade back and forth for a few points of edge and the way to really make money is figure out when the market is going to go up and get balls long before that.” Going long means betting that prices will rise.

A month later, Sam Trabucco, Alameda’s other co-CEO, tweeted, “we got ... uh, really long in winter 2020.” As a rationale for why, he added, “it’s where the money is.” Both Ellison and Trabucco had just a couple of years of trading experience in conventional markets before joining Bankman-Fried to deal in crypto. That’s a shallow pool of knowledge and experience to draw on.

According to several traders, many of Alameda’s long bets probably suffered big losses beginning in May 2022, after the dramatic collapse of the stable coin terraUSD and its sister cryptocurrency luna sharpened the decline in the crypto market. “What makes you a hero in bull markets kills you in bear markets,” says Marina Gurevich, chief operating officer of London-based Wintermute, one of the most active crypto trading firms in the world. Indeed, Bankman-Fried acknowledged in a Twitter conversation with a Vox reporter that it was around the time of luna’s crash when a lot of risky leverage built up in his business.

Layering Leverage on Top of Big Bets

On top of making big bets, Alameda was likely taking on too much leverage–that is, debt that can amplify wins and losses. One way the firm’s executives apparently did that was by using largely illiquid cryptocurrencies–including FTX’s own token, FTT, and a related one, serum–as collateral to take out loans.

For example, Bankman-Fried helped incubate the creation of serum, which was released in 2020. Serum has a low circulating supply of coins–initially, only 10% of it was freely tradeable, while the other 90% was locked up for years. But technically, he could extrapolate and assume that, if the circulating supply of serum was worth $1 billion, then the market value of all the coins in existence was $10 billion. Then he could get loans based on that higher valuation. Bankman-Fried ran this playbook with other digital assets too, which became known as “Sam coins” to industry insiders, crypto investor Jason Choi has written.

Choi concluded recently in a tweet, “This is likely how Alameda/FTX incurred the multi-billion-dollar hole: Alameda pledging illiquid collateral to borrow money to finance bets, which got margin called as markets went down this year.”

Investing Borrowed Money in Other Crypto Players

Another capital drain was venture investments. According to PitchBook, Alameda made more than 150 investments across the crypto industry, including in bitcoin miner Genesis Digital Mining and now-bankrupt crypto broker Voyager Digital. Alameda apparently took out loans to fund those bets. As the crypto market crashed, lenders reportedly attempted to recall those funds that were tied up in these illiquid investments. FTX’s and Alameda’s executives then took the questionable step of trying to pay back some of those Alameda loans using FTX customer funds, the Wall Street Journal has reported.

Borrowing for Other Big Spending

The finances of Bankman-Fried’s cluster of companies are so complex and entangled that huge chunks of it remain a mystery–even to the lawyers, financial investigators and bankruptcy veterans who have taken charge. But according to bankruptcy court filings, FTX executives also took out billions of dollars in loans from Alameda to fund everything from political contributions to Bankman-Fried’s purchase for $650 million of a 7.6% stake in Robinhood. It’s unclear how these loans may have also added to Alameda’s losses on top of everything else. Alameda itself has outstanding liabilities of $5.1 billion according to a filing Thursday in the Chapter 11 bankruptcy case in Delaware.

Shoddy Record-Keeping and Accounting Controls

A final–and perhaps substantial–contributor to Alameda’s losses: Bankman-Fried’s companies had terrible record-keeping and accounting systems. FTX customer deposits were not tracked, according to a bankruptcy filing, leaving it unclear in the bankruptcy proceedings what’s owed to customers. An example of this confusion: the leaked FTX balance sheet shows $8.8 billion in liabilities, while the Thursday filing in the Delaware bankruptcy case shows only $6.4 billion. It’s not clear what accounts for the discrepancy, but regardless, the numbers are still in flux. “This balance sheet was produced while the Debtors were controlled by Mr. Bankman-Fried, I do not have confidence in it,” workout veteran John J. Ray III, the new CEO of FTX overseeing the bankruptcy wrote in the filing. Bankman-Fried has tried to chalk up nearly the entire problem to “messy accounting + margin.”

Bankman-Fried’s careless accounting habits appear to date back to the earliest days of Alameda. When crypto venture capitalist Alex Pack was considering investing in Alameda in early 2019 and conducting due diligence, he saw they had lost $10 million in a single month–a hefty sum for such a small firm. When Pack asked about it, Bankman-Fried said it was due to “trade errors,’’ Pack recalls.

Pack says he kept probing, but he could never figure out what happened. “At one point, they just said, ‘Sorry, we didn't have great record keeping back then. We can’t answer all these questions.’” Pack passed on the deal. He thought they seemed like smart traders but walked away due to what he saw as “significant recklessness around risk taking and extremely poor infrastructure and accounting.”

Today, Pack says tracking positions in crypto can be particularly hard because you have to build your own trading systems, and the task gets “exponentially more difficult” as your book of business grows. And if Alameda started with bad accounting systems, Pack says it’s “not inconceivable” that they could have ended up with much more debt than they realized, as Bankman-Fried has claimed.

Leituras Relacionadas

Deep Insight: Decentralized Inference is Not Hype, but a Key Track for AI to Break Through Centralized Monopoly

Decentralized Reasoning: Beyond the Hype, a Key to Breaking AI's Centralized Monopoly A future scenario where a powerful AI model is banned by a major government illustrates the core value proposition of decentralized AI: resistance to censorship. The core bet of decentralized inference networks is mitigating this risk, with other benefits like cost being secondary. The path is extremely difficult, involving four key challenges: 1. **Running Massive Models:** Distributing a single model across a decentralized GPU swarm requires sophisticated techniques like pipeline and speculative decoding to overcome crippling network latency, aiming for usable speeds (e.g., 30-40 tokens/second). 2. **Proving Model Integrity:** Verifying that a node runs the correct model is critical. Solutions range from cryptographically secure but slow ZKML to faster, economically-secure methods like statistical fingerprints, deterministic re-execution, or live-weight proofs, each involving trade-offs between integrity, latency, and cost. 3. **Ensuring Prompt Privacy:** Simply sharding a model does not protect user inputs from nodes. Robust solutions currently require trusted hardware (TEEs) or advanced cryptography (FHE), which are not yet widely deployed in consumer swarms. 4. **Building a Real Market:** Identifying the ideal customer is tough. Beyond speculative AI agents, the viable market currently consists of startups embedding AI and projects needing batch processing (e.g., synthetic data generation), where decentralized aggregation can be an advantage over low-latency needs. The article analyzes several projects tackling these problems, such as Dolphin Network (live-weight proofs), Inference.net (statistical verification), Morpheus (TEE-based), and Darkbloom (Apple Secure Enclave). It provides a framework: decentralization is a "tax" for latency-sensitive applications (e.g., chat) but a potential supply-side advantage for throughput-oriented tasks (e.g., batch processing). The long-term vision is a closed data loop where decentralized inference generates valuable data (traces, preferences) to feed decentralized training networks, which in turn produce better open-weight models for the inference networks. A due diligence checklist advises focusing on projects that: are truly decentralized at specific layers; have a credible integrity method; offer real cost benefits; ensure genuine privacy; handle node reliability; have paying users; and are built by teams with deep AI expertise. The ultimate goal should be products that appeal beyond the crypto-native audience, using crypto mechanisms invisibly to deliver better cost, performance, or privacy.

Foresight NewsHá 13m

Deep Insight: Decentralized Inference is Not Hype, but a Key Track for AI to Break Through Centralized Monopoly

Foresight NewsHá 13m

The Final Piece of Franklin Templeton's Crypto Ambition

Franklin Templeton Completes Crypto Ambition with Acquisition of 250 Digital On June 22, Franklin Templeton announced the acquisition of 250 Digital and established Franklin Crypto, a new division focused on actively managed cryptocurrency strategies for institutional investors. The unit is led by Christopher Perkins and Seth Ginns. This acquisition marks a key piece in Franklin Templeton's multi-year crypto strategy, which began in 2018 with a digital assets team. The firm's crypto product suite now spans three layers: tokenized funds like the blockchain-based money market fund BENJI (~$831M AUM); a series of passive ETFs including Bitcoin (EZBC, ~$368M), Ethereum (EZET), XRP (XRPZ, ~$252M), Solana (SOEZ), and a multi-crypto index fund (EZPZ); and the newly added active management strategies from Franklin Crypto. The company has also expanded its crypto ecosystem through investments in projects like Ethena and Crossmint, and collaborations with blockchains such as Aptos and Sui. With approximately $18B in digital asset AUM and a total firm AUM of ~$1.78T, Franklin Templeton is positioning itself as a comprehensive crypto asset manager for pensions and sovereign wealth funds. In contrast, competitor Fidelity Investments has taken a different path, focusing early on building its own custody and trading infrastructure. Fidelity's Bitcoin ETF (FBTC) holds over $11B, significantly larger than Franklin Templeton's equivalent offering. Both giants' moves underscore the deepening trend of traditional finance entering the crypto space.

Foresight NewsHá 37m

The Final Piece of Franklin Templeton's Crypto Ambition

Foresight NewsHá 37m

Trading

Spot
Futuros
活动图片